Method and system for generating answers

Abstract

The present invention relates to a method and a system for generating answers. More specifically, the present invention relates to a method and a system for generating answers by using a large-scale foundation model. The present invention relates to an answer generation platform based on a large-scale foundation model. In addition, the present invention relates to a chemical reaction prediction system, a control method therefor, and a method for training the chemical reaction prediction system. More specifically, the present invention relates to: a chemical reaction prediction system for performing an electron flow-based forward reaction prediction; a control method therefor; and a method for training the chemical reaction prediction system.

Description

Method and system for generating answers

[0001] The present invention relates to a method and system for generating answers, and provides a method and system for generating answers using a generative model or a foundation model. The present invention relates to an answer generation platform based on a super-large foundation model. The present invention also relates to a chemical reaction prediction system, a control method thereof, and a learning method for a chemical reaction prediction system. More specifically, the present invention relates to a chemical reaction prediction system that performs forward reaction prediction based on electron flow, a control method thereof, and a learning method for a chemical reaction prediction system.

[0002] Recently, there has been a rapid increase in cases where artificial intelligence, especially deep learning, which extracts data characteristics using deep neural network structures, is achieving outstanding results in various fields such as voice recognition, image recognition, natural language processing, and autonomous driving.

[0003] Along with these advancements in deep learning technology, generative AI technology has recently been attracting attention. More specifically, generative AI models can generate new data in various forms, such as text, images, and voice, from given data. This offers a new level of application potential beyond simply classifying or predicting existing data.

[0004] In other words, as sentences, images, voices, etc. that were previously created by humans can now be automatically generated using generative artificial intelligence models, services using generative artificial intelligence (e.g., ChatGPT) have shown activity and accuracy that differentiate them from existing chatbot services, and are receiving great attention worldwide.

[0005] Meanwhile, efforts to solve various scientific problems in natural sciences (e.g., physics, chemistry, biology, etc.) are continuously being made. For example, research is actively underway to design new materials and develop new drugs, and such research is playing a crucial role in future technological advancements and industrial innovation.

[0006] In this regard, organic synthesis is a crucial task in new drug development and materials science, and predicting the outcome of chemical reactions is crucial for designing new molecules. However, it takes considerable time and money for researchers to directly perform chemical synthesis, such as molecular synthesis.

[0007] Accordingly, research is actively being conducted on methods to increase the efficiency of natural science research based on generative artificial intelligence.

[0008] The present invention provides an answer generation method and system that can suggest optimal research methods to researchers in the field of natural science.

[0009] More specifically, the present invention provides a method and system for generating answers that can minimize the risk of failure in natural science research and thereby increase the efficiency of natural science research.

[0010] Furthermore, the present invention provides a method and system for generating answers that can solve the problem of time and cost required for material research and development and increase the efficiency of material research and development.

[0011] Furthermore, the present invention is to provide a chemical reaction prediction model that can predict the results of various types of chemical reactions by understanding the chemical reaction mechanism.

[0012] More specifically, the present invention provides a chemical reaction prediction model based on electron flow using graph diffusion, in which both input and output structures are formed in graph space.

[0013] Furthermore, the present invention provides a learning method of a chemical reaction prediction model that enables a good understanding of chemical reaction mechanisms and generation of more accurate and interpretable chemical reaction prediction results.

[0014] According to the present invention, a method in which a memory and at least one processor are cooperatively performed may include a step of specifying a plurality of molecular structures to be predicted based on a user input received through a service page, a step of storing information on the plurality of molecular structures in the memory, a step of obtaining a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures stored in the memory, a step of receiving a user query for chemical reaction prediction related to the plurality of molecular structures through the service page, a step of processing the molecular graph for the plurality of molecular structures as an input of a chemical reaction prediction model so that a chemical reaction corresponding to the user query is predicted, a step of performing chemical reaction prediction for the plurality of molecular structures in the chemical reaction prediction model, a step of obtaining a result of chemical reaction prediction for the plurality of structures from the chemical reaction prediction model, and a step of generating an answer to the user query using the result of chemical reaction prediction.

[0015] In an embodiment, the chemical reaction prediction model includes a first module that receives a molecular structure as input and predicts a graph-based chemical reaction, and a second module that analyzes information related to chemical reaction prediction related to the plurality of molecular structures from text data, and in the step of performing the chemical reaction prediction, the result according to the chemical reaction prediction is generated using the prediction result of the first module and the analysis result of the second module.

[0016] In an embodiment, in the step of performing the chemical reaction prediction, the result predicted through the first module is characterized by being verified using output data analyzed in the second module.

[0017] In an embodiment, the plurality of molecular structures include a first molecular structure and a second molecular structure, and in the step of obtaining a molecular graph for the plurality of molecular structures, the first molecular graph including nodes and edges corresponding to the first molecular structure is obtained by converting atoms constituting the first molecular structure into nodes and bonding relationships between atoms constituting the first molecular structure into edges, and the second molecular graph including nodes and edges corresponding to the second molecular structure is obtained by converting atoms constituting the second molecular structure into nodes and bonding relationships between atoms constituting the second molecular structure into edges.

[0018] In an embodiment, the method further comprises a step of performing labeling so that different labels are assigned to each of the plurality of molecular structures, and a step of storing, in the memory, information about the first molecular graph and the label assigned to the first molecular graph obtained through the process of obtaining the molecular graph, and information about the second molecular graph and the label assigned to the second molecular graph obtained through the process of obtaining the molecular graph.

[0019] In an embodiment, when the user query including the plurality of molecular structures to which the different labels are assigned is received, in the step of generating the answer, the plurality of molecular structures to which the different labels are assigned are processed as inputs of the chemical reaction prediction model, and an answer to the user query is generated using the result of the chemical reaction prediction obtained from the chemical reaction prediction model, and when the answer to the user query includes a specific molecular structure generated through the chemical reaction prediction model, a label is assigned to the specific molecular structure.

[0020] In an embodiment, the method further includes providing a plurality of graphic objects corresponding to each of the plurality of molecular structures to which different labels are assigned, in an area of ​​the service page where the user query is received, wherein each of the plurality of graphic objects includes the first molecular graph and the second molecular graph.

[0021] In an embodiment, based on receiving a user input for selecting one of the plurality of graphic objects, the service page is characterized in that detailed information corresponding to the graphic object selected according to the user input is provided.

[0022] In an embodiment, in the step of performing chemical reaction prediction for the plurality of molecular structures, an embedding vector corresponding to the molecular graph is obtained using information on the plurality of molecular structures, an attention operation related to interactions between atoms of the plurality of molecular structures is performed, and the embedding vector is updated based on the operation, and bond prediction and atom prediction predicted as a chemical reaction of the plurality of molecular structures are performed using the updated embedding vector, and a result of the bond prediction and a result of the atom prediction are obtained.

[0023] In an embodiment, the updated embedding vector is characterized in that it is updated by adding different biases according to the types of coupling between nodes constituting the first molecular graph and the second molecular graph to the attention score calculated based on the result of performing the attention operation.

[0024] In an embodiment, the combined prediction is performed by performing a dot-product operation using the updated embedding vector, and the atomic prediction is performed by predicting atomic properties of atoms corresponding to the updated embedding vector using an atomic property probability distribution of each of the atoms corresponding to the updated embedding vector, wherein the atomic properties include charge states of the atoms that may change during a chemical reaction process of the plurality of molecular structures.

[0025] In an embodiment, the result of the chemical reaction prediction is characterized in that it corresponds to a final result stabilized through a diffusion feedback process.

[0026] In an embodiment, the service page includes at least one of a first region in which the plurality of graphic objects corresponding to each of the plurality of molecular structures to which the different labels are assigned and detailed information about the plurality of molecular structures are provided, a second region in which an answer to the user query is provided, and a third region in which the user query is received, wherein the detailed information about the plurality of molecular structures includes at least one of a molecular structure image, a name, a physical property characteristic, and a string according to a SMILES notation for each of the plurality of molecular structures, and wherein the detailed information is extracted from the user input or acquired from at least one pre-learned predictive model, wherein the pre-learned predictive model includes at least one of a chemical reaction prediction reaction model that predicts a chemical reaction between molecular structures and a molecular property prediction model that predicts a physical property of a molecular structure.

[0027] In an embodiment, the method further comprises: receiving an editing request for the plurality of molecular structures through a service page that provides an answer to the user query; and providing an editing interface that provides an editing function for the plurality of molecular structures on the service page.

[0028] In an embodiment, the editing interface is characterized in that a molecular graph corresponding to at least one of the first molecular structure and the second molecular structure obtained through the molecular graph transformation is provided in an editable state.

[0029] In an embodiment, the editing of the plurality of molecular structures is characterized by the deletion or position change of at least one of the nodes corresponding to each of the atoms constituting each of the plurality of molecular structures and the edges indicating the bonding relationship of the atoms, the addition of a new node corresponding to a new atom, or the addition of a new edge creating a new bonding relationship for the atoms.

[0030] In an embodiment, a molecular structure among the plurality of molecular structures that has been edited is stored in the memory, a new label that specifies the edited molecular structure is assigned to the edited molecular structure, and when a user query including the new label is input to the ultra-large foundation model, the ultra-large foundation model is characterized in that it generates an answer using the edited molecular structure corresponding to the new label.

[0031] The answer generation system according to the present invention comprises a memory and at least one processor, wherein the memory and the processor cooperate to specify a plurality of molecular structures to be predicted based on a user input received through a service page, store information on the plurality of molecular structures in the memory, and obtain a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures stored in the memory, receive a user query for chemical reaction prediction related to the plurality of molecular structures through the service page, process the molecular graph for the plurality of molecular structures as an input of a chemical reaction prediction model so that a chemical reaction corresponding to the user query is predicted, perform chemical reaction prediction for the plurality of molecular structures in the chemical reaction prediction model, obtain a result of chemical reaction prediction for the plurality of structures from the chemical reaction prediction model, and generate an answer to the user query using the result of chemical reaction prediction.

[0032] According to the present invention, a program stored in a computer-readable medium, which is executed by one or more processes in an electronic device, may include instructions for performing the steps of: specifying a plurality of molecular structures to be predicted based on a user input received through a service page; storing information about the plurality of molecular structures in the memory; obtaining a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures stored in the memory; receiving a user query for chemical reaction prediction related to the plurality of molecular structures through the service page; processing the molecular graph for the plurality of molecular structures as an input of a chemical reaction prediction model so that a chemical reaction corresponding to the user query is predicted; performing chemical reaction prediction for the plurality of molecular structures in the chemical reaction prediction model; obtaining a result of chemical reaction prediction for the plurality of structures from the chemical reaction prediction model; and generating an answer to the user query using the result of chemical reaction prediction.

[0033] A method for predicting a chemical reaction in which a memory and a processor are cooperatively performed according to the present invention comprises the steps of: receiving information related to a plurality of molecular structures as an input of an encoder; obtaining an embedding vector corresponding to the plurality of molecular structures by using the information related to the plurality of molecular structures in an embedding layer of the encoder; performing an attention operation related to the interaction between atoms of the plurality of molecular structures in a multi-head self-attention layer and updating the embedding vector based on the operation; storing the updated embedding vector through the updating step in a memory, and inputting the updated embedding vector stored in the memory into a decoder; performing bond prediction and atom prediction predicted as a chemical reaction of the plurality of molecular structures by using the updated embedding vector in the decoder; and deriving a final chemical reaction result predicted from a chemical reaction of the plurality of molecular structures by using a result of the bond prediction and a result of the atom prediction. It may include an acquiring step.

[0034] In an embodiment, the step of obtaining the chemical reaction result is characterized by including a step of sampling an initial chemical reaction result using a result of the bond prediction and a result of the atomic prediction, a step of stabilizing the sampled initial chemical reaction result through a diffusion feedback process, and a step of obtaining the final chemical reaction result stabilized through the diffusion feedback process.

[0035] In an embodiment, based on the plurality of molecular structures, the method further includes a step of converting atoms into nodes and bonds between atoms into edges to obtain a molecular graph for the plurality of molecular structures, wherein the plurality of molecular structures include a first molecular structure and the second molecular structure, and the step of obtaining the molecular graph further includes a step of converting atoms constituting the first molecular structure into nodes and bonding relationships between atoms constituting the first molecular structure into edges using a pre-specified graph transformation algorithm, thereby obtaining a first molecular graph including nodes and edges corresponding to the first molecular structure, and a step of converting atoms constituting the second molecular structure into nodes and bonding relationships between atoms constituting the second molecular structure into edges using the pre-specified graph transformation algorithm, thereby obtaining a second molecular graph including nodes and edges corresponding to the second molecular structure.

[0036] In an embodiment, the information related to the plurality of molecular structures includes information about nodes and edges corresponding to the first molecular structure and information about nodes and edges corresponding to the second molecular structure, and the embedding vector is characterized in that it includes at least one piece of information about an atom type (Type), an atomic charge (Charge), a number of hydrogens (Hydrogen), a number of radical electrons (Radical), and a degree (Degree) of a node corresponding to the first molecular structure and a node corresponding to the second molecular structure.

[0037] In an embodiment, in the step of updating the embedding vector, different biases are added to the attention score calculated by the multi-head self-attention layer according to the type of connection between nodes constituting the first molecular graph and the second molecular graph.

[0038] In the embodiment, the bond type is characterized by including a single bond type, a double bond type, a triple bond type, and an aromatic bond type.

[0039] In an embodiment, the method further includes a step of extracting at least one of an adjacency matrix, a bond type matrix, a shortest path matrix, and K-hop neighbors corresponding to the first molecular graph and the second molecular graph, using nodes and edges corresponding to the first molecular graph and nodes and edges corresponding to the second molecular graph, wherein the adjacency matrix includes information on direct connections between nodes constituting the first molecular graph and the second molecular graph, the bond type matrix includes information on bond types between nodes constituting the first molecular graph and the second molecular graph, the shortest path matrix includes information on a shortest path length between nodes constituting the first molecular graph and the second molecular graph, and the K-hop neighbors information includes information on neighboring nodes within K steps for each of the nodes constituting the first molecular graph and the second molecular graph.

[0040] In an embodiment, in the step of updating the embedding vector, the output vector of the multi-head self-attention layer is input to a feed-forward neural network layer, and in the feed-forward neural network layer, the output vector of the multi-head self-attention layer is updated using at least one of the adjacency matrix, the bond type matrix, the shortest paths matrix, and the K-hop neighbors, and the output vector of the feed-forward neural network layer is specified as the updated embedding vector.

[0041] In an embodiment, the combined prediction is performed by performing a dot-product operation using the updated embedding vector, and the dot-product operation is characterized in that it is performed for each vector corresponding to each pair of atoms corresponding to the updated embedding vector.

[0042] In an embodiment, the step of performing the bond prediction includes the step of obtaining an inner product value for each of a plurality of bond types for each of the atom pairs based on the inner product operation, wherein the plurality of bond types are characterized by being associated with at least one of a single bond, a double bond, bond formation, bond collapse, and no change.

[0043] In an embodiment, the step of performing the bond prediction is characterized in that it further includes the step of generating a probability distribution for each of the plurality of bond types for each of the atom pairs using the inner product value according to the inner product operation, and the step of obtaining a transformation matrix that predicts a change in the bonding state of each of the atom pairs using the probability distribution.

[0044] In an embodiment, in the step of generating the probability distribution, a softmax function is applied to the inner product values ​​obtained for the plurality of bonding types for each atom pair, thereby generating the probability distribution for the plurality of bonding types for each atom pair.

[0045] In an embodiment, the step of performing the atomic prediction further includes a step of generating an atomic characteristic probability distribution of each atom corresponding to the updated embedding vector using a softmax output layer, and a step of predicting the atomic characteristics of the atoms corresponding to the updated embedding vector using the probability distribution, wherein the atomic characteristics include charge states of the atoms that may change during a chemical reaction process of the structure of the plurality of molecules.

[0046] In an embodiment, the diffusion feedback process is characterized by repeatedly evaluating each bond transformation of the initial chemical reaction product and removing or changing unstable bonds.

[0047] In an embodiment, the diffusion feedback process is characterized in that, in order to predict a change in the bonding state of the initial chemical reaction product, the predicted bonding transformation is evaluated for each repeatedly performed step using a transformation probability matrix and a target transformation matrix, and the final chemical reaction product is generated using an interpolation factor.

[0048] A method for predicting a chemical reaction according to the present invention, wherein a memory and a processor are cooperatively performed, comprising: a step of receiving information related to a plurality of molecular structures as input to an encoder; A step of obtaining a molecular graph using atoms as nodes and bonds as edges based on the plurality of molecular structures, a step of obtaining an embedding vector corresponding to the molecular graph using information related to the plurality of molecular structures in an embedding layer of the encoder, a step of performing an attention operation related to interactions between atoms of the plurality of molecular structures in a multi-head self-attention layer and updating the embedding vector based on the operation, a step of storing the updated embedding vector through the updating step in a memory, and a step of inputting the updated embedding vector stored in the memory to a decoder, a step of performing bond prediction and atom prediction predicted as a chemical reaction of the plurality of molecular structures using the updated embedding vector in the decoder, a step of obtaining a final chemical reaction result predicted from a chemical reaction of the plurality of molecular structures using a result of the bond prediction and a result of the atom prediction, and an actual bond corresponding to the final chemical reaction result and the plurality of molecular structures. The method may include calculating a loss function between label data including states and atomic states, and optimizing a parameter of at least one of the encoder and the decoder to minimize the loss function.

[0049] In a chemical reaction prediction system including a memory, an encoder, a decoder and at least one processor according to the present invention, the encoder receives information related to a plurality of molecular structures, obtains a molecular graph using atoms as nodes and bonds as edges based on the plurality of molecular structures, obtains an embedding vector corresponding to the molecular graph in an embedding layer of the encoder, performs an attention operation related to interactions between atoms of the plurality of molecular structures in a multi-head self-attention layer of the encoder, and updates the embedding vector based on the operation, the processor stores the updated embedding vector through the updating step in a memory, and inputs the updated embedding vector stored in the memory to a decoder, and the decoder performs bond prediction and atom prediction predicted as a chemical reaction of the plurality of molecular structures using the updated embedding vector, and outputs a result of the bond prediction and a result of the atom prediction. By using this, the final chemical reaction result predicted from the chemical reaction of the above multiple molecular structures can be obtained.

[0050] A program stored in a computer-readable medium, which is executed by one or more processes in an electronic device according to the present invention, wherein the program comprises: a step of receiving information related to a plurality of molecular structures as an input of an encoder; a step of obtaining a molecular graph using atoms as nodes and bonds as edges based on the plurality of molecular structures; a step of obtaining an embedding vector corresponding to the molecular graph in an embedding layer of the encoder; a step of performing an attention operation related to interactions between atoms of the plurality of molecular structures in a multi-head self-attention layer and updating the embedding vector based on the operation; a step of storing the embedding vector updated through the updating step in a memory, and inputting the updated embedding vector stored in the memory into a decoder; a step of performing bond prediction and atom prediction predicted as a chemical reaction of the plurality of molecular structures using the updated embedding vector in the decoder; and a step of performing the bond prediction of the bond prediction. The method may include instructions for performing a step of obtaining a final chemical reaction result predicted from a chemical reaction of the plurality of molecular structures using the results and the results of the above atomic prediction.

[0051] As discussed above, according to the method and system for generating answers according to the present invention, an answer suitable for a user's query is generated and provided based on data extracted from a document, thereby allowing the user to be suggested an optimal research method and minimize the risk of research failure.

[0052] Furthermore, the answer generation method and system according to the present invention can provide answers to user queries using data extracted from documents or generated from a pre-trained predictive model. This allows users to quickly and accurately receive the information they need, while reducing the time and costs required for research and / or development.

[0053] Furthermore, the answer generation method and system according to the present invention can generate answers to user queries using predicted results from a pre-trained predictive model and provide the generated answers to the user. This allows users to shorten the time required for research and / or development and reduce the number of trial and errors in research and / or development.

[0054] Furthermore, according to the answer generation method and system according to the present invention, the extracted molecular structure and related data are provided in a visualized manner through a user interface, thereby enabling the user to intuitively recognize the information required and understand it more quickly, thereby increasing the accuracy and efficiency of research.

[0055] Meanwhile, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, the movement of electrons is modeled through the graph structure of molecules, thereby enabling a good understanding of the chemical reaction mechanism and accurate prediction of the chemical reaction result.

[0056] Furthermore, the chemical reaction prediction system and its control method according to the present invention, as well as the learning method for the chemical reaction prediction system, provide a chemical reaction prediction model that enables a thorough understanding of chemical reaction mechanisms and accurate prediction of chemical reaction results, thereby reducing the time and cost required for experiments by users. This reduces research and development costs and shortens the time to market for new products.

[0057] Meanwhile, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, both the input of the molecular structure and the output of the predicted result are performed in the graph space, thereby enabling forward reaction prediction to be performed regardless of the SMILES permutation and order.

[0058] Furthermore, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, it is possible to simultaneously sample multiple interdependent transformations occurring in parallel within a molecular graph, thereby ensuring consistency between transformations.

[0059] Furthermore, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, in order to solve problems that may arise due to a symmetrical structure, symmetry is broken by including a noise or sampling mechanism, and a valid output structure is formed, thereby preventing the occurrence of an invalid configuration.

[0060] FIG. 1 is a conceptual diagram for explaining a chemical reaction prediction system and a control method thereof according to the present invention, and an answer generation system to which a learning method of the chemical reaction prediction system is applied.

[0061] FIG. 2a, FIG. 2b, FIG. 2c, FIG. 2d, FIG. 2e and FIG. 2f are conceptual diagrams for explaining a chemical reaction prediction model according to the present invention.

[0062] Figure 3 is a flowchart for explaining a learning method of a chemical reaction prediction model according to the present invention.

[0063] Figures 4a, 4b, 4c and 4d are conceptual diagrams for explaining a chemical reaction prediction model according to the present invention.

[0064] FIG. 5a, FIG. 5b, FIG. 5c, FIG. 5d, FIG. 5e, FIG. 6 and FIG. 7 are conceptual diagrams for explaining examples of use in an answer system to which a chemical reaction prediction model according to the present invention is applied.

[0065] Figures 8 and 9 are conceptual diagrams for explaining a super-large foundation model according to the present invention.

[0066] Figure 10 is a flowchart illustrating a method for generating an answer according to the present invention.

[0067] FIG. 11, FIG. 12, FIG. 13, FIG. 14, FIG. 15, FIG. 16, FIG. 17, FIG. 18, FIG. 19, FIG. 20, FIG. 21, FIG. 22, FIG. 23, FIG. 24, FIG. 25, FIG. 26, FIG. 27, FIG. 28, FIG. 29, FIG. 30, FIG. 31, and FIG. 32 are conceptual diagrams for explaining an answer generation method according to the present invention.

[0068] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Regardless of the drawing numbers, identical or similar components will be given the same reference numbers, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably only for the convenience of writing the specification, and do not in themselves have distinct meanings or roles. In addition, when describing the embodiments disclosed in this specification, if it is determined that a specific description of a related known technology may obscure the gist of the embodiments disclosed in this specification, a detailed description thereof will be omitted. In addition, the attached drawings are only intended to facilitate easy understanding of the embodiments disclosed in this specification, and the technical ideas disclosed in this specification are not limited by the attached drawings, and should be understood to include all modifications, equivalents, and substitutes included in the spirit and technical scope of the present invention.

[0069] Terms that include ordinal numbers, such as first, second, etc., may be used to describe various components, but the components are not limited by these terms. These terms are used solely to distinguish one component from another.

[0070] When a component is referred to as being "connected" or "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but that there may be other components intervening. Conversely, when a component is referred to as being "directly connected" or "connected" to another component, it should be understood that there are no other components intervening.

[0071] Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0072] In this application, terms such as “include” or “have” are intended to specify the presence of a feature, number, step, operation, component, part or combination thereof described in the specification, but should be understood not to exclude in advance the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

[0073] The present invention relates to a method and system for generating answers. The answer generation system according to the present invention performs answer generation based on generative artificial intelligence (Generative AI) or a Foundation model, and may also be referred to as an answer generation platform based on a large Foundation model. However, the “large Foundation model” may also be referred to as a generative model, a Foundation model, or a Large Language Model (LLM). The answer generation system according to the present invention may be a system that generates property prediction results for molecular structures or designs molecules having characteristics desired by a user. In addition, the answer generation system according to the present invention may be a system that generates prediction results for new types of molecules and / or chemical reactions between multiple molecules. Furthermore, the answer generation system according to the present invention may be a system that generates prediction results for modifications of existing materials and synthesis of various materials (e.g., new materials, polymer materials, nanomaterials, composite materials, organic materials, pharmaceutical materials, etc.).

[0074] The answer generation system according to the present invention includes a super-large foundation model (or a super-large foundation artificial intelligence model), and the present invention aims to increase the efficiency of natural science research by minimizing the risk of research failure.

[0075] Hereinafter, we will look into it in more detail with the attached drawings. Fig. 1 is a conceptual diagram for explaining a chemical reaction prediction system and a control method thereof according to the present invention, and an answer generation system to which a learning method of the chemical reaction prediction system is applied, and Figs. 2a, 2b, 2c, 2d, 2e, and 2f are conceptual diagrams for explaining a chemical reaction prediction model according to the present invention. Fig. 3 is a flowchart for explaining a learning method of a chemical reaction prediction model according to the present invention, and Figs. 4a, 4b, 4c, and 4d are conceptual diagrams for explaining a chemical reaction prediction model according to the present invention, and Figs. 5a, 5b, 5c, 5d, 5e, 6, and 7 are conceptual diagrams for explaining an example of use in an answer system to which a chemical reaction prediction model according to the present invention is applied. Figs. 8 and 9 are conceptual diagrams for explaining a super-large foundation model according to the present invention, and Fig. 10 is a flowchart for explaining a method of generating an answer according to the present invention. FIG. 11, FIG. 12, FIG. 13, FIG. 14, FIG. 15, FIG. 16, FIG. 17, FIG. 18, FIG. 19, FIG. 20, FIG. 21, FIG. 22, FIG. 23, FIG. 24, FIG. 25, FIG. 26, FIG. 27, FIG. 28, FIG. 29, FIG. 30, FIG. 31, and FIG. 32 are conceptual diagrams for explaining an answer generation method according to the present invention.

[0076] Meanwhile, as illustrated in FIG. 1, the answer generation system (100) may include at least one of an input unit (110), an output unit (120), a communication unit (130), a storage unit (140), and a super-large foundation model (200). Here, the super-large foundation model (200) may also be referred to as a foundation model, and the foundation model may mean a super-large AI core base model trained with a massive dataset.

[0077] Although not shown, the answer generation system (100) may include one or more processors, which may include one or more general-purpose processors and / or one or more special-purpose processors (e.g., digital signal processors, tensor processing units (TPUs), graphics processing units (GPUs), neural network processing units (NPUs), application-specific integrated circuits, application-specific integrated circuits (ASICs), etc.). The one or more processors may be configured to execute instructions, computer-readable instructions, and / or other instructions described herein stored (or included) in the storage (140). The answer generation system and method may enable the memory and at least one processor to cooperate to perform the data processing described below. The processor may perform a series of operations and data processing using data and information stored in the memory. In this case, the memory may be a component of the storage (140).

[0078] Meanwhile, the input unit (110) may be configured as a means for data input and may be configured in various types. For example, the input unit (110) may be configured to receive user input. The input unit (110) may be configured to receive user input from a user terminal (10). Here, “receiving input” may mean receiving an input signal (or selection signal) corresponding to the user’s input based on the input being made by the user through the input unit configuration provided in the user terminal (10).

[0079] The input unit (110) may also be referred to as a user interface module. The input unit (110) may include a touch screen, a computer mouse, a keyboard, a keypad, a touch pad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not place any limitations on the type of input unit (110). Furthermore, the input unit (110) in the present invention does not necessarily refer to a hardware means, but can be understood as a channel for receiving input from a user.

[0080] Here, the user input may include documents, text, images (or videos), voice, etc. In this case, the answer generation system (100) may further include a module that converts voice into text.

[0081] Next, the output unit (120) can output information through an output unit configuration (e.g., a display unit, a touch screen, a speaker, etc.) provided in a user terminal (10) linked to the answer generation system (100). For example, the output unit (120) can output a page (or service page, 1000) linked to the answer generation system (100) to the display unit of the user terminal (10). In addition, the output unit (120) does not necessarily mean a hardware means, and can be understood as a channel for outputting results to the user.

[0082] Next, the communication unit (130) may be connected to a user terminal (10), a server (e.g., a central server, an external server, etc.), a device, and at least one network through a wireless or wired network, and may be configured to receive or transmit overall data and information necessary for the operation of the answer generation system (100).

[0083] Here, the user terminal (10) may include at least one of a mobile phone, a smart phone, a notebook computer, a laptop computer, a slate PC, a tablet PC, an ultrabook, a desktop computer, a digital broadcasting terminal, a personal digital assistant (PDA), a portable multimedia player (PMP), a navigation device, and a wearable device (e.g., a smartwatch, a smart glass, a head mounted display (HMD)).

[0084] Furthermore, the communication unit (130) can support various communication methods according to the communication standards of the communicating device.

[0085] For example, the communication unit (130) may be configured to communicate with a communication target using at least one of WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth™, RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus) technologies.

[0086] Meanwhile, the storage unit (140) serves to store various data related to the present invention, and may include one or more non-transitory computer-readable storage media that can be read and / or accessed by at least one of the one or more processors (140).

[0087] The one or more computer-readable storage media may include volatile and / or non-volatile storage components, such as optical, magnetic, organic, or other memory or disk storage devices. In some examples, the storage (140) may be implemented using a single physical device (e.g., a single optical, magnetic, organic, or other memory or disk storage device), while in other examples, the storage (140) may be implemented using two or more physical devices.

[0088] The storage unit (140) may include computer-readable instructions and additional data. The storage unit (140) may include storage necessary to perform at least some of the methods, scenarios, and techniques described herein and / or at least some of the functions of the devices and networks.

[0089] Furthermore, at least a portion of the storage unit (140) may be a cloud storage or a cloud server. At least a portion of data corresponding to user input received from the input unit (110) and learning data may be stored in the storage unit (140).

[0090] That is, it can be understood that the storage unit (140) is sufficient as a space in which information necessary for the operation of the answer generation system (100) is stored, and there are no restrictions on the physical space.

[0091] Meanwhile, the super-large foundation model (200) can be configured to generate property prediction results for molecular structures or to design molecules with desired characteristics by the user. Furthermore, the super-large foundation model (200) can generate prediction results for new types of molecules and / or chemical reactions between multiple molecules, or to generate prediction results for the modification of existing materials and the synthesis of various materials (e.g., new materials, polymer materials, nanomaterials, composite materials, organic materials, pharmaceutical materials, etc.).

[0092] In this regard, the super-large foundation model (200) may include at least one of a document understanding model (300), a chemical reaction prediction model (400), and a molecular property prediction model (500).

[0093] A document understanding model (300) can extract various types of content that satisfy preset content criteria from documents (e.g., papers, books, patent documents, reports, etc.). More specifically, the document understanding model (300) can be a model trained to understand structured data, unstructured data, linguistic data (or linguistic elements), and non-linguistic data (or non-linguistic elements) contained in a document, and extract various contents (or data) and knowledge based on the understood contents.

[0094] Here, the preset content criteria can be set in various ways and can be determined according to the purpose or use of the answer generation system (100) according to the present invention.

[0095] For example, if the purpose of utilizing the answer generation system (100) is chemistry, bio, new materials, new substances, and new drug development, the document understanding model (300) can be trained to understand and extract content related to chemistry, bio, new materials, new substances, and new drug development from the document to be analyzed.

[0096] In this case, the preset content criteria may include content related to molecular structures related to at least one of chemistry, biology, new materials, new substances, and new drug development. Here, the document understanding model (300) can extract content related to chemistry, biology, new materials, new substances, and new drug development from the analysis target document based on the preset content criteria.

[0097] For convenience of explanation, in this specification, the established content criteria are explained as being related to at least one of chemistry, bio, new materials, new substances, and new drug development, but are not necessarily limited thereto.

[0098] The document understanding model (300) can extract at least one of text, molecular structure, formula, chart, table, and image that satisfies the preset content criteria from the document to be analyzed based on the preset content criteria.

[0099] For example, the document understanding model (300) can understand the chemical structure of the molecular structure included in the analysis target document (20), and based on the result of the understanding, can convert the molecular structure into a SMILES string expression and extract it. In addition, the document understanding model (300) can understand the chemical structure of the molecular structure, and based on the result of the understanding, can perform a graph transformation corresponding to the molecular structure.

[0100] Meanwhile, in the present invention, receiving "molecular structure" can be understood as receiving information that can identify a molecule. In this case, the information that can identify a molecular structure can take various forms, such as a molecular structural formula, a molecular graph, a chemical formula, a molecular structural formula according to the SMILES notation, or a molecular structure image.

[0101] In addition, the document understanding model (300) can understand text (23) related to the molecular structure (21) among the texts included in the document to be analyzed (20) and extract it as text data (23) related to the molecular structure (21).

[0102] Furthermore, the document understanding model (300) can recognize rows and columns forming a table (24) associated with a molecular structure (21) from the document to be analyzed (20), and extract them by converting them into structured data (25) in a format such as HTML or Excel.

[0103] Furthermore, the document understanding model (300) can extract relationship information (or relationships) between molecular structures included in the document.

[0104] As discussed above, the document understanding model (300) can extract various types of data contained in a document by converting them into data in a form (e.g., machine-readable data) that the super-large foundation model (200) can understand. The data extracted using the document understanding model (300) can be sorted by page or document unit and stored in a storage unit (or memory, 140). In the present invention, the document understanding model (300) can also be referred to as a deep document understanding model.

[0105] Next, the chemical reaction prediction model (400) can be configured to predict a chemical reaction between input molecular structures (or compounds or reactants) and output (or generate) a predicted chemical reaction result (or product). For example, as illustrated in FIG. 2A, when information on the molecular structure (401a, 401b) of a specific compound is input, the chemical reaction prediction model (400) can predict a product (402) formed by a chemical reaction between the molecular structure (401a, 401b) of the specific compound.

[0106] The chemical reaction prediction model (400) according to the present invention can model the movement of electrons through the graph structure of molecules. The chemical reaction prediction model (400) according to the present invention can be an electron flow-based prediction model using graph diffusion that can well understand the chemical reaction mechanism and accurately predict the results of a chemical reaction. The chemical reaction prediction model (400) can express both inputs and outputs in a graph space. The chemical reaction prediction model (400) can output a reaction product and / or byproduct prediction result based on the graph structure of an input compound by learning the graph transformation of a chemical reaction. The chemical reaction prediction model (400) in the present invention can also be called an electron flow model, an electron flow-based forward reaction prediction model, or an electron-flow inspired graph diffusion model for interpretable forward reaction prediction using a graph diffusion model that simulates electron flow.

[0107] Meanwhile, the chemical reaction prediction model (400) discussed above may include at least one of an encoder (410) and a decoder (420).

[0108] The encoder (410) according to the present invention can be configured to extract the characteristics of each atom and bond through a graph representation of a molecular structure and generate an embedding vector based on the extracted characteristics. In addition, by performing an attention operation related to the interaction between atoms of the plurality of molecular structures through a neural network, the characteristics of the molecular structure can be better understood. The decoder (420) can perform bond prediction and atom prediction predicted as chemical reactions of the plurality of molecular structures using the embedding vector obtained from the encoder (410). The chemical reaction prediction model (400) can perform a diffusion feedback process to stabilize the chemical reaction results according to the bond prediction and atom prediction, thereby generating the final chemical reaction results. The configuration and detailed process of the chemical reaction prediction model (400) according to the present invention will be described in more detail later.

[0109] Meanwhile, the chemical reaction prediction model (400) according to the present invention can utilize not only molecular structural information but also various chemical information such as reaction conditions, catalysts, and temperature to more accurately predict chemical reactions. This chemical information may include molecular structural data (e.g., molecular structure graph) and descriptive data (e.g., text data). Since chemical reactions are affected by various factors such as molecular structural changes, reaction mechanisms, reaction conditions, and reaction environments, more accurate predictions can be possible only when these factors are comprehensively considered. Meanwhile, such chemical information may be included in various literature, textbooks, theses, patent literature, articles, and academic journals, and the chemical reaction prediction model (400) according to the present invention can perform more accurate chemical reaction predictions by integrating such chemical information.

[0110] To this end, as illustrated in FIG. 2b, the chemical reaction prediction model (400) may include at least one of a first module (400a) and a second module (400b). Here, the first module (400a) may also be named a “ChemExpert-Graph” module, a graph module, a graph processing module, a graph model, a graph processing model, etc., and the second module (400b) may also be named a “ChemExpert-Text” module, a text module, a text processing module, a text model, a text processing model, etc.

[0111] The first module (400a) can receive a molecular (or chemical) structure as input and predict a graph-based chemical reaction. For example, referring to (a) of FIG. 2d, an example of a chemical reaction can be seen. Furthermore, as illustrated in (b) of FIG. 2d, the first module (400a) can include multiple layers (400a-1) to predict a graph-based chemical reaction. More specific details regarding the multiple layers (400a-1) will be described later.

[0112] The molecular structure (403a, or molecular structural formula) input to the first module (400a) is converted into a molecular graph in the form of a graph, and in the molecular graph, atoms can be expressed as nodes and bonds can be expressed as edges.

[0113] The molecular structure (403a) may correspond to at least one of data extracted from a document (403) including a molecular structure (411) using a document understanding model (300) or information extracted through a storage unit (or memory, 140).

[0114] The first module (400a) can analyze changes in the structural properties of a molecule based on an input molecular graph, predict a chemical reaction path and a product to be generated as a result of the chemical reaction, and output the predicted chemical reaction path and product.

[0115] In one embodiment, the first module (400a) can analyze structural changes of a molecule based on a molecular graph to predict the process in which a specific bond is broken and a new bond is formed.

[0116] In another embodiment, the first module (400a) can analyze the interactions between atoms within a molecule based on a molecular graph and predict radical formation and bonding changes that may occur at each step.

[0117] That is, the first module (400a) can be configured to receive a molecular graph as input and output a chemical reaction path and product (404a) predicted based on the molecular graph.

[0118] Next, the second module (400b) may be configured to process text data (403b) to understand and predict the reaction mechanism. In this case, the text data (403b) may be data extracted from a document containing a molecular structure (403a) using a document understanding model (300), or may correspond to at least one piece of information extracted through a storage unit (or memory, 140) related to the molecular structure (411). The second module (400b) may be a model that has learned a large amount of data related to chemical reactions.

[0119] In one embodiment, the text data (403b) input to the second module (400b) is data including a description of the molecular structure (403a), and may include at least one of chemical reaction conditions, chemical reaction mechanism (or reaction path), and chemical properties of the molecular structure (403a).

[0120] The second module (400b) can analyze the input text data (403b) to understand and predict the chemical reaction mechanism. More specifically, the second module (400b) can analyze the input text data (403b) and output at least one of the chemical reaction conditions, chemical reaction mechanism, and chemical properties predicted based on the text data (403b).

[0121] In one embodiment, the second module (400b) may analyze text data (403b) using natural language processing (NLP) technology and extract at least one text among chemical reaction conditions, chemical reaction mechanism (or reaction path), chemical properties, and experimental data included in the text data (403b).

[0122] In another embodiment, the second module (400b) can predict a chemical reaction mechanism (e.g., how a specific catalyst or condition affects a reaction) based on text extracted through analysis of text data (403b), and output the predicted chemical reaction mechanism and chemical properties.

[0123] The second module (400b) can analyze information related to prediction of chemical reactions related to the plurality of molecular structures from text data.

[0124] The chemical reaction prediction model (400) can combine the output data (404a) of the first module (400a) and the output data (404b) of the second module (400b) to output the final chemical reaction prediction result (e.g., product, chemical reaction path, chemical reaction mechanism, etc.).

[0125] In one embodiment, as illustrated in (a) to (c) of FIG. 2c, the chemical reaction prediction model (400) can generate electron flow, reaction conditions, and structural effects of a molecular structure (or chemical structure) using output data from the first module (400a) and the second module (400b). In this case, the electron flow, reaction conditions, and structural effects can be displayed together as graphs and text, and a molecular structure reflecting the positions before and after electron movement can be generated, or the molecular structure of a product generated according to the reaction conditions can be generated.

[0126] That is, the chemical reaction prediction model (400) can make more accurate predictions than using only a single data source by fusing and utilizing output data (404a, 404b) output from each of the first module (400a) and the second module (400b), and can enable users to intuitively recognize various elements related to chemical reactions.

[0127] In addition, the chemical reaction prediction model (400) can verify the chemical reaction result predicted by the first module (400a) using the output data analyzed by the second module (400b). That is, the second module (400b) can obtain at least one of the chemical reaction conditions, chemical reaction mechanism, and chemical characteristics analyzed based on the text data (403b). The chemical reaction prediction model (400) can verify whether the chemical reaction result predicted and obtained by the first module (400a) matches experimental data or theoretical expectations based on the data analyzed by the second module (400b).

[0128] Meanwhile, the chemical reaction prediction model (400) according to the present invention is characterized by performing chemical transformation based on a molecular graph, and for understanding, the composition of the molecular structure graph will be briefly examined.

[0129] Figure 2e (a) represents nodes and edges, and in a molecular graph, atoms can be represented as nodes and bonds between atoms as edges. For example, as shown in Figure 2e (b), a water molecule has a molecular formula of “H2O”, and at this time, the atoms are composed of H, H, and O. In this case, there are three nodes, and as shown in Figure 2e (c), the atoms H, H, and O can be represented by modes n1, n2, and n3, respectively. In addition, there can be two bonds between atoms, OH and OH, and these bonds can be represented by edges (e1, e2), as shown in Figure 2e (c). In this way, when a molecular structure is converted into a molecular graph, the unique positional and topological relationships of the molecular structure can be preserved, enabling more accurate prediction. In the present invention, a molecular structure to be predicted is converted into a molecular graph and input into an encoder (410) to predict a chemical reaction.

[0130] Meanwhile, in the present invention, the encoder (410) can embed a molecular structure into a vector using a molecular graph. At this time, briefly looking at the vector, as illustrated in (a) of Fig. 2f, an ethanol (C2H5OH) molecule can represent ethanol through a vector having a specific dimension, as illustrated in (b) of Fig. 2f, and for example, a vector corresponding to each atom of ethanol can be expressed as shown in the figure. At this time, the dimension of the vector and the information included can be set in various ways. For example, the figure represents a five-dimensional vector, and each vector can include atom type, bonding information, charge information, hybridization information, directionality information, etc. Meanwhile, although not illustrated, the encoder (410) can embed a vector for the bonding information of the molecule and utilize it for analysis.

[0131] Below, a method for predicting chemical reactions and obtaining chemical reaction results based on molecular graphs using a chemical reaction prediction model (400) according to the present invention will be described in more detail. The description below can be performed in the first module (410) illustrated in FIG. 2b.

[0132] Chemical reaction prediction and organic synthesis are key challenges in drug discovery and / or materials science. Predicting the outcome of chemical reactions is crucial for designing novel molecules and can significantly contribute to shortening product development cycles in various industries. Traditionally, sequence-based models using SMILES strings have been utilized to predict chemical reactions and / or retrosynthetic pathways. SMILES strings encode chemical structures (or molecules or molecular structures) as ASCII strings, allowing them to be processed as text data using natural language processing techniques.

[0133] However, SMILES strings do not match the natural graph representation of molecules, which reduces learning efficiency and can generate chemically invalid transformations (or transformations). Specifically, for chemical reactions, transformations in SMILES space may not directly correspond to valid molecular graphs, resulting in chemically infeasible outputs. In other words, these transformations have the limitation that they do not always lead to valid transformations in graph space, a more natural representation of molecules.

[0134] To address these issues, various graph-based (or graph-centric) approaches have been proposed to represent molecules as graphs. Graphical representations can accurately represent molecular structures, enhance the interpretability of chemical reaction mechanisms, and facilitate more accurate predictions of chemical reactions based on electron flow theory.

[0135] Accordingly, the present invention proposes a chemical reaction prediction model based on electron flow using graph diffusion, which models the movement of electrons through the graph structure of molecules, thereby enabling a thorough understanding of chemical reaction mechanisms and accurate prediction of chemical reaction results. The prediction method according to the present invention may also be called an interpretable forward reaction prediction method using a graph diffusion model that simulates electron flow (Electron-Flow Inspired Graph Diffusion Model for Interpretable Forward Reaction Prediction).

[0136] The chemical reaction prediction model described in the present invention utilizes graph diffusion, with both input and output operating in graph space. This maintains invariance with respect to the permutation and order of SMILES strings, while providing users with more interpretable transformations. Rather than starting with random graph sampling, the chemical reaction prediction model according to the present invention can learn the transformation process from an initial set of graphs to the final product.

[0137] Below, the present invention will be examined in more detail with reference to the attached drawings.

[0138] As illustrated in FIG. 4A, the chemical reaction prediction model (400) according to the present invention may include an encoder (410), a decoder (420), a graph diffusion module (430), and a learning unit (440). As illustrated, the chemical reaction prediction model (400) may perform a series of data processing according to the present invention in cooperation with a memory and at least one processor. Various information may be stored in the memory. For example, the memory may store various information such as information input to the chemical reaction prediction model (400), results generated according to a series of data processing in the chemical reaction prediction model (400), information on intermediate products, and information on final results. The processor may cooperate with each component of the chemical reaction prediction model to allow the process for chemical reaction prediction according to the present invention to proceed in each component and neural network layer.

[0139] In the chemical reaction prediction model (400) according to the present invention, during the learning stage, as shown in FIGS. 4a and 4b, a learning unit (440) and a ground-truth product exist, but after learning is completed, as shown in FIGS. 4c and 4d, the learning unit and the ground-truth product may be excluded.

[0140] The encoder (410) according to the present invention may be configured as a backbone encoder. The encoder (410) according to the present invention is configured as an attention-based graph neural network (GNN) and can perform relative position encoding based on the shortest path between nodes (atoms). Meanwhile, the diffusion process described in the present invention does not depend on a backbone network and can be implemented with various encoders.

[0141] The encoder (410) according to the present invention may include an embedding layer (411) and a graph neural network (GNN, Graph Neural Networks, 412), as illustrated. The graph neural network may be an attention-based graph neural network or a backbone neural network.

[0142] Meanwhile, the graph neural network (412) may include a multi-head self-attention layer (413) and a feed-forward neural network layer. The multi-head self-attention layer may also be referred to as a topologically weighted multi-head self-attention layer.

[0143] The embedding layer (411) is configured to receive information related to a plurality of molecular structures and convert it into an embedding vector. The embedding layer (411) may be configured to generate a vector including atoms of the molecular structure and bonding relationships between atoms. The graph neural network (412) may receive the embedding vector embedded in the embedding layer (411), perform an attention operation related to the interaction between atoms of the plurality of molecular structures, and update the embedding vector based on the operation. At this time, the embedding layer may receive a molecular graph and obtain an embedding vector based on the molecular graph.

[0144] Meanwhile, node features contained in the embedding vector output by the embedding layer (411) may include atom type, atomic charge, the number of implicit hydrogens, and the number of radical electrons. The embedding layer (411) can encode nodes according to the molecular graph to generate an embedding vector containing information usable within the encoder. This allows the molecular graph to be fully reconstructed into individual molecules, and the node degrees can also be tokenized as input features and later used to measure the consistency between the bond prediction and the predicted atomic degrees.

[0145] The features of the atom v can be represented as (f), (g), and (h) of Fig. 5a, and can be tokenized and embedded through the embedding layer (411).

[0146] As illustrated in (g) of FIG. 5a, the features of the atom v are tokenized and embedded through the embedding layer (411), and the tokenized features of type i according to (f) of FIG. 5a may have features such as those illustrated in (g) of FIG. 5a. i is an index indicating the type of the atomic feature, and may belong to the set of type, charge, hydrogen, radical, and order. i may mean at least one of the features of type, charge, hydrogen, radical, and order, where type may mean the atom type (e.g., carbon, oxygen, etc.), charge may mean the formal charge of the atom, hydrogen may mean the number of implicit hydrogens, radical may mean the number of radical electrons of the atom, and order may mean the degree of the node (the number of other atoms bonded to the atom). These features are embedded through the embedding layer (411), as illustrated in (h) of FIG. 5a, to form an initial embedding of the atom. The initial embedding of atom v can be the average of all embeddings.

[0147] The embedding layer (411) can perform embedding for multiple molecules based on a molecular graph. The molecular graph for the molecular structure may be stored in memory, a dataset, a database, etc. Alternatively, the molecular graph for the molecular structure may be extracted by the molecular graph extraction module (450). The chemical reaction prediction model (400) may further include the molecular graph extraction module (450) or may cooperate with the molecular graph extraction module (450). Furthermore, the molecular graph extraction module (450) may of course be included as a component of the encoder (410).

[0148] A molecular graph can maintain multiple molecular structures by representing atoms as nodes and bonds between atoms as edges.

[0149] The structure of the plurality of molecules may include a first molecular structure and a second molecular structure, and these first and second molecular structures may be referred to as reactants. They correspond to the drawing symbols 401a and 401b in Fig. 4a.

[0150] The molecular graph extraction module (450) can obtain a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures using a pre-specified graph transformation algorithm.

[0151] The molecular graph extraction module (450) can obtain a first molecular graph including nodes and edges corresponding to the first molecular structure by converting atoms constituting the first molecular structure into nodes and bonding relationships between atoms constituting the first molecular structure into edges. Similarly, the molecular graph extraction module (450) can obtain a second molecular graph including nodes and edges corresponding to the second molecular structure by converting atoms constituting the second molecular structure into nodes and bonding relationships between atoms constituting the second molecular structure into edges using the specified graph transformation algorithm.

[0152] In the embedding layer (411) of the encoder (410), an embedding vector corresponding to the plurality of molecular structures can be obtained using information related to the plurality of molecular structures. At this time, the information related to the plurality of molecular structures can include information on nodes and edges corresponding to the first molecular structure and information on nodes and edges corresponding to the second molecular structure. The information related to the molecular structure can be a molecular graph extracted by the molecular graph extraction module (450) or information extracted from such a molecular graph.

[0153] The embedding vector may include at least one piece of information about the atomic type (Type), atomic charge (Charge), number of hydrogens (Hydrogen), number of radical electrons (Radical), and degree (Degree) of the node corresponding to the first molecular structure and the node corresponding to the second molecular structure. Furthermore, the embedding vector may be configured to further include information about the bonding relationship between atoms of the first and second molecular structures.

[0154] Meanwhile, when the reactants to be predicted, i.e., a plurality of molecular structures (e.g., the first molecular structure and the second molecular structure, 401a, 401b) are specified, the chemical reaction prediction model (400) can extract at least one of an adjacency matrix, a bond type matrix, a shortest path, and K-hop neighbors corresponding to the first molecular graph and the second molecular graph, respectively, by using the nodes and edges corresponding to the first molecular graph and the nodes and edges corresponding to the second molecular graph. The time point at which such information (480) is extracted is not limited and may vary.

[0155] The above adjacency matrix may include information about direct connections between nodes constituting the first molecular graph and the second molecular graph.

[0156] The above coupling type matrix may include information on coupling types between nodes constituting the first molecular graph and the second molecular graph.

[0157] The above shortest path matrix may include information about the shortest path length between nodes constituting the first molecular graph and the second molecular graph.

[0158] The above K-step neighbor information may include information about neighboring nodes within K steps for each of the nodes constituting the first molecular graph and the second molecular graph.

[0159] At least one (480) of the adjacency matrix, the bond type matrix, the shortest paths, and the K-hop neighbors extracted in this manner can be stored in memory. At least one of the adjacency matrix, the bond type matrix, the shortest paths, and the K-hop neighbors stored in the memory can be input to at least one of the embedding layer (411), the neural network (412), the multi-head self-attention layer (413), and the feedforward neural network (414).

[0160] In the embedding layer (411), when an embedding vector (v1) corresponding to a plurality of molecular structures (or reactants, for example, the first molecular structure, the second molecular structure, 401a, 401b) is obtained, the obtained embedding vector (v1) can be input to a graph neural network (412). In particular, the embedding vector (v1) can be input to a multi-head self-attention layer (413) of the graph neural network.

[0161] The multi-head self-attention layer (413) can perform attention operations related to interactions between atoms of multiple molecular structures using an embedding vector (v1) embedded based on a molecular graph, and update the embedding vector based on the operations. At this time, the embedding vector (v1) to be updated may be a vector output from the embedding layer.

[0162] The multi-head self-attention layer (413) can perform global attention-based pooling. To account for the molecular graph topology, the multi-head self-attention layer (413) can add a bias to the attention weight scores derived from the shortest path between pairs of nodes. This can act as an encoding of the relative positions between atoms. Consequently, this can be free from spectral-based encoding issues (e.g., lack of sign invariance for Laplacian matrix eigenvectors) because the distance measurement between nodes is not absolute. This encoding is computationally efficient, allowing global message passing while considering the entire graph topology. At this time, the calculation time of the shortest path can be varied.

[0163] Global attention-based pooling is a method that considers the entire topology of a graph, and the multi-head self-attention layer (413) can globally consider the correlation between each node of the graph by performing global attention-based pooling.

[0164] For example, assuming that the dimension of the input vector from the embedding layer is 128 and the number of heads in the multi-head attention is 8, each head can be input with a 16-dimensional sub-vector, which is obtained by dividing the input vector by the 128 dimension, the number of heads, 8.

[0165] Each head of the multi-head self-attention layer (413) is configured to independently calculate an attention value for a 16-dimensional sub-vector by applying a preset attention mechanism. For example, the first head may mainly reflect information of adjacent nodes in the molecular graph, and the second head may mainly reflect information of a specific combination type in the molecular graph. The 16-dimensional attention vectors calculated from multiple heads, for example, eight heads, of the multi-head self-attention layer (413) may be combined again to form a 128-dimensional vector. This combined vector has the same dimension as the embedding vector output from the embedding layer (411), but may include richer information by integrating various information extracted from each head. In the present invention, this may be expressed as an updated embedding vector.

[0166] Meanwhile, in order to reflect the shortest path between nodes in the weights, in the present invention, as discussed above, the shortest path between all pairs of nodes in the molecular graph can be calculated. The shortest path can be calculated using a preset algorithm. The multi-head self-attention layer (413) can use the calculated shortest path information to encode the relative positions between nodes. This encoding can be performed based on the shortest path distance between nodes. The multi-head self-attention layer (413) can use the shortest path information as a bias when calculating the attention weight score between nodes. For example, the multi-head self-attention layer (413) can adjust the attention weight by increasing the attention weight as the shortest path distance between two nodes is close, and decreasing it as the shortest path between two nodes is farther away. The multi-head self-attention layer (413) can add the bias derived from the shortest path between pairs of nodes to the attention weight score.

[0167] The multi-head self-attention layer (413) encodes each node of the molecular graph using attention weights adjusted based on the shortest path, thereby enabling encoding that takes into account the topology of the entire molecular graph.

[0168] Additionally, in the multi-head self-attention layer (413), a bias can be added to the attention score depending on the combination type (e.g., single, double, triple, aromatic combination).

[0169] For example, in the multi-head self-attention layer (413), different biases can be added to the attention score calculated by the multi-head self-attention layer according to the type of connection between the nodes constituting the first molecular graph and the second molecular graph.

[0170] Here, the bond type may include a single bond type, a double bond type, a triple bond type, and an aromatic bond type.

[0171] This may mean that instead of a single bias matrix for all heads, each head has its own bias matrix. For example, the first head may have a shortest-path bias, and the second head may have a bias based on a single bond. The remaining heads may use masks that focus on global attention between atoms or molecules. In the multi-head self-attention layer (413) of the present invention, these various bias matrices may be added to each layer corresponding to the multi-head. In the present invention, by adding a bias matrix to each layer corresponding to the multi-head, the attention score of each head may emphasize information masked by the bias matrix. For example, when a bias matrix based on a bond type is added, the head is masked based on the bond type and can only receive information from neighbors connected by a specific bond type. For example, a specific head may receive information from nodes connected only by a single bond.

[0172] In this way, the weights for the attention score according to the present invention, which add biases reflecting various characteristics of the molecular structure, can be calculated according to the formula and meaning shown in Fig. 5. Here, A(Qk,Kk,Vk) can mean the attention weights calculated by applying a softmax function to the attention score.

[0173] In this way, in the multi-head self-attention layer (413), when an attention operation related to the interaction between atoms of the plurality of molecular structures is performed and the embedding vector is updated based on the operation, the updated embedding vector can be stored in memory through the update step. Then, the updated embedding vector stored in the memory can be input to a feed-forward neural network layer (414). The feed-forward neural network layer (414) can receive the output of the multi-head attention layer (413) and additionally update the updated embedding vector. The feed-forward neural network layer (414) can update the output vector of the multi-head self-attention layer using at least one of the adjacency matrix, the bond type matrix, the shortest paths, and the K-hop neighbors. The above feedforward neural network layer (414) can update the output vector of the multi-head self-attention layer to include more structural information about molecular structures. The output vector of the feedforward neural network layer can be specified as the final updated embedding vector (v2) input to the decoder (420).

[0174] Furthermore, the decoder (420) according to the present invention may include at least one projection layer (Projection Layer, 421). In the decoder (420), bond prediction and atom prediction based on a plurality of molecular structures (reactants) may be performed based on the updated embedding vector (v2) input to the projection layer. Bond prediction is to predict bonds between atoms, and atom prediction is to predict state changes of atoms. The decoder (420) may include a plurality of projection layers (422a, 422b). In each projection layer (422a, 422b), bond prediction (Bond Prediction, 423) and atom prediction (Atom Prediction, 424) may be performed.

[0175] In the decoder (420), joint prediction can be achieved by performing a dot product operation using the updated embedding vector. The dot product operation can be performed for each vector corresponding to each pair of atoms corresponding to the updated embedding vector.

[0176] In block 423 of the drawing, each box on the x-axis and y-axis (423a, 423b) conceptually represents each atom of the molecules to be reacted, and these atoms can be represented by the updated embedding vector (v2). The decoder (420) can predict the bonding type between each atom through an inner product operation based on the information contained in the updated embedding vector.

[0177] The decoder (420) can obtain inner product values ​​for each of multiple bond types for each atom pair based on the inner product operation. The multiple bond types can be associated with at least one of single bond, double bond, bond formation, bond collapse, and no change.

[0178] The decoder (420) can generate a probability distribution for each of the plurality of bonding types for each atom pair using the inner product value according to the inner product operation. In addition, the decoder (420) can obtain a transformation matrix that predicts the change in the bonding state of each atom pair using the probability distribution.

[0179] The decoder (420) can generate the probability distribution for the plurality of combination types for each atom pair by applying a softmax function to the inner product values ​​obtained for the plurality of combination types for each atom pair.

[0180] The decoder (420) can generate an atomic characteristic probability distribution of each atom corresponding to the updated embedding (v2) vector using a softmax output layer. Furthermore, the decoder (420) can predict the atomic characteristics of the atoms corresponding to the updated embedding vector using the probability distribution. The atomic characteristics may include charge states of the atoms, which may change during a chemical reaction process of the structure of the plurality of molecules.

[0181] By performing an inner product between the updated embedding vector (v2, or atomic embedding vector, as shown in (a) of Fig. 5b) for the decoder (420) atoms, a score for the potential combined transformation can be calculated, as shown in (b) of Fig. 5. Here, the lowercase letter t can represent a step of the diffusion process.

[0182] This is based on the standard query-key matrix multiplication of the transformer attention mechanism, but in the present invention, the key and the query can share the same projection head. The setting that the key and the query have the same projection head may require multiple projections to accommodate various bond types such as none, single bond, double defect, triple bond, aromatic, etc. This corresponds to the multi-head attention in the transformer, and if expressed in a formula, it can be represented as (c) of Fig. 5b. In this way, the bond prediction can be performed in the Bond prediction heads. This can correspond to the first projection layer (projection layer) 423a illustrated. That is, the bond prediction can be performed in the first projection layer (423a). The first projection layer (423a) can be composed of layers that constitute the bond prediction head.

[0183] Each joint type prediction can be viewed as a head in this multi-head attention scheme, where Softmax regularization is applied to a state such as (d) in Fig. 5b to obtain a distribution of joint transformations for each atom pair. Here, (d) in Fig. 5b may denote a joint transformation.

[0184] In the combined prediction head, a score S is calculated for each combined type k, as shown in (c) of Fig. 5b, where the meaning of each symbol is as shown in (d) of Fig. 5b.

[0185] Meanwhile, in the combined prediction head, the probability for the combined type k of the combined transformation T according to (e) of Fig. 5d can be calculated as shown in (b) of Fig. 5b. In the combined prediction head, the probability distribution of the combined transformation can be obtained by applying softmax regularization as shown in (f) of Fig. 5b.

[0186] In the present invention, each combination type prediction can be understood as a head in a multi-head attention system. The decoder (420) can obtain a normalized probability distribution for the combination transformation of each atom pair by applying softmax normalization to the possible states of the combination transformation according to (e) of FIG. 5b. This can provide an efficient and scalable method for predicting combination transformations directly in the embedding space.

[0187] Next, atom prediction can be performed in the second projection layer (423b) of the decoder (420). The second projection layer (423b) can be composed of layers that constitute an atom prediction head. Since the characteristics of some atoms may change during a chemical reaction, the decoder (420) according to the present invention can perform predictions on atoms in addition to predicting bond transformations between atoms. Atom prediction can be performed in the second projection layer (423b). For this purpose, a standard softmax output layer can be used for each atomic feature. This includes the degree of the atom for each bond type (e.g., the number of single bonds the atom has), and such predictions can help ensure consistency between independent parallel samples of bond transformations and atomic transformations.

[0188] In the atomic prediction head, as illustrated in (c) of Fig. 5c, the probability that the feature Av of the atom v becomes a at step t of the diffusion process can be calculated. Here, (b) of Fig. 5c may represent the embedding of the atom v, and (d) of Fig. 5c may represent the weight matrix of the prediction head for the atomic feature a. In the atomic prediction head, the probability that the atomic feature Av becomes a can be calculated by multiplying the embedding of the atom v (h, see (b) of Fig. 5c)) by the weight matrix (W, see (d) of Fig. 5c) and then applying the softmax function.

[0189] The decoder (420) can obtain a predicted chemical reaction result (425, Predicted Product (Probability Distribution)) predicted from a chemical reaction of a plurality of molecular structures by combining the predicted probability of the combination of atoms and the predicted probability of atoms output from the projection layers (423a, 423b). The predicted chemical reaction result from the projection layers can also be called an initial chemical reaction result.

[0190] In the present invention, the predicted chemical reaction result can be input to a graph diffusion module (430), and the graph diffusion module (430) can sample the initial chemical reaction result using the result of the bond prediction and the result of the atom prediction. The graph diffusion module (430) can stabilize the sampled initial chemical reaction result through a diffusion feedback process. The chemical reaction prediction model can obtain the final chemical reaction result stabilized through the diffusion feedback process.

[0191] In the above diffusion feedback process, the graph diffusion module (430) can repeatedly evaluate each bond transformation of the initial chemical reaction product and remove or change unstable bonds. Furthermore, the graph diffusion module (430) can use the transformation probability matrix and the target transformation matrix to evaluate the predicted bond transformation for each repeatedly performed step in order to predict the bond state change of the initial chemical reaction product, and can generate the final chemical reaction product using an interpolation factor.

[0192] In the present invention, since bond prediction is performed independently across the entire molecular graph, a method is needed to obtain a final, valid molecular structure. For example, during the prediction process, bonds may be predicted to multiple atoms or bonding rules may be violated. To address these issues, a denoising process is required to achieve a valid final state.

[0193] As illustrated in Fig. 5e, a stable final state can be created through multiple sampling and denoising processes (432) for the predicted chemical reaction results.

[0194] FIG. 5e illustrates the diffusion process from the correct data (or target label, ground truth product, 440) to the initial distribution (425), which is based on the initial distribution derived from the dot product matrix of the predicted bond transformations between all atom pairs. Here, the target label corresponding to the correct data may include the actual bonding states and atomic states corresponding to multiple molecular structures input to the encoder. Here, the initial distribution (425) refers to the initial chemical reaction product (425) output from the decoder, which may correspond to a probability distribution. Meanwhile, the diffusion process to the target label is performed during the learning process in the learning unit (440), and in the actual inference stage, a graph diffusion module (430) that has completed learning may be used. In this case, the predicted chemical reaction product (425) may be input to the graph diffusion module (430) that has completed learning, and through a denoising process, the final chemical reaction product may be obtained.

[0195] The learning unit (440) may calculate a loss function between the final chemical reaction result output from the decoder and label data (hereinafter, referred to as “target label”) including actual bonding states and atomic states corresponding to the plurality of molecular structures, and may optimize a parameter of at least one of the encoder and the decoder to minimize the loss function.

[0196] Meanwhile, during the learning process, the target label can be randomly sampled from the interpolated space, interpolated with the initial distribution. Figure 5e shows a noise graph where these random points are denoised into the target label. During prediction, the results sampled from the initial distribution are updated at each step, resulting in a stable final product molecule. A similar process can be applied to atomic features.

[0197] In the diffusion step, the predicted product may correspond to the initial distribution obtained through the dot product matrix of the joint transformation prediction for all atom pairs in the decoder (420), as discussed above. This initial distribution represents the probability distribution of the formation of joints between atoms. The interpolated product is the target label (ground-truth product) interpolated with the initial distribution. In the present invention, a noisy graph can be obtained by sampling random points in this interpolated space. These random points represent the noisy graph that is denoised to the target label. The denoised product denoises the sampled noisy graph so that it gradually gets closer to the target label. This process is performed in multiple steps, and the graph can become more stable at each step. The alternative product means that in some cases, the alternative product may be derived through a different path, which may have a different joint configuration than the target label.

[0198] In this way, in the present invention, the decoder (420) obtains an initial distribution, and the graph diffusion module (430) samples the initial distribution to sample random points. The graph diffusion module (430) interpolates the target label with the initial distribution through interpolation and denoising processes to create an intermediate state. This intermediate state (v3) can be input back to the encoder during the learning process. The graph diffusion module (430) samples random points to obtain a noisy graph, and gradually approaches the target label through a denoising process. This process can be performed repeatedly, and at least one of the encoder, the decoder, and the graph diffusion module can be learned during this repetition.

[0199] The graph diffusion module (430) can be configured to learn how to denoise the predicted chemical reaction product (425), by spreading the target label over the previous output distribution, and then learning how to map the states within the spread interpolation space to the final target label.

[0200] For this purpose, a transformation probability matrix Mt and a desired target transformation matrix T can be defined, and the transformation matrix T can be derived from the atomic bond prediction results. The atomic bond prediction results can be obtained from the decoder and derived from the bond prediction head discussed above. This can be expressed as the atom-mapped label of the reaction. The transformation matrix T can identify whether a bond is formed or broken by finding the difference between the input and output adjacency matrices, and the transformation matrix T can be expressed as in (a) of Fig. 5d. This defines the difference between the adjacency matrices of the reactants and products, and the adjacency matrix can represent the entire bond type matrix including single / double / triple / aromatic bonds.

[0201] Meanwhile, the reactants are time-dependent because they continuously change at each diffusion step. This means that the target label also changes at each step. In the present invention, a method for predicting when and where the difference between the reactants and the target label occurred (the reaction center) and, if there is a difference, what changed (the final combined configuration) can be learned, as shown in (b) of Fig. 5d. The parameter at is a randomly sampled interpolation factor that determines the weights between the current transformation matrix and the target transformation matrix. During training, the diffusion process may include obtaining the transformation probabilities Mt from the initial reactant set {Gr} and sampling {Gr(t+1)} from Mt+1 to be used as the input for the next step. Both steps may aim to directly predict the target transformation matrix T. Therefore, the training loss can be defined as shown in (c) of Fig. 5d. During prediction, one can simply encode the input {Gr}, obtain the probability of the combined transformation Mt using the prediction head, and sample from this distribution at each step using standard categorical softmax sampling. In the learning process of the present invention, this process can be repeated continuously and the results can be fed back until a stable configuration is obtained in which the combination no longer changes.

[0202] Meanwhile, in the present invention, the final chemical reaction result can be derived by applying Top-K ranking to a plurality of results stabilized in the diffusion module (430). This can be performed in the Top-K Sampled Product module of the chemical reaction prediction model (400). The Top-K ranking process can be performed in the inference stage. That is, it can be utilized to extract results with a high probability in order to provide the final result to the user. In the present invention, at least one of Top-k sampling and Top-K ranking can be performed, which can be used to obtain more diverse results in the inference process. In Top-k sampling, multiple samples can be generated based on the learned content of the chemical reaction prediction model, and the distribution can be diversified by adjusting the temperature parameter τ. Top-k ranking is used to select the most likely result among the generated samples, calculate the probability of each sample, and select the result with the highest probability as the final prediction.

[0203] In the Top-k sampling process, when sampling the distribution Mt at time step t, the logits can be rescaled by the temperature parameter τ. This can make the softmax distribution depicted in (a) of Figure 6 sharper or more uniform.

[0204] Meanwhile, in order to obtain multiple samples, the present invention can use a temperature schedule τ. In the present invention, when N samples are to be extracted, τi can be set as in (b) of Fig. 6. Here, f(i) is a temperature sampling schedule, and in the present invention, f(i) can be defined as in (c) of Fig. 6. Here, x is a linspace between 0 and 1.

[0205] Next, we examine Top-K ranking. Given sampled candidates, we need a method to rank them by probability. This can be achieved by calculating the probability from the transformation matrix Mt. If we consider the sampled states as sampled tokens for each atom and bond, we can derive the probability for each state from the sampled softmax distribution Mt. This allows us to assign a total probability to the final overall configuration of the molecule (atoms and bond states) and calculate the sampled probability (S). This can be represented as in (d) of Fig. 6. Here, (e), (f), and (g) of Fig. 6 can each be derived from the softmax distribution for atomic features, bond features, and atomic orders. Furthermore, (d) of Fig. 6 can ensure consistency between the predicted atomic order and the actual order calculated from the bonds. In the present invention, since bonds are sampled independently, multiple bonds can be formed on the same atom. Therefore, the actual order of each atom in the sampled graph is directly calculated and compared with the probability distribution of the predicted order, and in this way, a high probability score is assigned when the actual order of the generated molecule matches the predicted order.

[0206] Furthermore, since independent sampling from a distribution does not guarantee that all solutions are unique, numerous duplicate samples may result. Therefore, during prediction, the present invention selects the top-k product by taking a simple eigengraph.

[0207] Meanwhile, the above description focuses on Figures 4a and 4b, which include the learning process of a chemical reaction prediction model. However, as previously discussed, the chemical reaction prediction model can have the configurations of Figures 4c and 4d during the inference process. In this case, since everything can be understood identically except for the learning process, the detailed description will be omitted from the above description.

[0208] In the inference process, as described above and illustrated in FIG. 3, a process of receiving information related to a plurality of molecular structures including a first molecular structure and a second molecular structure as inputs of an encoder (S310), a process of obtaining an embedding vector corresponding to the first molecular structure and the second molecular structure using the information related to the plurality of molecular structures in an embedding layer of the encoder (S320), a process of performing an attention operation related to the interaction between atoms of the first molecular structure and the second molecular structure in a multi-head self-attention layer and updating the embedding vector based on the operation (S330), a process of storing the updated embedding vector through the update step in a memory, and inputting the updated embedding vector stored in the memory to a decoder (S340), a process of using the updated embedding vector in the decoder, as a chemical reaction of the first molecular structure and the second molecular structure The final chemical reaction result can be obtained through a process (S350) of performing predicted bond prediction and atom prediction, and a process (S360) of obtaining a final chemical reaction result predicted from a chemical reaction of the first molecular structure and the second molecular structure using the result of the bond prediction and the result of the atom prediction.

[0209] The process for obtaining these final chemical reaction results can be accomplished through the answer generation system (100) illustrated in FIG. 1. At this time, the answer generation system (100) can receive a user query, and the user query can include information about the molecular structure that is the target of chemical reaction prediction.

[0210] In the chemical reaction prediction model (400) according to the present invention, as shown in Fig. 7(a), a chemical reaction result can be obtained as shown in Fig. 7(b) through bond prediction and atomic prediction for a plurality of molecular structures that are the target of a chemical reaction.

[0211] As described above, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, the movement of electrons is modeled through the graph structure of molecules, thereby enabling a good understanding of the chemical reaction mechanism and accurate prediction of the chemical reaction result.

[0212] Furthermore, the chemical reaction prediction system and its control method according to the present invention, as well as the learning method for the chemical reaction prediction system, provide a chemical reaction prediction model that enables a thorough understanding of chemical reaction mechanisms and accurate prediction of chemical reaction results, thereby reducing the time and cost required for experiments by users. This reduces research and development costs and shortens the time to market for new products.

[0213] Meanwhile, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, both the input of the molecular structure and the output of the predicted result are performed in the graph space, thereby enabling forward reaction prediction to be performed regardless of the SMILES permutation and order.

[0214] Furthermore, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, it is possible to simultaneously sample multiple interdependent transformations occurring in parallel within a molecular graph, thereby ensuring consistency between transformations.

[0215] Furthermore, according to the chemical reaction prediction system and control method thereof according to the present invention, and the learning method of the chemical reaction prediction system, in order to solve problems that may arise due to a symmetrical structure, symmetry is broken by including a noise or sampling mechanism, and a valid output structure is formed, thereby preventing the occurrence of an invalid configuration.

[0216] Next, the document understanding model (300) can extract various types of content that satisfy preset content criteria from documents (e.g., papers, books, patent documents, reports, etc.). More specifically, the document understanding model (300) can be a model trained to understand structured data, unstructured data, linguistic data (or linguistic elements), and non-linguistic data (or non-linguistic elements) contained in a document, and extract various contents (or data) and knowledge based on the understood contents.

[0217] Here, the preset content criteria can be set in various ways and can be determined according to the purpose or use of the answer generation system (100) according to the present invention.

[0218] For example, if the purpose of utilizing the answer generation system (100) is chemistry, bio, new materials, new substances, and new drug development, the document understanding model (300) can be trained to understand and extract content related to chemistry, bio, new materials, new substances, and new drug development from the document to be analyzed.

[0219] In this case, the preset content criteria may include content related to molecular structures related to at least one of chemistry, biology, new materials, new substances, and new drug development. Here, the document understanding model (300) can extract content related to chemistry, biology, new materials, new substances, and new drug development from the analysis target document based on the preset content criteria.

[0220] For convenience of explanation, in this specification, the established content criteria are explained as being related to at least one of chemistry, bio, new materials, new substances, and new drug development, but are not necessarily limited thereto.

[0221] The document understanding model (300) can extract at least one of text, molecular structure, formula, chart, table, and image that satisfies the preset content criteria from the document to be analyzed based on the preset content criteria.

[0222] In one embodiment, as illustrated in FIG. 8, the document understanding model (300) can understand the chemical structure of the molecular structural formula (21) included in the analysis target document (20), and based on the result of the understanding, can convert the molecular structural formula (21) into a SMILES string expression (22) and extract it. In addition, the document understanding model (300) can understand the chemical structure of the molecular structural formula (21), and based on the result of the understanding, can perform a graph transformation corresponding to the molecular structural formula (21).

[0223] In another embodiment, the document understanding model (300) can understand text (23) related to the molecular structure (21) among the texts included in the document to be analyzed (20) and extract it as text data (23) related to the molecular structure (21).

[0224] In another embodiment, the document understanding model (300) can recognize rows and columns constituting a table (24) associated with a molecular structure (21) from a document to be analyzed (20), and extract them by converting them into structured data (25) in a format such as HTML or Excel.

[0225] Furthermore, the document understanding model (300) can extract relationship information (or relationships) between molecular structures included in the document.

[0226] In one embodiment, as illustrated in FIG. 7, among the multiple molecular structures included in the document to be analyzed (600), a third molecular structure assigned a third label (M2) may be understood as a molecular structure (or compound) generated as a result of a chemical reaction between a first molecular structure assigned a first label (M1) and a second molecular structure assigned a second label (M2). The document understanding model (300) may understand the relationship between the first molecular structure and the second molecular structure included in the document to be analyzed (600), and extract relationship information that the third molecular structure is generated through a chemical reaction between the first molecular structure and the second molecular structure.

[0227] In this case, relationship information between molecular structures can be extracted by understanding the text contained in the document to be analyzed, or by understanding non-verbal data contained in the document to be analyzed.

[0228] In one embodiment, the document understanding model (300) can understand the relationship between the first molecular structure and the second molecular structure through symbols (e.g., plus, arrow, etc.) present in one region among multiple regions included in the document to be analyzed (600) where the first molecular structure and the second molecular structure are located, and can extract relationship information that a third molecular structure is generated through a chemical reaction between the first molecular structure and the second molecular structure.

[0229] As discussed above, the document understanding model (300) can extract various types of data contained in a document by converting them into data in a form (e.g., machine-readable data) that the super-large foundation model (200) can understand. The data extracted using the document understanding model (300) can be sorted by page or document unit and stored in a storage unit (or memory, 140). In the present invention, the document understanding model (300) can also be referred to as a deep document understanding model.

[0230] Next, the molecular property prediction model (500) may be a pre-learned model based on various learning data to predict the properties of a material (or molecule) or design a material structure.

[0231] For example, learning data used for learning a molecular property prediction model (500) may be data including unique characteristic information of a material and physical property characteristic information of a material.

[0232] Here, the unique characteristic information of the material may include the name of the material, molecular structure, and / or chemical formula, etc. In addition, the physical property characteristic information of the material may include physical property (i.e., domain) values ​​such as boiling point, melting point, refractive index, solubility, viscosity, surface tension, density, strength, and / or thermal conductivity of the material.

[0233] This molecular property prediction model (500) can predict the physical properties (or physical property information) of a material or design a material having the physical properties desired by the user.

[0234] Specifically, the molecular property prediction model (500) can receive information on the unique characteristics of a material and / or information on the physical properties of a material, and output predicted data based on the input information and learned knowledge.

[0235] In one embodiment, the molecular property prediction model (500) can input unique characteristic information of a specific material and output predicted physical property information of the specific material based on the input information and learned knowledge.

[0236] In another embodiment, the molecular property prediction model (500) can receive property characteristic information of a specific material and output unique characteristic information of the specific material predicted based on the input information and learned knowledge.

[0237] In another embodiment, the molecular property prediction model (500) can receive unique characteristic information of a specific material and physical property information of a specific material, and output optimal material unique characteristic information and material property information predicted based on the input information and learned knowledge.

[0238] As discussed above, the answer generation system (100) based on the super-large foundation model (200) is intended to suggest optimal research methods to researchers in the field of natural sciences, minimize the risk of failure in natural science research, and thereby increase the efficiency of natural science research. More specifically, the present invention solves the problems of time and cost required for material research and development, and increases the efficiency of material research and development. Below, the overall process of the answer generation method and system of the super-large foundation model will be examined.

[0239] The response generation system (100) can specify an analysis target based on user input received from a user terminal (10). The user input may include at least one of a document, an image, voice, video, and text. For example, if user input regarding a document is received, the response generation system (100) can specify the document corresponding to the user input as the analysis target. The following description assumes that user input regarding a document is received.

[0240] As illustrated in FIG. 9, the answer generation system (100) can specify the analysis target document (30) to be analyzed.

[0241] In the present invention, there may be various methods (or methods or criteria) for specifying the document (30) to be analyzed.

[0242] In one embodiment, the answer generation system (100) may specify the input document as the analysis target document (30) based on the input of at least one document corresponding to the user's selection among documents stored (or built-in) in the storage (or memory or storage space or database) of the user terminal (10) in the document upload page (or interface) provided on the service page (1000).

[0243] In another embodiment, the answer generation system (100) may receive link information (e.g., URL) of a document or link information of an external storage service (e.g., Google Drive, Dropbox, etc.) storing the document from the user terminal (10). Furthermore, the answer generation system (100) may directly access the document through the link information of the document or download the document to specify the document (30) to be analyzed.

[0244] However, the method for specifying the document to be analyzed in the present invention is not necessarily limited to the above-described embodiments. For convenience of explanation, the following description assumes that a document received through a user terminal (10) on which a service page (1000) is output is specified as the document to be analyzed (30).

[0245] Once the target document (30) for analysis is specified, the answer generation system (100) can extract various types of content from the target document (30) for analysis using a document understanding model (300). Here, the various contents extracted from the target document (30) for analysis can be understood as contents that satisfy preset content criteria.

[0246] As discussed above, based on the fact that the purpose of utilizing the answer generation system (100) is chemistry, bio, new materials, new substances, and new drug development, the preset content criteria can be determined as content related to at least one of chemistry, bio, new materials, new substances, and new drug development.

[0247] Accordingly, the document understanding model (300) can extract multiple contents (31) related to chemistry, biotechnology, new materials, new substances, and new drug development from the analysis target document (30) based on preset content criteria. For example, the multiple contents (31) may include at least one of text, molecular structures, formulas, charts, tables, and images.

[0248] Furthermore, when multiple contents (31) are extracted from the document understanding model (300), the processor can store the extracted multiple contents (31) in a storage unit (or memory, 140).

[0249] Meanwhile, the answer generation system (or processor, 100) can analyze the relationships between multiple pieces of content (31) stored in the storage (or memory, 140). Here, the relationship represents a semantic association between different pieces of content (e.g., molecular structures, text, formulas, tables, etc.), and may refer to a relationship based on semantic, thematic, and / or structural similarities that connect different pieces of content. Such relationships can be analyzed based on the meaning of each piece of content.

[0250] Specifically, the answer generation system (100) can analyze (32) the relationship between the plurality of contents (31) based on the meaning of each of the plurality of contents (31). For example, let's assume that a first molecular structure, a second molecular structure, a first text, a second text, a first formula, a second formula, a first table, and a second table are extracted as the plurality of contents (31). The answer generation system (100) can, through the relationship analysis (32) of the plurality of contents (31), determine that the first molecular structure, the first text, the first formula, and the first table have a relationship with each other, and determine that the second molecular structure, the second text, the second formula, and the second table have a relationship with each other.

[0251] Furthermore, the answer generation system (100) can perform grouping of multiple contents (31). In the present invention, grouping can be performed between multiple contents (31) that have a relationship with each other.

[0252] The answer generation system (100) can group (33) related contents among the plurality of contents (31) based on the relationship between the plurality of contents.

[0253] More specifically, the answer generation system (100) can group (33) content about the same molecular structure among at least one text, molecular structure, formula, chart, table, and image included in a plurality of contents (31) into related content to generate grouped content (34).

[0254] In one embodiment, the answer generation system (100) can group (33) a first text, a first formula, and a first table containing content related to a first molecular structure among a plurality of contents (31) into content related to the first molecular structure, thereby generating grouped first content.

[0255] In another embodiment, the answer generation system (100) can group the second text, the second formula, and the second table, which contain content related to the second molecular structure among the plurality of contents (31), into content related to the second molecular structure, and generate grouped second content.

[0256] Furthermore, the answer generation system (or processor, 100) can store the grouped content in a storage (or memory, 140) in association with a user account.

[0257] Through the process discussed above, content (34) grouped based on a specific molecular structure may include at least one of a molecular structure image of a specific molecular structure corresponding to the grouped content (34), a name of the molecular structure, a description of the molecular structure, physical properties of the molecular structure (e.g., molecular weight, density, melting point, boiling point, flash point, surface tension, etc.), and a string according to the SMILES notation of the molecular structure.

[0258] Meanwhile, at least some of the content included in the content (34) grouped for a specific molecular structure may include content generated by a pre-learned prediction model.

[0259] Specifically, at least some of the content included in the grouped content (34) for a specific molecular structure may include content generated by at least one of a super-large foundation model (200), a pre-learned chemical reaction prediction model (400), and a pre-learned molecular property prediction model (500).

[0260] In one embodiment, let us assume that a plurality of contents (31) extracted from a document to be analyzed (30) contain molecular structure images and names of specific molecular structures, but no description of the specific molecular structure exists. The answer generation system (100) can generate a description of the specific molecular structure using a pre-learned chemical reaction prediction model (400). In addition, the answer generation system (100) can group (33) the molecular structure images and names of the specific molecular structures extracted from the document to be analyzed (30) and the description of the specific molecular structure generated by the chemical reaction prediction model (400) to generate grouped contents (34).

[0261] In another embodiment, let us assume that a plurality of contents (31) extracted from a document to be analyzed (30) contain molecular structure images, names, and descriptions of a specific molecular structure, and that no physical properties exist for the specific molecular structure. The answer generation system (100) can generate physical properties for the specific molecular structure using a pre-learned molecular property prediction model (500). In addition, the answer generation system (100) can group (33) the molecular structure images, names, and descriptions of the specific molecular structure extracted from the document to be analyzed (30) and the physical properties for the specific molecular structure generated by the molecular property prediction model (500) to generate grouped contents (34).

[0262] That is, the answer generation system (100) can generate content not included in the document to be analyzed using at least one of the ultra-large foundation model (200), the chemical reaction prediction model (400), and the molecular property prediction model (500), and can generate grouped content (34) including content generated by at least one of the models (200, 400, 500).

[0263] Meanwhile, the answer generation system (100) may perform labeling (35) so that at least some of the plurality of contents (31) are assigned a label. Here, at least some of the contents assigned a label among the plurality of contents (31) may correspond to contents (34) grouped through grouping (33).

[0264] In this case, grouped content (34) containing related content may be given the same label.

[0265] Specifically, the answer generation system (100) can assign the same label to grouped contents (34) through labeling (35). For example, the answer generation system (100) can assign a first label to first contents grouped based on a first molecular structure, and a second label to second contents grouped based on a second molecular structure, through labeling (35).

[0266] In this regard, as discussed above, when there are multiple grouped contents (34) in the present invention, each of the multiple grouped contents (34) may be assigned a different label (for example, a first label may be assigned to the first grouped contents, and a second label may be assigned to the second grouped contents).

[0267] However, as discussed above, the present invention has described the labeling target as grouped content, but this is not necessarily limited to this. In addition to grouped content, the present invention can also assign labels to individual pieces of content with independent meanings, thereby enabling the assignment of labels to each piece of content.

[0268] Furthermore, grouped contents with different labels can be stored in a storage unit (140) in association with a user account.

[0269] Meanwhile, the answer generation system (100) can provide grouped content (34) stored in the storage unit (140) to the user terminal (10) where the service page (1000) is output.

[0270] Specifically, the answer generation system (100) can provide a graphic object corresponding to each grouped content (34) assigned a label through labeling performance (35) to an area of ​​a service page (1000) where a user query is received (e.g., see FIG. 13).

[0271] And, the answer generation system (100) can receive a user query corresponding to the user's input through one area of ​​the service page (1000).

[0272] Here, the user query (36) may include a label (or label information) assigned to the grouped content (34), or information that can express a molecular structure (e.g., a name of the molecular structure, a chemical formula of the molecular structure, etc.).

[0273] For convenience of explanation, the following description assumes that a user query including label information (e.g., “Can you predict the chemical reaction of the first label (m1) and the second label (m2)?”, 36) has been received. However, the present invention does not limit the information included in the user query (36) to any one, and any information that can express a specific molecule (or compound or material) may be included in the user query.

[0274] The answer generation system (100) can process the user query (36) as input to the super-large foundation model (200) based on the user query (36) received.

[0275] The super-large foundation model (200) receives a user query (36) as input, understands the contents included in the user query (36), and can identify specific content related to the user query (36).

[0276] More specifically, the super-large foundation model (200) can extract labels assigned to grouped content (34) from the user query (36) through analysis of the user query (36) and specify specific grouped content (37) corresponding to the extracted labels. For example, based on the analysis result of the user query (36), which indicates that the user query (36) includes content corresponding to the first label and the second label indicating specific grouped content (37), the super-large foundation model (200) can specify the first content corresponding to the first label and the second content corresponding to the second label as specific grouped content (37) associated with the user query (36).

[0277] Furthermore, the super-large foundation model (200) can process specific content (i.e., the molecular structure of specific grouped content) as input to a pre-trained chemical reaction prediction model (400).

[0278] At this time, the super-large foundation model (200) can change the name of the molecular structure included in a specific grouped content (37) into a language that the computer can understand.

[0279] More specifically, the super-large foundation model (200) can convert (or change) the names of molecular structures included in specific grouped contents (37) into strings according to the SMILES notation, which is a computer-understandable language, and process the converted strings and information about the specific grouped contents (37) as inputs to a chemical reaction prediction model (400) that understands chemical reaction mechanisms. For example, the super-large foundation model (200) can convert the names of the first molecular structure and the second molecular structure included in each of the specific grouped contents (37) into strings according to the SMILES notation, and input the converted strings and information about the first molecular structure and the second molecular structure into a pre-trained chemical reaction prediction model (400).

[0280] The chemical reaction prediction model (400) can predict the intermolecular chemical reaction of a specific grouped content (37) and output the predicted result as output data. As described above, the chemical reaction prediction model (400), which receives the string converted from the super-large foundation model (200) and information on the first molecular structure and the second molecular structure, can predict the chemical reaction (or synthesis result) between the first molecule corresponding to the first molecular structure and the second molecule corresponding to the second molecular structure, and output the chemical reaction prediction result (38) between the first molecule and the second molecule.

[0281] In one embodiment, the predicted result of a chemical reaction may include a third molecular structure generated as a result of the chemical reaction between the first molecular structure and the second molecular structure, and may include at least one of chemical properties of the third molecular structure, reaction conditions, expected yield, reaction energy, reaction path, and expected reaction time.

[0282] Meanwhile, the super-large foundation model (200) can generate an answer (39) to a user query (36) by using the output data (chemical reaction prediction result (38)) of the chemical reaction prediction model (400) and the contents constituting the grouped content (or specific grouped content).

[0283] At this time, the super-large foundation model (200) can determine which procedures and tools to use to generate an answer (39) to a user query (36). More specifically, the super-large foundation model (200) can determine the answer generation procedure performed to make a prediction corresponding to the user query (36) and the tool used in the answer generation procedure.

[0284] In this case, the answer generation system (100) can provide information about the answer generation procedure determined from the super-large foundation model (200) and the tool used in the answer generation procedure to the service page (1000) (e.g., see FIG. 17).

[0285] The super-large foundation model (200) can execute an operation to generate an answer (39) to a user query (36) based on the determined answer generation procedure and tool. The super-large foundation model (200) can generate an answer (39) to a user query (36) using the output data of the chemical reaction prediction model (400) discussed above, the content constituting the specific grouped content (37), and the determined answer generation procedure and tool.

[0286] At this time, the answer generation system (100) can assign a new label (e.g., the third label (M3)) through labeling of the new molecular structure based on the fact that the answer (39) generated from the super-large foundation model (200) includes a new molecular structure (e.g., the third molecular structure).

[0287] Based on the generation of an answer (39) to a user query (36), the answer generation system (100) can provide the answer (39) generated from the super-large foundation model (200) through a user terminal (10) on which a service page (1000) is output.

[0288] Meanwhile, the answer generation system (100) can receive new (or additional) user inquiries through the service page (1000).

[0289] The answer generation system (100) can receive an input for a new user query (40) when a new query (40) is input from a user terminal (10) after an answer (39) for a user query (36) is provided.

[0290] For example, a new user query (40) may be a query containing content related to at least one labeled molecular structure, or a query containing content related to another unlabeled molecule. For convenience of explanation, the following description will be given on the assumption that a new user query (40) containing a specific molecular structure (e.g., a third molecular structure 41) to which a new label (e.g., a third label (M3)) has been assigned has been received.

[0291] The answer generation system (100) can input a new user query (40) into the super-large foundation model (200) based on the receipt of a new user query (40) including a label (M3) assigned to the third molecular structure (41).

[0292] The super-large foundation model (200) can utilize at least one predictive model to understand a new user query (40) and generate an answer to the new user query (40). In one embodiment, the super-large foundation model (200) can input information about a third molecular structure (41) into the molecular property prediction model (500) based on the fact that the user query (40) includes the content “Can you predict the surface tension of m3?”

[0293] The molecular property prediction model (500) can predict the property characteristics (e.g., surface tension) for a third molecular structure (41) corresponding to a new user query (40). In addition, the molecular property prediction model (500) can output a property prediction result (42) for the third molecular structure (41).

[0294] In one embodiment, the property prediction result (42) may include at least one of surface tension, boiling point and melting point, density, solubility, viscosity, thermal properties, mechanical properties, and electrical properties for the third molecular structure (41).

[0295] The super-large foundation model (200) determines an answer generation procedure and a tool used in the answer generation procedure to make a prediction corresponding to a new user query (40), and can generate an answer (e.g., “m3 surface tension is OO..., 43) for the user query (40) using the determined answer generation procedure and tool and the output data of the molecular property prediction model (500).

[0296] Furthermore, the answer generation system (100) can provide an answer (43) generated from the super-large foundation model (200) to the user terminal (10).

[0297] In this way, the present invention generates and provides answers tailored to user queries, thereby suggesting optimal research methods and minimizing the risk of research failure. Furthermore, by utilizing a pre-trained predictive model to provide predictive information in response to user queries, it supports user decision-making and provides the information users need, thereby enhancing research efficiency.

[0298] Below, based on the overall process of the system and the method of generating answers for the super-large foundation model discussed above, the method of generating answers for the super-large foundation model will be described in more detail.

[0299] First, in the present invention, a process of specifying a plurality of molecular structures to be predicted can be performed based on user input received through a service page (S1010, see FIG. 10).

[0300] The answer generation system (100) according to the present invention can be implemented in various platform forms, such as applications, software, and websites. For convenience of explanation, the present specification does not limit the form in which the answer generation system (100) is implemented. In the present invention, the answer generation system (100) may also be referred to as an "answer generation platform."

[0301] The user discussed above may have a user account pre-registered in the answer generation system (100) according to the present invention. In this case, the account may be created through a page (or screen) linked to the answer generation system (100). Alternatively, the account may also be created in at least one other system linked to the answer generation system (100). However, in this specification, the system where the user account is issued is not separately distinguished, and all accounts that can utilize the various services (or functions) provided by the answer generation system (100) according to the present invention are referred to as "accounts pre-registered in the answer generation system (100)."

[0302] Meanwhile, in the present invention, inputting a "molecular structure" can be understood as inputting information that can identify a molecule. At this time, the information that can identify a molecular structure can take various forms, such as a molecular structural formula, a molecular graph, a chemical formula, a molecular structural formula according to the SMILES notation, or a molecular structure image.

[0303] As illustrated in FIG. 11, an answer generation platform based on a super-large foundation model (200) can provide a service page (1000) linked to the platform to a user terminal (10).

[0304] The answer generation system (100) can specify a plurality of molecular structures to be predicted based on user input received through the service page (1000).

[0305] As discussed above, user input may include at least one of documents, images, audio, video, and text. The following description assumes that user input for a document is received.

[0306] The answer generation system (100) can receive user input for at least one document from a user terminal (10) provided with a service page (1000). At this time, the answer generation system (100) can receive one or more documents (i.e., multiple documents are also possible) from the user terminal (10). For convenience of explanation, in this specification, it is assumed that one document has been received.

[0307] The answer generation system (100) may provide (or display) a graphic object (601) linked to a document input function in an area of ​​a service page (1000) to receive user input regarding a document. For example, when a graphic object (601) is selected from a user terminal (10), the answer generation system (100) may activate (or output) a document upload page (or window) on the user terminal (10). The user may select a document through the document upload page or upload the document using a drag and drop method. The answer generation system (100) may receive user input regarding the document based on the input of a document corresponding to the user's selection.

[0308] However, user input for a document in the present invention is not necessarily limited to the aforementioned embodiments. For example, a user may input link information (e.g., a URL) for a document or link information for an external storage service (e.g., Google Drive, Dropbox, etc.) storing the document. In this case, the answer generation system (100) may directly access the document through the link information or download the document to receive user input for the document.

[0309] Furthermore, the answer generation system (100) can specify a document received from a user terminal (10) as a document to be analyzed. For example, as illustrated in FIG. 12, the answer generation system (100) can specify a document (600) corresponding to a user input as a target for analysis using a document understanding model (300).

[0310] Once the document to be analyzed is specified, the answer generation system (100) can specify at least one molecular structure to be predicted from the document to be analyzed (600).

[0311] Specifically, the answer generation system (100) can extract a plurality of contents (or various contents related to the molecular structure or information on the molecular structure) from the document to be analyzed (600) using the document understanding model (300), and can specify the molecular structure to be predicted based on the extracted contents.

[0312] As discussed above, the document understanding model (300) can extract multiple contents satisfying preset content criteria from at least one document. Here, the preset content criteria can be understood as contents related to molecular structures related to at least one of chemistry, biology, new materials, new substances, and new drug development. Accordingly, the answer generation system (100) can extract contents related to molecular structures related to at least one of chemistry, biology, new materials, new substances, and new drug development from the document using the document understanding model (300).

[0313] The document understanding model (300) can understand the contents included in the document to be analyzed (600) and, based on the results of the understanding, can extract various contents (or information) related to the molecular structure from the document to be analyzed (600).

[0314] In one embodiment, as illustrated in FIG. 12, when a document to be analyzed (600) includes a first molecular structure (611) and a second molecular structure (621), the document understanding model (300) can understand the first molecular structure (611) and the second molecular structure (621), and extract the first molecular structure (611) and the second molecular structure (621) included in the document to be analyzed (600) based on the understanding result. In this case, the answer generation system (100) can specify the first molecular structure (611) and the second molecular structure (621) extracted using the document understanding model (300) as the predicted target molecular structure.

[0315] In another embodiment, when the document to be analyzed (600) includes a first molecular graph for a first molecular structure (611) and a second molecular graph for a second molecular structure (621), the document understanding model (300) can understand the first molecular graph and the second molecular graph, and extract the first molecular graph and the second molecular graph included in the document to be analyzed (600) based on the understanding result. In this case, the answer generation system (100) can specify the first molecular structure (611) for the first molecular graph and the second molecular structure (621) for the second molecular graph as the target molecular structures to be predicted.

[0316] In addition, the answer generation system (100) can extract at least one of text, formula, chart, table, and image as multiple contents from the analysis target document (600) using the document understanding model (300). For example, the document understanding model (300) can understand the texts (612, 613, 614, 615, 622, 623, 624, 625, 632, 633, 634, 635) included in the analysis target document (600), and extract the texts (612, 613, 614, 615, 622, 623, 624, 625, 632, 633, 634, 635) included in the analysis target document (600) based on the results of the understanding.

[0317] Furthermore, in the present invention, a process of storing information on a plurality of molecular structures in memory can be performed (S1020, see FIG. 10).

[0318] A plurality of contents extracted from the document understanding model (300) can be stored in a storage unit (or memory, 140).

[0319] More specifically, the answer generation system (100) can store various information about the extracted plurality of molecular structures (611, 621, 631) in the storage (or memory, 140). For example, the answer generation system (100) can store the extracted plurality of molecular structures (611, 621, 631) and the extracted texts (612, 613, 614, 615, 622, 623, 624, 625, 632, 633, 634, 635) together with the plurality of molecular structures (611, 621, 631) in the storage (140).

[0320] Meanwhile, the answer generation system (100) can analyze the relationships between multiple pieces of content stored in the storage unit (140) based on the meaning of each piece of content. That is, the answer generation system (100) or the processor of the system (100) can perform a series of processes described in the present invention using the content stored in the storage unit (140, or memory).

[0321] Here, relationship analysis can be understood as the process of identifying and understanding the interrelationships (or relevance) between multiple pieces of content based on the meaning each piece of content carries. This may involve grouping or deriving specific patterns by identifying the similarity, interdependence, or related meaning of the information expressed by each piece of content.

[0322] The answer generation system (100) can analyze the meaning of each of the multiple contents and identify (or extract) the meaning of each of the multiple contents. The analysis of the meaning of each of the multiple contents can be performed using at least one of the ultra-large foundation model (200), the chemical reaction prediction model (400), and the molecular property prediction model (500). For example, the answer generation system (100) analyzes the meaning of each of the texts (612, 613, 614, 615, 622, 623, 624, 625, 632, 633, 634, 635) extracted from the document understanding model (300), and based on the analysis results, the meaning of each of the texts (612, 613, 614, 615, 622, 623, 624, 625, 632, 633, 634, 635) is determined by the name of the first molecular structure (612), the description of the first molecular structure (613), the SMILES notation of the first molecular structure (614), the physical properties of the first molecular structure (615), the name of the second molecular structure (622), the description of the second molecular structure (623), and the SMILES notation of the second molecular structure. It can be specified as having a notation (624), a physical property characteristic of the second molecular structure (625), a name of the third molecular structure (632), a description of the third molecular structure (633), a SMILES notation of the third molecular structure (634), and a physical property characteristic of the third molecular structure (635).

[0323] Furthermore, the answer generation system (100) can specify contents that have a relationship with each other based on the meaning of each of the specified plurality of contents. More specifically, the answer generation system (100) can specify a relationship between a plurality of molecular structures (611, 621, 631) extracted from the document understanding model (300), a name (612) of a first molecular structure having different meanings, a description (613) of the first molecular structure, a SMILES notation (614) of the first molecular structure, a physical property characteristic (615) of the first molecular structure, a name (622) of the second molecular structure, a description (623) of the second molecular structure, a SMILES notation (624) of the second molecular structure, a physical property characteristic (625) of the second molecular structure, a name (632) of the third molecular structure, a description (633) of the third molecular structure, a SMILES notation (634) of the third molecular structure, and a physical property characteristic (635) of the third molecular structure. For example, the answer generation system (100) can specify that a first molecular structure (611), a name of the first molecular structure (612), a description of the first molecular structure (613), a SMILES notation (614) of the first molecular structure, and a physical property characteristic (615) of the first molecular structure have a relationship with each other, and a second molecular structure (621), a name of the second molecular structure (622), a description of the second molecular structure (623), a SMILES notation (624) of the second molecular structure, and a physical property characteristic (625) of the second molecular structure have a relationship with each other. In addition, the answer generation system (100) can specify that a third molecular structure (631), a name of the third molecular structure (632), a description of the third molecular structure (633), a SMILES notation (634) of the third molecular structure, and a physical property characteristic (635) of the third molecular structure have a relationship with each other.

[0324] Meanwhile, the answer generation system (100) can group contents regarding the same molecular structure among the plurality of contents into related contents based on the relationships between the plurality of contents. More specifically, the answer generation system (100) can group, based on a specific relationship, the name (612) of the first molecular structure, a description (613) of the first molecular structure, a SMILES notation (614) of the first molecular structure, and physical properties (615) of the first molecular structure, which contain contents related to the first molecular structure (611). In addition, the answer generation system (100) can group, based on a specific relationship, the name (622) of the second molecular structure, the description (623) of the second molecular structure, the SMILES notation (624) of the second molecular structure, and the physical properties (625) of the second molecular structure, which contain content related to the second molecular structure (621), as content related to the second molecular structure (621). Furthermore, the answer generation system (100) can group, based on a specific relationship, the name (632) of the third molecular structure, the description (633) of the third molecular structure, the SMILES notation (634) of the third molecular structure, and the physical properties (635) of the third molecular structure, which contain content related to the third molecular structure (631).

[0325] Through the grouping process described above, at least one grouped content can be generated (or extracted). Referring to FIG. 12, a first molecular structure (611), a name of the first molecular structure (612), a description of the first molecular structure (613), a SMILES notation (614) of the first molecular structure, and a physical property characteristic (615) of the first molecular structure can be grouped to generate a grouped first content (610). In addition, a second molecular structure (621), a name of the second molecular structure (622), a description of the second molecular structure (623), a SMILES notation (624) of the second molecular structure, and a physical property characteristic (625) of the second molecular structure can be grouped to generate a grouped second content (620). Furthermore, the third molecular structure (631), the name of the third molecular structure (632), the description of the third molecular structure (633), the SMILES notation of the third molecular structure (634), and the physical properties of the third molecular structure (635) may be grouped to generate grouped third content (630).

[0326] The contents (610, 620, 630) grouped based on each of a plurality of molecular structures (611, 621, 631) may include at least one of a molecular structure image of a specific molecular structure corresponding to each of the grouped contents (610, 620, 630), a name of the molecular structure (612, 622, 632), a description of the molecular structure (613, 623, 633), a string according to the SMILES notation of the molecular structure (614, 624, 634), and a physical property characteristic (615, 625, 635) of the molecular structure.

[0327] Furthermore, grouped contents (610, 620, 630) can be stored in the storage unit (140) in connection with a user account.

[0328] Meanwhile, in the present invention, a process of obtaining a molecular graph for a plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on a plurality of molecular structures stored in memory can be performed (S1030, see FIG. 10).

[0329] Here, the molecular graph may include nodes and edges corresponding to each atom and bond between atoms included in the molecular structure. That is, in the molecular graph, atoms may be represented as nodes and bonds may be represented as edges.

[0330] The answer generation system (100) can obtain a molecular graph for a molecular structure by transforming the molecular structure. More specifically, the answer generation system (100) can obtain a molecular graph for each of the multiple molecular structures by transforming the atoms contained in each of the multiple molecular structures into nodes and the bonds between atoms into edges based on the multiple molecular structures stored in the storage unit (140).

[0331] In one embodiment, as illustrated in FIG. 12, the answer generation system (100) can convert the atoms constituting the first molecular structure (611) stored in the storage unit (140) into nodes (611a, 611b, 611c) and the bonding relationships between the atoms constituting the first molecular structure (611) into edges (611d, 611e, 611f), thereby obtaining a first molecular graph including nodes (611a, 611b, 611c) and edges (611d, 611e, 611f) corresponding to the first molecular structure (611).

[0332] In another embodiment, the answer generation system (100) can convert the atoms constituting the second molecular structure (621) stored in the storage unit (140) into nodes (621a, 621b, 621c) and the bonding relationships between the atoms constituting the second molecular structure (621) into edges (621d, 621e, 621f), thereby obtaining a second molecular graph including nodes (621a, 621b, 621c) and edges (621d, 621e, 621f) corresponding to the second molecular structure (621).

[0333] However, in the present invention, the molecular graph may be obtained using a document understanding model (300) or may be obtained (or extracted) from various information related to the molecular structure stored in the storage unit (140).

[0334] In one embodiment, when a document to be analyzed (600) includes a first molecular structure (611) and a second molecular structure (621), the document understanding model (300) understands the first molecular structure (611) and the second molecular structure (621), and based on the result of the understanding, converts the atoms included in each of the first molecular structure (611) and the second molecular structure (621) into nodes (611a, 611b, 611c, 621a, 621b, 621c) and the bonds between atoms into edges (611d, 611e, 611f, 621d, 621e, 621f), thereby obtaining a molecular graph for each of the first molecular structure (611) and the second molecular structure (621).

[0335] In another embodiment, when the text extracted from the document understanding model (300) includes descriptions of the first molecular structure (611) and the second molecular structure (621), the answer generation system (100) can extract molecular graphs for each of the first molecular structure (611) and the second molecular structure (621) from the storage (140) based on the descriptions for each of the first molecular structure (611) and the second molecular structure (621).

[0336] That is, as examined above, there may be various methods (or methods) for obtaining a molecular graph in the present invention, and for convenience of explanation, the following description will be given on the assumption that the molecular graph is obtained through a conversion process by the answer generation system (100).

[0337] Meanwhile, the answer generation system (100) can perform labeling for each of the grouped contents (610, 620, 630) stored in the storage unit (140) so that a label is assigned to each of the grouped contents (610, 620, 630).

[0338] The answer generation system (100) can perform labeling for a plurality of molecular structures (611, 621, 631) so that a different label is assigned to each of the plurality of molecular structures (611, 621, 631).

[0339] In one embodiment, the answer generation system (100) can assign a first label (M1) to the first molecular structure (611) and a second label (M2) to the second molecular structure (621) through labeling the first molecular structure (611) and the second molecular structure (621).

[0340] In another embodiment, the answer generation system (100) may assign a first label (M1) to first content (610) grouped based on a first molecular structure (611) and a second label (M2) different from the first label (M1) to second content (620) grouped based on a second molecular structure (621) through labeling of a plurality of molecular structures (611, 621, 631). In addition, the answer generation system (100) may assign a third label (M3) different from the first label (M1) and the second label (M2) to third content (630) grouped based on a third molecular structure (631). In this case, the labeling of molecular structures may be understood as assigning the same label to content grouped based on a specific molecular structure (i.e., all content included in the grouped content has the same label).

[0341] The grouped contents (610, 620, 630) assigned different labels as discussed above can be stored in the storage (140) in connection with a user account. More specifically, the answer generation system (100) can store information (M1) about a first molecular graph acquired through a process of acquiring a molecular graph and a label assigned to the first molecular graph, and information (M2) about a second molecular graph acquired through a process of acquiring a molecular graph and a label assigned to the second molecular graph, in the storage (or memory, 140).

[0342] Meanwhile, the answer generation system (100) can provide grouped contents (610, 620, 630) stored in the storage unit (140) to the user terminal (10) where the service page (1000) is output. The answer generation system (100) can provide a graphic object corresponding to each labeled content to an area of ​​the service page (1000) where the user query is received.

[0343] Here, the labeled content may correspond to content (e.g., an image of a molecular structure, a name of a molecular structure, a description of a molecular structure, a SMILES notation for a molecular structure, physical properties of a molecular structure, etc.) related to a molecular structure (e.g., a first molecular structure (611), a second molecular structure (621), a third molecular structure (631)) extracted from the analysis target document (600) discussed above.

[0344] In this regard, the answer generation system (100) can provide a plurality of graphic objects corresponding to each of a plurality of molecular structures (611, 621, 631) assigned different labels to an area of ​​a service page (1000) where a user query is received.

[0345] As illustrated in FIG. 13, the service page (1000) may include at least one of a first area (710) in which information related to a plurality of molecular structures specified from a document to be analyzed (600) is provided, a second area (720) in which at least a portion of the document is provided or an answer to a user query is provided, and a third area (730) in which a user query is received.

[0346] First, the first area (710) of the service page (1000) may include at least one grouped content with a label. The first area (710) may display information about a specific (or extracted) molecular structure, and such information may be provided in various forms, such as a molecular graph, text, or image.

[0347] Specifically, the first region (710) may include at least one graphic object (711, 712, 713) corresponding to an extracted molecular structure (e.g., a first molecular structure (611) assigned with a first label (M1), a second molecular structure (621) assigned with a second label (M2), a third molecular structure (631) assigned with a third label (M3)) each assigned with a different label through labeling, and at least one of detailed information about the extracted molecular structure (611, 621, 631).

[0348] The first area (710) of the service page (1000) may include a first sub-area (710a) containing graphic objects (711, 712, 713) and a second sub-area (710b) containing detailed information.

[0349] In this regard, when a plurality of molecular structures (611, 621, 631) are specified from a document (600), a first sub-area (710a) may include a plurality of graphic objects (711, 712, 713) corresponding to each of the plurality of molecular structures. More specifically, a first graphic object (711) among the plurality of graphic objects may include a graph of a first molecular structure (611) corresponding to the first graphic object (711) among the plurality of molecular structures (611, 621, 631), and a second graphic object (712) may include a graph of a second molecular structure (621) corresponding to the second graphic object (712). In this case, at least some of the graphic objects may include a molecular structure image of the molecular structure.

[0350] Additionally, the second sub-area (710b) may provide detailed information on the molecular structure corresponding to one of the plurality of graphic objects (711, 712, 713) selected by user input. The answer generation system (100) may provide detailed information on the selected graphic object based on the user input received for selecting one of the plurality of graphic objects (711, 712, 713).

[0351] In this regard, detailed information of a plurality of molecular structures (611, 621, 631) corresponding to each of the plurality of graphic objects (711, 712, 713) included in the first sub-area (710a) may be linked (or associated) to exist. For example, assume that a first graphic object (711) corresponding to a first molecular structure (611) is selected from a user terminal (10). The answer generation system (100) can provide detailed information (711a, 711b, 711c, 711d, 711e) about the first molecular structure (611) linked to the first graphic object (711) and information about the first label (M1) assigned to the first molecular structure to the second sub-area (710b) based on the selection of the first graphic object (711) included in the first sub-area (710a) from the user terminal (10).

[0352] Here, detailed information about the molecular structure may include at least one of a molecular structure image (711a) of the molecular structure, a name of the molecular structure (711b), a description of the molecular structure (711c), a string according to the SMILES notation (711d), and a physical property characteristic (711e) of the molecular structure. In this case, the molecular structure image of the molecular structure may also be provided (or displayed) in the form of a molecular graph obtained through a process of obtaining the molecular graph.

[0353] Meanwhile, detailed information (or at least one content included in the grouped content) may be extracted from a document or acquired from at least one pre-trained predictive model. As discussed above, the pre-trained predictive model may include at least one of a chemical reaction prediction model (400) that predicts chemical reactions between molecular structures and a molecular property prediction model (500) that predicts physical properties of molecular structures.

[0354] As an example, if a plurality of contents extracted from a document to be analyzed (600) include a molecular structure image, a name, a description, and a string according to the SMILES notation of the first molecular structure (611), and no physical properties for the first molecular structure (611) exist, the answer generation system (100) can predict the physical properties for the first molecular structure (611) using a pre-learned molecular property prediction model (500). The molecular property prediction model (500) outputs the physical property prediction result for the first molecular structure (611) as output data, and the answer generation system (100) can obtain the physical property prediction result for the first molecular structure (611) and generate detailed information for the first molecular structure (611). In this case, the molecular structure image (711a), name (711b), description (711c) and smile notation (711d) of the first molecular structure (611) included in the first area (710) of the service page (1000) are extracted from the document (600), and the physical properties (711e) for the first molecular structure (611) can be understood to have been generated by the molecular physical properties prediction model (500).

[0355] Next, a document (600) received from a user terminal (10) may be provided in a second area (720) that is distinct from the first area (710) of the service page (1000).

[0356] In the second region (720), highlighting objects corresponding to each of the plurality of molecular structures (611, 621, 631) may be overlapped in one region of the document provided on the service page (1000) so that it is identifiable that a plurality of molecular structures (611, 621, 631) have been extracted from the document (600). More specifically, in the second region (720), highlighting objects (721, 722) may be overlapped and displayed in the first region (or first sub-region, 720a) including the first molecular structure (611) of the document provided on the service page (1000) and the second region (or second sub-region, 720b) ​​including the second molecular structure (621), so that it is identifiable that the first molecular structure (611) and the second molecular structure (621) have been extracted from the document (600).

[0357] Here, in the first region (720a) including the first molecular structure (611), a first label (M1) assigned to correspond to the first molecular structure (611) may be provided around the first highlighting object (721) overlapped with the first region (720a). In addition, in the second region (720b) ​​including the second molecular structure (621), a second label (M2) assigned to correspond to the second molecular structure (621) may be provided around the second highlighting object (722) overlapped with the second region (720b).

[0358] Furthermore, the highlighting objects (721, 722) may be visually highlighted and displayed to be distinguished from other objects in a user interface (e.g., a service page (1000)). For example, the answer generation system (100) may display the highlighting objects (721, 722) to be distinguished from other objects in the service page (1000) by at least one of changing the color of the highlighting objects (721, 722), adding a border, or changing the background color of the highlighting objects (721, 722).

[0359] Meanwhile, when a highlighting object is selected by user input, information about a specific molecular structure corresponding to the selected highlighting object may be provided in another area distinct from an area of ​​the service page (1000) where the highlighting object is displayed.

[0360] Specifically, the answer generation system (100) may provide detailed information about a specific molecular structure linked to the highlighted object selected according to the user input in a first area (710) of a service page (1000) based on receiving a user input for selecting one of a plurality of highlighting objects (721, 722). For example, assume that a user input for a first highlighting object (721) in a first area (720a) including a first molecular structure (611) is received. The answer generation system (100) may provide detailed information (711a, 711b, 711c, 711d, 711e) about the first molecular structure (611) linked to the first highlighting object (721) in the first area (710) based on receiving a user input for selecting the first highlighting object (721).

[0361] At this time, depending on the selected highlighting object, the display of multiple graphic objects (711, 712, 713) included in the first area (710) of the service page (1000) may also be changed. More specifically, when the first highlighting object (721) is selected, the first graphic object (711) corresponding to the first highlighting object (721) may be highlighted and displayed in the first sub-area (710a) of the first area (710) so that it can be identified that the first highlighting object (721) has been selected by the user.

[0362] That is, when a highlighting object is selected, a graphic object corresponding to the selected highlighting object may be displayed in the first area (710) with a first visual appearance so that the user can intuitively recognize it. Alternatively, a graphic object corresponding to an unselected highlighting object may be displayed with a second visual appearance.

[0363] In this way, as a highlighting object is selected by the user, information about a specific molecular structure linked to the highlighting object can be provided (or displayed) in another area (first area) that is distinct from the area (second area) in which the highlighting object is displayed.

[0364] In other words, the present invention visually presents information through graphic objects and labels, enabling users to easily understand and utilize data related to complex molecular structures. This enhances user convenience and comprehension, and can increase research efficiency.

[0365] Meanwhile, the answer generation system (100) can receive an editing request for a plurality of molecular structures through a service page (1000) that provides answers to user queries.

[0366] The answer generation system (100) may provide a graphic object linked to an editing request receiving function for a plurality of molecular structures in one area of ​​the service page (1000). For example, as illustrated in FIG. 13, the answer generation system (100) may provide a graphic object (714) linked to an editing request receiving function for a molecular structure (first molecular structure (611)) corresponding to the molecular structure image (or graph) in a first area (710) of the service page (1000) where a molecular structure image corresponding to the selected graphic object (first graphic object (711)) is displayed.

[0367] For another example, although not shown, the answer generation system (100) may provide a graphic object linked to an editing request receiving function for a plurality of molecular structures to each of a plurality of graphic objects (711, 712, 713) provided in the first region (710). The answer generation system (100) may receive an editing request for a specific molecular structure corresponding to the selected graphic object based on receiving a user input selecting one of the graphic objects provided in each of the plurality of graphic objects (711, 712, 713).

[0368] As another example, although not shown, the answer generation system (100) may provide a graphic object linked to an editing request receiving function for a plurality of molecular structures to each of a plurality of highlighting objects (721, 722) provided in the second area (720). The answer generation system (100) may receive an editing request for a specific molecular structure corresponding to the selected highlighting object based on receiving a user input selecting any one of the graphic objects provided to each of the plurality of highlighting objects (721, 722).

[0369] That is, the present invention does not limit the method of receiving an editing request for a molecular structure to any one method. For convenience of explanation, the following description assumes that the graphic object (714) has been selected.

[0370] The answer generation system (100) can receive an editing request for a plurality of molecular structures (e.g., the first molecular structure) based on the selection of a graphic object (714) included in the first region (710) from the user terminal (10). And, as illustrated in FIG. 14, the answer generation system (100) can provide an editing interface (800) that provides an editing function for the first molecular structure (611) based on receiving an editing request for the first molecular structure (611) from the user terminal (10).

[0371] The editing interface (800) may provide a molecular graph corresponding to either a first molecular structure or a second molecular structure obtained through molecular graph transformation among multiple molecular structures in an editable state. For example, the answer generation system (100) may provide a molecular structure graph (810) of the first molecular structure (611) on the editing interface (800) activated on the user terminal (10) based on receiving an editing request for the first molecular structure (611) from the user terminal (10).

[0372] Here, the molecular structure image (810) can also be understood as a molecular graph including nodes (nodes, 811a, 811b, 811c, 811d, 811e, 811f, 811g) corresponding to each of the atoms constituting the first molecular structure (611) and edges (edges, 812a, 812b, 812c, 812d, 812e) indicating the bonding relationship of the atoms. The first molecular structure (611) can be edited based on a user input for at least one of the nodes (811a, 811b, 811c, 811d, 811e, 811f, 811g) and the edges (812a, 812b, 812c, 812d, 812e).

[0373] Specifically, editing of the first molecular structure (611) may be a deletion or position change of at least one of the nodes (811a, 811b, 811c, 811d, 811e, 811f, 811g) corresponding to each of the atoms constituting the first molecular structure (611) and the edges (812a, 812b, 812c, 812d, 812e) indicating the bonding relationship between the atoms, or an addition of a new node corresponding to a new atom or an addition of a new edge creating a new bonding relationship between the atoms.

[0374] For example, the answer generation system (100) may activate a node and edge deletion mode based on the selection of a graphic object (801) associated with a node and edge deletion function included in an area of ​​the editing interface (800). In addition, the answer generation system (100) may edit the first molecular structure (611) such that a specific node (811b, 811e) at a position corresponding to the user input is deleted based on the reception of a user input for selecting a specific node (811b, 811e) among a plurality of nodes (811a, 811b, 811c, 811d, 811e, 811f, 811g) included in an image (810) of the first molecular structure.

[0375] In this case, when editing is performed on the first molecular structure (611) to be edited based on the user's input, a molecular structure image (820) corresponding to the edited molecular structure, which is different from the molecular structure image (810) of the first molecular structure (611) before editing, may be continuously displayed on the editing interface (800). For example, based on the deletion of a specific node (811b, 811e) corresponding to the user's selection, a molecular structure image (820) in which the specific node (811b, 811e) is deleted may be displayed on the editing interface (800).

[0376] Furthermore, the answer generation system (100) can store the molecular structure (or molecular structure image or molecular graph, 820) that has been edited (or has been edited) in a predetermined storage (e.g., storage (140) or memory) in association with a user account. For example, the answer generation system (100) can generate a molecular structure (820) in which the first molecular structure (611) is edited based on the selection of a graphic object (802) associated with an edit save (or completion) function included in the editing interface (800) from the user terminal (10), and store the edited molecular structure (820) in association with the user account in the storage (or memory, 140).

[0377] In this way, the present invention can provide a user environment in which a user can design a desired molecule through an editing interface.

[0378] Meanwhile, the edited molecular structure may be assigned a new label that specifies the edited molecular structure. The answer generation system (100) may perform labeling on the edited molecular structure stored in the storage unit (140) so that a new label for specifying the edited molecular structure is assigned. For example, as illustrated in FIG. 15, the answer generation system (100) may perform labeling on the edited molecular structure (941) so that the edited molecular structure (941) is assigned a fourth label (M4) that is different from the first label (M1) assigned to the molecular structure before editing (e.g., the first molecular structure (911)).

[0379] The answer generation system (100) can generate a fourth graphic object corresponding to the edited molecular structure (941) and provide the generated fourth graphic object in one area of ​​the service page (1000).

[0380] Specifically, as illustrated in FIG. 16, a graphic object (or fourth graphic object, 1014) corresponding to an edited molecular structure (941) may be provided in a first area (or first sub-area, 1010) of a service page (1000). Here, the graphic object (1014) corresponding to the edited molecular structure (941) may include a molecular structure image of the edited molecular structure (941).

[0381] Additionally, the first area (or second sub-area, 1010) of the service page (1000) may include a molecular structure image (1014a) of the edited molecular structure (941). In addition, a fourth label (M4) assigned to the edited molecular structure (941) may be provided in a surrounding area of ​​the molecular structure image (1014a).

[0382] Furthermore, the graphic object (1014) corresponding to the edited molecular structure (941) may further include detailed information about the edited molecular structure (941). For example, it may include at least one of a name (1014b) of the edited molecular structure (941), a description (1014c) of the edited molecular structure (941), a SMILES notation (1014d) of the edited molecular structure (941), and a physical property characteristic (1014e) of the edited molecular structure (941).

[0383] In this case, at least one of the molecular structure image (1014a) of the edited molecular structure (941), the name (1014b) of the edited molecular structure (941), the description (1014c) of the edited molecular structure (941), the SMILES notation (1014d) of the edited molecular structure (941), and the physical properties (1014e) of the edited molecular structure (941) may be generated by at least one of the pre-learned chemical reaction prediction model (400) and the molecular physical properties prediction model (500). Hereinafter, for convenience of explanation, the edited molecular structure (941) is named “the fourth molecular structure (941)”, and the detailed information of the edited molecular structure (941) is named “the grouped fourth content (940)” (see FIG. 15).

[0384] Meanwhile, in the present invention, a process of receiving a user query for chemical reaction prediction related to multiple molecular structures can be performed through a service page (S1040, see FIG. 10).

[0385] The answer generation system (100) can receive a user query containing at least one label assigned through labeling via the service page (1000). More specifically, the answer generation system (100) can receive a user query regarding the prediction of a chemical reaction related to a plurality of labeled molecular structures.

[0386] As illustrated in FIG. 16, the third area (1030) of the service page (1000) may be configured to receive user queries. The third area (1030) may include a graphic object (1031) associated with a user query receiving function.

[0387] The answer generation system (100) can receive a user query including a label assigned to a specific molecular structure through the third area (1030) of the service page (1000).

[0388] Specifically, the answer generation system (100) can receive a user query including the plurality of molecular structures to which different labels are assigned. For example, the answer generation system (100) can receive a user query (e.g., “Can you predict the reaction between m2 and m4?”, 1032) including a second label (M2) assigned to a second molecular structure (621) and a fourth label (M4) assigned to an edited molecular structure (941) based on the selection of a graphic object (1031) included in a third region (1030) from a user terminal (10). In this case, the user query (1032) may be a query related to predicting a chemical reaction between a second molecular structure (621) to which a second label (M2) is assigned and a fourth molecular structure (941) to which a fourth label (M4) is assigned.

[0389] In this way, users can input queries more intuitively and simply by utilizing labels assigned to specific molecular structures, without having to input complex information about specific molecular structures.

[0390] When a user query is received, the answer generation system (100) can identify a specific content (or molecular structure) related to the user query among a plurality of contents.

[0391] The answer generation system (100) can input the user query (1032) into the super-large foundation model (200) based on the user query (1032) received.

[0392] The super-large foundation model (200) receives a user query (1032) as input, understands the question (or content) included in the user query (1032), and can identify specific content (or molecular structure) related to the user query (1032).

[0393] The super-large foundation model (200) can analyze a user query (1032) and extract labels representing grouped content from the user query (1032). For example, the super-large foundation model (200) can extract the second label (M2) and the fourth label (M4) based on the analysis result of the user query (1032) that the user query (1032) includes text corresponding to the second label (M2) representing the grouped second content (620) and the fourth label (M4) representing the grouped fourth content (940).

[0394] Furthermore, the super-large foundation model (200) can specify specific grouped content corresponding to the extracted labels. For example, the super-large foundation model (200) can specify the grouped second content (620) corresponding to the extracted second label (M2) and the grouped fourth content (940) corresponding to the extracted fourth label (M4) as specific grouped content.

[0395] Here, the fact that the grouped second content (620) and the grouped fourth content (940) are specified can also be understood to mean that the second molecular structure (621) corresponding to the grouped second content (620) and the fourth molecular structure (941) corresponding to the grouped fourth content (940) are specified.

[0396] In this way, the present invention can shorten the time required to identify specific content corresponding to a user query through a label assigned to an extracted molecular structure.

[0397] When content related to a user query is specified, the present invention may proceed with a process of processing molecular graphs for multiple molecular structures as input to a chemical reaction prediction model so as to predict a chemical reaction corresponding to the user query (S1050, see FIG. 10).

[0398] The super-large foundation model (200) can process the molecular structure of a specific grouped content as input to a pre-trained prediction model. More specifically, the super-large foundation model (200) can process a second grouped content (620) assigned a second label (M2) and a fourth grouped content (940) assigned a fourth label (M4) as input to the pre-trained prediction model.

[0399] Here, processing a specific grouped content as input to a pre-trained prediction model can also be understood to mean processing a molecular structure corresponding to the grouped content as input.

[0400] In this regard, the answer generation system (100) can process a second molecular structure (621) corresponding to a specific grouped second content (620) and a fourth molecular structure (941) corresponding to a specific grouped fourth content (940) as inputs of a pre-learned prediction model.

[0401] More specifically, the answer generation system (100) can process the second molecular graph for the second molecular structure (621) and the fourth molecular graph for the fourth molecular structure (941) as inputs of the pre-learned prediction model. The second molecular graph for the second molecular structure (621) can be confirmed with reference to FIG. 12, and the fourth molecular graph for the fourth molecular structure (941) can be confirmed with reference to FIG. 15. In this case, which of the multiple pre-learned prediction models will process the molecular graph (or specific content) as input can be determined based on the user query (1032). For example, the super-large foundation model (200) can understand the content included in the user query (1032) and determine that the user query (1032) is related to predicting a chemical reaction for multiple molecular structures based on the fact that the user query (1032) includes the content “Can you predict a chemical reaction between m2 and m4?” Based on the judgment result, the super-large foundation model (200) can determine a prediction model to which multiple molecular graphs (e.g., the second molecular graph and the fourth molecular graph) are input, as a chemical reaction prediction model (400).

[0402] Furthermore, the super-large foundation model (200) can process the second molecular graph and the fourth molecular graph as inputs to a chemical reaction prediction model (400) that understands the chemical reaction mechanism.

[0403] When a molecular graph for a molecular structure is input into a chemical reaction prediction model, a process of performing chemical reaction prediction for multiple molecular structures can be performed in the chemical reaction prediction model (S1060, see FIG. 10).

[0404] Furthermore, in the present invention, a process of obtaining a result of chemical reaction prediction for a plurality of molecular structures from a chemical reaction prediction model can be performed (S1070, see FIG. 10).

[0405] The chemical reaction prediction process of the chemical reaction prediction model (400) has been described in more detail above, so it will be briefly described below to avoid duplication of content.

[0406] When a second molecular graph and a fourth molecular graph are input, the chemical reaction prediction model (400) can obtain an embedding vector corresponding to the second molecular graph and the fourth molecular graph from the embedding layer (411) of the encoder (410) using information related to a plurality of molecular structures (the second molecular structure (621) and the fourth molecular structure (941)). In addition, the chemical reaction prediction model (400) can perform an attention operation related to the interaction between atoms of the plurality of molecular structures (621, 941) using a multi-head self-attention layer (413) and update the embedding vector based on the operation. Next, the chemical reaction prediction model (400) performs bond and atom prediction predicted as a chemical reaction of a plurality of molecular structures (621, 641) using the embedding vector updated through the update process, and can output the final chemical reaction result predicted from the chemical reaction of the plurality of molecular structures (621, 641) using the result of bond prediction and the result of atom prediction.

[0407] As discussed above, the updated embedding vector can be understood as being updated by adding different biases according to the coupling types between the nodes constituting the second molecular graph and the fourth molecular graph to the attention scores calculated in each of the multiple heads of the multi-head self-attention layer (413). Furthermore, coupling prediction is performed by performing a dot-product operation using the updated embedding vector, and atomic prediction predicts the atomic properties of the atoms corresponding to the updated embedding vector using the atomic property probability distribution of each of the atoms corresponding to the updated embedding vector. The atomic properties may include charge states of the atoms that may change during a chemical reaction process of the multiple molecular structures.

[0408] That is, the chemical reaction prediction model (400) can output the final chemical reaction result predicted from the chemical reaction between the second molecular structure (621) and the fourth molecular structure (941) through the process mentioned above. Here, the result of the chemical reaction prediction obtained from the chemical reaction prediction model (400) can correspond to the final result stabilized through the diffusion feedback process.

[0409] In one embodiment, the output data (or result of chemical reaction prediction) of the chemical reaction prediction model (400) may include a specific molecular structure generated as a result of chemical reaction prediction between a plurality of molecular structures and at least one piece of information (or content) related to the specific molecular structure. For example, the output data may include at least one of a fifth molecular structure generated as a result of chemical reaction prediction between the second molecular structure (621) and the fourth molecular structure (941), a molecular structure image of the fifth molecular structure, a name of the fifth molecular structure, a description of the fifth molecular structure, and a SMILES notation of the fifth molecular structure.

[0410] Meanwhile, in the present invention, a process of generating an answer to a user's query using the result of chemical reaction prediction can be performed (S1080, see FIG. 10).

[0411] The super-large foundation model (200) can determine the answer generation procedure performed for prediction corresponding to a user query (1032) and the tool used in the answer generation procedure. For example, as illustrated in FIG. 17, the super-large foundation model (200) can determine which procedure and tool to use to generate an answer to a user query (1032).

[0412] In addition, the answer generation system (100) can provide information on the answer generation procedure and tool determined from the super-large foundation model (200) to the service page (1000). For example, the answer generation system (100) can provide information (1101) on the answer generation procedure and tool determined from the super-large foundation model (200) through the service page (1000) to perform prediction corresponding to a user query (ex: “Can you predict the reaction between m2 and m4?”, 1100).

[0413] Furthermore, the super-large foundation model (200) can generate an answer to the user query (1032) by using a plurality of molecular structures (e.g., the second molecular structure (621) and the fourth molecular structure (941)) corresponding to specific labels (the second label (M2) and the fourth label (M4)) included in the user query (1032). More specifically, the super-large foundation model (200) can generate an answer (1110) to the user query (1100) by using the chemical reaction prediction result of the chemical reaction prediction model (400) and the content constituting the grouped content (e.g., content related to the second molecular structure (621) and the fourth molecular structure (941)) and information (1101) about the determined answer generation procedure and tool.

[0414] Meanwhile, the answer generation system (100) can use a pre-learned molecular property prediction model (500) to predict the properties of a specific molecular structure, and provide information on the properties of a specific molecular structure predicted from the molecular property prediction model (500) as an answer (1110) to a user query (1100).

[0415] As discussed above, the molecular property prediction model (500) may be a model constructed for material structure design. The molecular property prediction model (500) may be configured to predict properties from molecular structures or to design molecules with desired properties (or new properties) by a user.

[0416] Specifically, the answer generation system (100) can process the fifth molecular structure as input to the molecular property prediction model (500). The molecular property prediction model (500) can receive the fifth molecular structure as input and output the property prediction result for the fifth molecular structure as output data. The answer generation system (100) can obtain the property prediction result for the fifth molecular structure output from the molecular property prediction model (500).

[0417] Furthermore, the answer generation system (100) can input the property prediction results for the fifth molecular structure into the super-large foundation model (200). The super-large foundation model (200) can generate an answer (1110) to a user query (1100) using the property prediction results for the fifth molecular structure. In this case, the answer generation system (100) can provide information on the properties of the fifth molecular structure predicted from the molecular property prediction model (500) as an answer (1110) to the user query (1100).

[0418] Meanwhile, if the answer (1110) to the user query (1100) includes a specific molecular structure (or a new molecular structure) generated through a previously learned prediction model, a label may be assigned to the specific molecular structure.

[0419] The answer generation system (100) can perform labeling on a specific molecular structure so that a new label is assigned to identify the specific molecular structure. For example, the answer generation system (100) can perform labeling on a fifth molecular structure so that a fifth label (M5) is assigned to the fifth molecular structure generated through the chemical reaction prediction model (400).

[0420] Here, a specific molecular structure (the fifth molecular structure) and a label assigned to the specific molecular structure (the fifth label (M5)) can be stored in the storage unit (140) together with the extracted molecular structure and the label assigned to the extracted molecular structure, linked to the user account.

[0421] In addition, the answer generation system (100) can generate a specific graphic object corresponding to a specific molecular structure based on the generation of a specific molecular structure from a previously learned prediction model. For example, the answer generation system (100) can generate a fifth graphic object corresponding to a fifth molecular structure using a fifth molecular structure stored in the storage unit (140).

[0422] Furthermore, the answer generation system (100) may perform an update on the service page (1000) so that a specific graphic object corresponding to a specific molecular structure is included in one area of ​​the service page (1000). More specifically, as illustrated in FIG. 18, the answer generation system (100) may perform an update on the first area (1210) so that a graphic object (or the fifth graphic object, 1215) corresponding to the fifth molecular structure is included in the first area (1210) of the service page (1000).

[0423] Based on the update, the first area (1210) of the service page (1000) may provide detailed information on a molecular structure corresponding to a specific graphic object, together with a specific graphic object. For example, the first area (1210) may include a molecular structure image (1215a) of a fifth molecular structure corresponding to a fifth graphic object (1215), a name of the fifth molecular structure (1215b), a description of the fifth molecular structure (1215c), a SMILES notation (1215d) for the fifth molecular structure, and physical properties (1215e) for the fifth molecular structure.

[0424] Meanwhile, the answer generation system (100) can provide an answer (e.g., “The product generated through the chemical reaction between m2 and m4 is m5...”, 1221) to a user query (1100) generated from a super-large foundation model (200) in one area (or a second area, 1220) of a service page (1000).

[0425] The answer (1221) provided in the second area (1220) of the service page (1000) may include content indicating that a new specific molecular structure (e.g., a fifth molecular structure) has been created as a result of a chemical reaction prediction between multiple molecular structures (e.g., a second molecular structure (621) and a fourth molecular structure (941)).

[0426] Specifically, the answer (1221) may include a molecular structure image (1212a) of the second molecular structure (621) and a molecular structure image (1214a) of the fourth molecular structure (941), and may include a molecular structure image (1215a) of the fifth molecular structure generated as a result of a chemical reaction between the second molecular structure (621) and the fourth molecular structure (941).

[0427] Additionally, between the image of the second molecular structure (1212a), the image of the fourth molecular structure (1214a), and the image of the fifth molecular structure (1215a) included in the answer (1221), a graphic object (e.g., a plus sign, an arrow, etc.) indicating a relationship between the molecular structures corresponding to each image may be displayed together.

[0428] Furthermore, the answer (1221) may include at least one of detailed information about the fifth molecular structure generated through the chemical reaction prediction model (400) (e.g., the name of the fifth molecular structure (1215b) and the description of the fifth molecular structure (1215c)).

[0429] Furthermore, the answer (1221) may be provided with information on the properties of the fifth molecular structure predicted from the property prediction model (500), and information on the fifth label (M5) assigned to the fifth molecular structure may be displayed together in the peripheral area of ​​the molecular structure image (1215a) of the fifth molecular structure.

[0430] Meanwhile, the method of generating an answer for the super-large foundation model discussed above was explained on the assumption that a document was received. However, below, we will examine the method of generating an answer for the super-large foundation model in more detail, assuming that a text was received.

[0431] First, the answer generation system (100) can receive a user query in text form from a user terminal (10) provided with a service page (1000).

[0432] Specifically, the answer generation system (100) can receive a user query including information on at least one molecular structure through an area of ​​the service page (1000). Here, the information on the molecular structure included in the user query may exist in various ways. For example, the information on the molecular structure may include the name of a specific molecular structure, a description of the specific molecular structure, a SMILES notation for the specific molecular structure, a formula for the specific molecular structure, etc. However, the information on the molecular structure described above is merely an example, and the information on the molecular structure included in the user query in the present invention is not necessarily limited thereto, and it is obvious that various other information related to the molecular structure may be included.

[0433] For example, as illustrated in FIG. 19, the answer generation system (100) can receive a user query (ex: “Can you predict the reaction between Molecular structure A and Molecular structure B?”, 1532) including a name for a specific molecular structure (ex: “Molecular structure A”, “Molecular structure B”) based on the selection of a graphic object (1531) included in a third area (1530) from the user terminal (10).

[0434] The answer generation system (100) can input the user query (1532) into the super-large foundation model (200) based on the user query (1532) received.

[0435] The super-large foundation model (200) can specify a molecular structure to be predicted based on the name of the specific molecular structure included in the user query (1532). For example, the super-large foundation model (200) can specify a first molecular structure and a second molecular structure corresponding to each specific molecular structure based on the name of the specific molecular structure included in the user query (1532) (e.g., “Molecular structure A”, “Molecular structure B”).

[0436] Once the molecular structure is determined, the answer generation system (100) can extract information about the specified molecular structure. In this case, the extraction of information about the specified molecular structure can be accomplished by the super-large foundation model (200).

[0437] In one embodiment, as illustrated in FIG. 20, the answer generation system (100) can extract at least one of a molecular structure image of a first molecular structure (1611), a name of the first molecular structure (1612), a description of the first molecular structure (1613), a SMILES notation of the first molecular structure (1614), and a physical property characteristic (1614) of the first molecular structure related to the first molecular structure (1611) specified from a user query (1600).

[0438] In another embodiment, the answer generation system (100) can extract at least one of a molecular structure image of a second molecular structure (1621), a name of the second molecular structure (1622), a description of the second molecular structure (1623), a SMILES notation of the second molecular structure (1624), and a physical property characteristic (1625) of the first molecular structure related to the second molecular structure (1621) specified from the user query (1600).

[0439] Here, various information (e.g., detailed information, multiple contents, etc.) related to a specific molecular structure may be i) extracted from contents (or information or data) related to various molecular structures stored in the storage unit (140), or ii) generated by at least one of a chemical reaction prediction model (400) and a molecular property prediction model (500).

[0440] In one embodiment, the information related to various molecular structures stored in the storage unit (140) may include content related to molecular structures related to at least one of chemistry, biology, new materials, new substances, and new drug development extracted from each of a plurality of documents using the document understanding model (300). The ultra-large foundation model (200) may extract, as content, at least one content related to a specific molecular structure among the content related to molecular structures stored in the storage unit (140).

[0441] In another embodiment, when a specific molecular structure is specified from a user query, the answer generation system (100) can generate at least one content related to the specified molecular structure by using at least one of a pre-learned chemical reaction prediction model (400) and a molecular property prediction model (500).

[0442] Meanwhile, the answer generation system (100) can group contents regarding the same molecular structure among the plurality of contents into related contents based on the relationship between the plurality of contents. More specifically, the answer generation system (100) can group the molecular structure image of the first molecular structure (1611), the name of the first molecular structure (1612), the description of the first molecular structure (1613), the SMILES notation of the first molecular structure (1614), and the physical properties of the first molecular structure (1615) containing contents related to the first molecular structure into contents related to the first molecular structure. In addition, the answer generation system (100) can group the molecular structure image of the second molecular structure (1621), which includes content related to the second molecular structure (1621), the name of the second molecular structure (1622), the description of the second molecular structure (1623), the SMILES notation of the second molecular structure (1624), and the physical properties of the first molecular structure (1625) as content related to the second molecular structure (621).

[0443] Through the grouping process described above, content grouped based on molecular structures can be generated. For example, a first molecular structure (1611), a molecular structure image of the first molecular structure (1611), a name of the first molecular structure (1612), a description of the first molecular structure (1613), a SMILES notation of the first molecular structure (1614), and a physical property characteristic (1615) of the first molecular structure can be grouped to generate grouped first content (1610). In addition, a second molecular structure (1621), a molecular structure image of the second molecular structure (1621), a name of the second molecular structure (1622), a description of the second molecular structure (1623), a SMILES notation of the second molecular structure (1624), and a physical property characteristic (1625) of the first molecular structure can be grouped to generate grouped second content (1620).

[0444] Furthermore, grouped contents (1610, 1620) can be stored in the storage unit (140) in association with a user account.

[0445] Meanwhile, the answer generation system (100) can obtain a molecular graph for multiple molecular structures by converting atoms into nodes and bonds between atoms into edges based on multiple molecular structures stored in the storage unit (or memory, 140).

[0446] As discussed above, the answer generation system (100) can obtain a molecular graph for a molecular structure by converting atoms included in the molecular structure into nodes and bonds between atoms into edges based on the molecular structure.

[0447] In one embodiment, as illustrated in FIG. 16, the answer generation system (100) can convert the atoms constituting the first molecular structure (1611) stored in the storage unit (140) into nodes (1611a, 1611b, 1611c) and the bonding relationships between the atoms constituting the first molecular structure (1611) into edges (1611d, 1611e, 1611f), thereby obtaining a first molecular graph including nodes (1611a, 1611b, 1611c) and edges (1611d, 1611e, 1611f) corresponding to the first molecular structure (1611).

[0448] In another embodiment, the answer generation system (100) can convert the atoms constituting the second molecular structure (1621) stored in the storage unit (140) into nodes (1621a, 1621b, 1621c) and the bonding relationships between the atoms constituting the second molecular structure (1621) into edges (1621d, 1621e, 1621f), thereby obtaining a second molecular graph including nodes (1621a, 1621b, 1621c) and edges (1621d, 1621e, 1621f) corresponding to the second molecular structure (1621).

[0449] Furthermore, the answer generation system (100) can store information on a plurality of specific molecular structures in the storage unit (140). More specifically, the answer generation system (100) can store various information (e.g., the first molecular graph and the second molecular graph) on the first molecular structure (1611) and the second molecular structure (1621) in the storage unit (or memory, 140).

[0450] Meanwhile, the answer generation system (100) can perform labeling on the plurality of molecular structures (1611, 1621) so that different labels are assigned to each of the plurality of molecular structures (1611, 1621). For example, the answer generation system (100) can assign a first label (M1) to the first molecular structure (1611) and a second label (M2) to the second molecular structure (1621) through labeling the first molecular structure (or the first molecular structure graph, 1611) and the second molecular structure (or the second molecular structure graph, 1621).

[0451] And, the answer generation system (100) can store information (M1) about a first molecular graph obtained through a process of obtaining a molecular graph and a label assigned to the first molecular graph, and information (M2) about a second molecular graph obtained through a process of obtaining a molecular graph and a label assigned to the second molecular graph in a storage unit (or memory, 140).

[0452] Furthermore, the super-large foundation model (200) understands the content included in the user query (1600) and, based on the fact that the user query (1032) includes the content “Can you predict the reaction between molecular structure A and molecular structure B?”, can process the first molecular structure (1611) assigned with the first label (M1) and the second molecular structure (1621) assigned with the second label (M2) as inputs to the chemical reaction prediction model (400).

[0453] Meanwhile, since the chemical reaction prediction process of the chemical reaction prediction model (400) has been described in more detail above, it will be briefly described below to avoid duplication of content.

[0454] When a first molecular graph and a second molecular graph are input, the chemical reaction prediction model (400) can obtain an embedding vector corresponding to the first molecular graph and the second molecular graph from the embedding layer (411) of the encoder (410) using information related to a plurality of molecular structures (the first molecular structure (1611) and the second molecular structure (1621)). In addition, the chemical reaction prediction model (400) can perform an attention operation related to the interaction between atoms of the plurality of molecular structures (1611, 1621) using a multi-head self-attention layer (413) and update the embedding vector based on the operation. Next, the chemical reaction prediction model (400) performs bond and atom prediction predicted as a chemical reaction of a plurality of molecular structures (1611, 1621) using the embedding vector updated through the update process, and can output the final chemical reaction result predicted from the chemical reaction of the plurality of molecular structures (1611, 1621) using the result of bond prediction and the result of atom prediction.

[0455] Furthermore, the super-large foundation model (200) can obtain the results of chemical reaction prediction for multiple molecular structures (1611, 1621) from the chemical reaction prediction model (400), and use the obtained results to generate an answer to a user query (1600) (e.g., “The product generated through the chemical reaction between m1 and m2 is m3...”, 1721).

[0456] Meanwhile, the answer generation system (100) can generate third content (or detailed information) grouped based on a specific molecular structure based on whether the answer (1721) to the user query (1600) includes a specific molecular structure (or a new molecular structure). For convenience of explanation, the specific molecular structure will be referred to as a "third molecular structure" in the following description.

[0457] In this case, at least some of the information included in the grouped third content may be generated by at least one of a chemical reaction prediction model (400) and a molecular property prediction model (500). As another example, at least some of the information included in the grouped third content may be generated using information related to various molecular structures stored in the storage unit (140).

[0458] Additionally, the answer generation system (100) can perform labeling on third content grouped based on the third molecular structure, so that a new label is assigned to identify the third molecular structure. For example, the answer generation system (100) can perform labeling on the third molecular structure generated through the chemical reaction prediction model (400) so that a third label (M3) is assigned to the third molecular structure.

[0459] Here, the third molecular structure and the third label assigned to the third molecular structure may be stored in the storage unit (140) together with the extracted molecular structure (e.g., the first molecular structure and the second molecular structure) and the label assigned to the extracted molecular structure (e.g., the first label and the second label) linked to the user account.

[0460] Furthermore, the answer generation system (100) can generate a specific graphic object corresponding to the third molecular structure. For example, the answer generation system (100) can generate a third graphic object corresponding to the third molecular structure generated through the chemical reaction prediction model (400).

[0461] Meanwhile, as illustrated in FIG. 21, the answer generation system (100) can provide graphic objects corresponding to each labeled content, along with an answer (1721) generated from the super-large foundation model (200), to the service page (1000). Here, the labeled content can include grouped first content (1610) and second content (1620), and third content grouped based on a third molecular structure included in the answer (1721) to the user query (1600).

[0462] First, the first area (1710) of the service page (1000) may include at least one grouped content to which a label has been assigned.

[0463] Specifically, the first region (1710) may include at least one graphic object (1711, 1712, 1713) corresponding to each of a plurality of molecular structures each assigned a different label through labeling (e.g., a first molecular structure assigned a first label (M1), a second molecular structure assigned a second label (M2), a third molecular structure assigned a third label (M3)) and detailed information about the plurality of molecular structures.

[0464] The first area (710) of the service page (1000) may include a first sub-area (1710a) containing graphic objects (1711, 1712, 1713) and a second sub-area (1710b) containing detailed information.

[0465] The first sub-area (1710a) may include a plurality of graphic objects (1711, 1712, 1713) corresponding to each of the plurality of molecular structures. More specifically, the first graphic object (1711) among the plurality of graphic objects may include an image of the first molecular structure corresponding to the first graphic object (1711) among the plurality of molecular structures, and the third graphic object (1713) may include an image of the third molecular structure corresponding to the third graphic object (1713).

[0466] Additionally, the second sub-area (1710b) may provide detailed information on the molecular structure corresponding to one of the plurality of graphic objects (1711, 1712, 1713) selected by user input. The answer generation system (100) may provide detailed information on the selected graphic object based on the user input received for selecting one of the plurality of graphic objects (1711, 1712, 1713).

[0467] In this regard, detailed information on molecular structures corresponding to each of the plurality of graphic objects (1711, 1712, 1713) included in the first sub-area (1710a) may exist in a linked (or associated) manner. For example, let us assume that a third graphic object (1713) corresponding to a third molecular structure is selected from the user terminal (10). Based on the selection of the third graphic object (1713) included in the first sub-area (710a) from the user terminal (10), the answer generation system (100) may provide detailed information (1711a, 1711b, 1711c, 1711d, 1711e) on the third molecular structure linked to the third graphic object (1713) to the second sub-area (1710b). Furthermore, in the second sub-area (1710b), information about the third label (M3) assigned to the third molecular structure may also be displayed.

[0468] As another example, if an answer to a user query includes a new molecular structure (e.g., a third molecular structure), a graphic object corresponding to the new molecular structure included in the first sub-area (1710a) may be automatically selected by the answer generation system (100), so that detailed information about the new molecular structure may be provided in the second sub-area (1710b).

[0469] Next, in a second area (1720) that is distinct from the first area (1710) of the service page (1000), an answer (1721) to a user query (1600) can be provided.

[0470] The answer (1721) provided in the second area (1720) of the service page (1000) may include content indicating that a new specific molecular structure (e.g., a third molecular structure) has been created as a result of a chemical reaction prediction between multiple molecular structures (e.g., a first molecular structure and a second molecular structure).

[0471] Specifically, the answer (1721) may include a molecular structure image (1711a) of a first molecular structure and a molecular structure image (1712a) of a second molecular structure, and may include a molecular structure image (1713a) of a third molecular structure generated as a result of a chemical reaction between the first molecular structure and the second molecular structure.

[0472] Additionally, between the molecular structure image (1711a) of the first molecular structure included in the answer (1721), the molecular structure image (1712a) of the second molecular structure, and the molecular structure image (1713a) of the third molecular structure, a graphic object (e.g., a plus sign, an arrow, etc.) indicating a relationship between the molecular structures corresponding to each image may be displayed together.

[0473] Furthermore, the answer (1721) may include at least one of detailed information about the third molecular structure generated through the chemical reaction prediction model (400) (e.g., the name of the third molecular structure (1713b) and the description of the third molecular structure (1713c)).

[0474] Furthermore, the answer (1221) may be provided with information on the properties of the third molecular structure predicted from the property prediction model (500), and information on the third label (M3) assigned to the third molecular structure may be displayed together in the surrounding area of ​​the molecular structure image (1713a) of the third molecular structure.

[0475] Meanwhile, the answer generation system (100) can receive a new user query including a third label (M3) assigned to a third molecular structure through the service page (1000).

[0476] Specifically, the answer generation system (100) can receive a new user query including a third label (M3) assigned to a third molecular structure through the third area of ​​the service page (1000). For example, as illustrated in FIG. 22, the answer generation system (100) can receive a new user query (e.g., “Is the product m3 the same as this one Molecular structure D?”, 1832) including a third label (M3) assigned to a third molecular structure, based on the selection of a graphic object (1831) included in the third area (1830) from the user terminal (10).

[0477] The answer generation system (100) can input a new user query (1832) into the super-large foundation model (200) based on the receipt of a new user query (1832) including a third label (M3) assigned to a third molecular structure.

[0478] Here, the super-large foundation model (200) can identify a fourth molecular structure corresponding to the name of the specific molecular structure (e.g., “Molecular structure D”) based on the fact that the new user query (1832) contains a name for a new specific molecular structure that is not a labeled content, and can extract content related to the fourth molecular structure. Then, the answer generation system (100) can group the content related to the fourth molecular structure, generate fourth content grouped based on the fourth molecular structure, and assign a fourth label (M4) to the grouped fourth content. Since more specific details related to this have been described above, they will be described only briefly.

[0479] The super-large foundation model (200) can generate an answer to a new user query (1832) by using at least a portion of the detailed information of the third molecular structure corresponding to the third label (M3) and the fourth molecular structure corresponding to the fourth label (M4). For example, the super-large foundation model (200) can generate an answer to a new user query (1832) by using at least a portion of the information included in the grouped third content assigned the third label (M3) and the grouped fourth content assigned the fourth label (M4).

[0480] In this case, the super-large foundation model (200) can utilize at least one prediction model to generate an answer to a new user query (1832). For example, the super-large foundation model (200) can understand the content included in the new user query (1832) and, based on the fact that the user query (1832) includes the content “Do m3 and molecular structure D have the same structure?”, input information about third content grouped based on the third molecular structure and fourth content grouped based on the fourth molecular structure into the chemical reaction prediction model (400).

[0481] The chemical reaction prediction model (400) can compare the connection structures of the third molecular structure and the fourth molecular structure corresponding to a new user query (1832). In addition, the chemical reaction prediction model (400) can output the comparison results for the third molecular structure and the fourth molecular structure as output data.

[0482] Meanwhile, the super-large foundation model (200) determines an answer generation procedure and a tool used in the answer generation procedure to make a prediction corresponding to a new user query (1832), and uses the determined answer generation procedure and tool, the output data of the chemical reaction prediction model (400), and information on the third molecular structure and the fourth molecular structure to generate an answer (e.g., “m3 and m4 are connected in different relationships...”, 1921) for the user query (1832) (see FIG. 23).

[0483] When an answer (1921) is generated, as illustrated in FIG. 23, the answer generation system (100) can provide an answer (1921) to a user query (1832) generated from a super-large foundation model (200) in a second area (1920) of a service page (1000).

[0484] In this case, the answer (1921) provided in the second area (1920) of the service page (1000) may include content indicating a comparison result for a connection structure between a plurality of molecular structures (e.g., a third molecular structure and a fourth molecular structure). For example, the answer (1921) may include at least one of a label (M3, M4) assigned to each of the third molecular structure and the fourth molecular structure, a molecular structure image (1913a, 1914a) of each of the third molecular structure and the fourth molecular structure, and content (1921a) describing a comparison result for a connection structure between the third molecular structure and the fourth molecular structure.

[0485] Furthermore, the first area (1910) of the service page (1000) may include at least one of a fourth graphic object (1914) corresponding to the fourth molecular structure, a molecular structure image (1914a) of the fourth molecular structure linked to the fourth graphic object (1914), a name of the fourth molecular structure (1914b), a description of the fourth molecular structure (1914c), a SMILES notation for the fourth molecular structure (1914d), and a physical property characteristic (1914e) of the fourth molecular structure.

[0486] Meanwhile, the service page provided by the answer generation system (100) may be provided to users in various forms, in addition to the form of the service page (1000) discussed above. Below, the service pages of the answer generation system (100) provided in various forms will be examined in more detail.

[0487] In one embodiment, as illustrated in FIG. 24, the answer generation system (100) can provide a service page (2400) linked to the system (100) to the user terminal (10).

[0488] The service page (2400) may include at least one of a first area (2410) for receiving an answer to a user query and a second area (2420) for providing an answer to the user query. The answer generation system (100) may receive a user query (e.g., “What about a reaction between 2-bromoethanol and ethane sulfonyl chloride. First look for their smiles and then predict their reaction.” , 2411) including information on at least one molecule input through the first area (2410) of the service page (2400).

[0489] The super-large foundation model (200) can extract SMILES strings for each of the first compound and the second compound from the storage unit (140) based on the fact that the user query (2411) includes the content “What is the reaction between the first compound (2-bromoethanol) and the second compound (ethane sulfonyl chloride)? First, find their SMILES strings.” and input information about the first compound and the second compound into the chemical reaction prediction model (400). The chemical reaction prediction model (400) can predict the result of the chemical reaction between the first molecular structure and the second molecular structure under specific conditions and output the predicted result.

[0490] Furthermore, as illustrated in FIGS. 24 and 25, the super-large foundation model (200) can generate an answer to a user's (2411) query (e.g., “The predicted product from the reaction of 2-bromoethanol and ethanesulfonyl chloride is S-ethyl isobutyl ethanesulfonothioate. This organic compound is commonly used as a pesticide. This reaction is a process in which the sulfonyl chloride group of ethanesulfonyl chloride substitutes the hydroxyl group of 2-bromoethanol, thereby forming a sulfonothioester.”, 2421) using the chemical reaction prediction result between the first molecular structure and the second molecular structure output from the chemical reaction prediction model (400). For example, the answer (2421) may include molecular structure images (or molecular graphs, 2421a, 2421b) of each of the specified first molecular structure and second molecular structure, and detailed information (e.g., a smiley string) for each of the molecular structures. In addition, the answer (2421) may include molecular structure images (2421c) of products generated as a result of chemical reaction prediction, and detailed information (e.g., a smiley string) for the products (or molecular structures).

[0491] Next, the answer generation system (100) may receive a new user query containing information about a product generated as a result of a chemical reaction prediction. For example, as illustrated in FIG. 25 , the answer generation system (100) may receive a user query containing information about a new product (e.g., “Is the generated compound identical to this compound?” 2511) entered through the first area (2410) of the service page (2400).

[0492] The super-large foundation model (200) can input information about the product and the comparison target compound included in the new user query (2511) into the chemical reaction prediction model (400) based on the content included in the new user query (2511). The chemical reaction prediction model (400) can compare the molecular structure of the new product corresponding to the new user query (2511) and the connection structure of the molecular structure of the comparison target compound, and output the comparison result as output data.

[0493] The super-large foundation model (200) can generate an answer to a user query (2511) using the output data of the chemical reaction prediction model (400). For example, as illustrated in FIG. 26, the super-large foundation model (200) can generate an answer (e.g., “The two compounds have different arrangements of atoms.”, 2521) to a user query (2511) and provide it through a service page (2400). In this case, the answer (2521) can include contents indicating a comparison result for the molecular structure of the new product and the molecular structure of the comparison target compound (e.g., a molecular structure image (or molecular graph, 2521a) of the new product and a molecular structure image (2521b) of the comparison target compound, SMILES strings for each of the new product and the comparison target compound, etc.).

[0494] Meanwhile, a service page linked to the answer generation system (100) may include various functions and be provided to the user.

[0495] In one embodiment, as illustrated in FIG. 27, the service page (2700) may include at least one of a first area (2701) including a function for setting various reaction conditions (or experimental conditions) such as temperature and pressure, a second area (2702) including a function for selecting a reactant to be used for predicting a chemical reaction, and a third area (2703) including a function for selecting a reagent and / or solvent required for the reaction.

[0496] In another embodiment, the service page (2700) may include a fourth area (2704) that includes a function for deriving the direction of the reaction or the starting material when the user selects at least one of the forward reaction simulation or the reverse reaction simulation. In addition, the service page (2700) may include a fifth area (2705) that comprehensively provides experimental protocols, conditions, compound information, reaction types, etc.

[0497] In another embodiment, the service page (2700) may include a sixth area (2706) that provides information about the main product predicted from the simulation results, information about the yield predicted based on the reaction conditions, and information about the predicted by-products and impurities.

[0498] In this way, the present invention can input reactants, reagents, and experimental conditions, utilize text data to predict reaction results, and provide visualizations of predicted products, yields, and byproducts. This allows the present invention to help users efficiently discover synthetic pathways and establish optimal experimental conditions.

[0499] Meanwhile, as discussed above, a user may have a user account pre-registered in the answer generation system (100) according to the present invention.

[0500] Accordingly, various information related to a user account can be stored in the storage unit (140) of the answer generation system (100). Here, the information related to the user account can include at least one of the user's (or user account's) history information and the user's metadata (e.g., name, gender, age, major, occupation, workplace (or company), etc.).

[0501] More specifically, the user's history information may include information related to various events occurring in the user account. For example, the events occurring in the user account may include at least one of: i) entering a user query to obtain an answer from the super-large foundation model (200), ii) entering (or selecting) a document, and iii) entering a new (or additional) query for an answer generated from the super-large foundation model (200).

[0502] Based on these events, the user's history information may include at least one of: i) a user query entered by the user, ii) document information entered by the user (or the user's document input history), iii) content extracted from a document entered by the user (e.g., a specific molecular structure and information related to the specific molecular structure), and iv) an answer of the super-large foundation model (200) to the user query.

[0503] Accordingly, in the storage unit (140) of the answer generation system (100), at least one of the document to be analyzed, the extracted molecular structure, the label for (or assigned to) the extracted molecular structure, the user query, and the answer to the user query may be stored in connection with a user account.

[0504] Here, the information related to the extracted molecular structure may be i) information extracted from the document to be analyzed, or ii) information extracted from a user query. Alternatively, the information related to the extracted molecular structure may be i) information generated from a chemical reaction prediction model (400), or ii) information generated from a molecular property prediction model (500).

[0505] Furthermore, the storage unit (140) of the answer generation system (100) may store various information related to at least one of chemistry, biology, new materials, new substances, and new drug development. For example, the storage unit (140) may store data such as text, molecular structures, formulas, charts, tables, and images related to at least one of chemistry, biology, new materials, new substances, and new drug development.

[0506] Accordingly, the present invention can provide a user environment that allows the user to utilize various functions by using the DB constructed in the answer generation system (100).

[0507] As illustrated in FIG. 28, the answer generation system (100) can provide a service page (2800) to the user terminal (10) that is configured to explore the physical and chemical properties of a specific compound (or product or molecule) or to predict missing properties by using various information stored in the storage unit (140).

[0508] In one embodiment, the first area (2810) of the service page (2800) may provide detailed information about a compound selected from the user terminal (10). For example, the first area (2810) of the service page (1000) may provide the structure of the compound along with its physical and chemical properties.

[0509] In another embodiment, the second area (2820) of the service page (2800) may include a function that allows a user to search for similar compounds. For example, the answer generation system (100) may receive a search request for a compound structurally similar to a specific compound from a user terminal (10) via the second area (2820), extract information on a compound structurally similar to the specific compound from the storage unit (140), and provide information on the extracted similar compound on the second area (2820).

[0510] In addition, the present invention can provide a user environment that allows a user to explore various chemical reaction data or explore chemical reaction data similar to a specific chemical reaction based on various chemical reaction data stored in a storage unit (140).

[0511] In one embodiment, as illustrated in FIG. 29, the answer generation system (100) may provide a service page (2900) to the user terminal (10) that allows the user to explore various chemical reaction data. The user may explore data on various chemical reactions through the service page (2900), and the chemical reaction data may include, for example, at least one of reactants, products, and reaction conditions.

[0512] In this case, the response generation system (100) can provide various chemical reaction data in visualized form with options selected by the user. For example, one area of ​​the service page (2900) may include graphic objects (2901, 2902, 2903) linked to various visualization options.

[0513] In one embodiment, the answer generation system (100) can provide various chemical reaction data (2911, 2912) stored in the storage unit (140) in a list format when a first graphic object (2901) is selected from the user terminal (10).

[0514] In another embodiment, the answer generation system (100) may provide various chemical reaction data (2911, 2912) stored in the storage unit (140) in a table format when a second graphic object (2902) is selected from the user terminal (10).

[0515] In another embodiment, the answer generation system (100) can provide various chemical reaction data (2911, 2912) stored in the storage unit (140) in a graph format when a third graphic object (2903) is selected from the user terminal (10).

[0516] Furthermore, the present invention can provide a user environment that enables users to conduct chemical synthesis research and experiments more efficiently and systematically.

[0517] As illustrated in FIG. 30, the answer generation system (100) may provide a service page (3000) that allows a user to conduct research on various synthetic pathways related to forward and reverse synthesis. In this case, the service page (3000) may support both forward and reverse simulations, and may include functions that provide information on various pathways for synthesizing compounds, provide information on synthesizable compounds having specific properties, or allow a user to select a compound having desired physical and chemical properties and plan a synthetic pathway.

[0518] In one embodiment, a first region (3010) of a service page (3000) may provide a reaction pathway between selected compounds. The first region (3010) may provide a visual representation of the reaction mechanism, including major products and by-products.

[0519] In another embodiment, the second area (3010) of the service page (3000) may provide a sorted list of synthesizable compounds and may include functionality to add reagents and solvents required for the reaction.

[0520] In another embodiment, various reaction pathways may be visualized and provided in the third area (3030) of the service page (3000). In this case, compounds produced along a specific reaction pathway may be visualized step by step and provided in the third area (3030). Furthermore, the fourth area (3040) of the service page (3000) may provide information on major products and by-products predicted from the simulation results, and for each reaction pathway, information on predicted yields, detailed information on major products and reaction conditions, expected impurities, and the amount of impurities may be provided.

[0521] Meanwhile, a service page where a user query is received may provide information on at least one molecular structure assigned a label. For example, as illustrated in FIG. 31, a third area (3130) of a service page (3100) may provide a plurality of graphic objects (3131, 3132, 3133, 3134, 3135) corresponding to a plurality of molecular structures, each of which includes different labels (M1, M2, M3, M4, M5). In this case, the plurality of graphic objects (3131, 3132, 3133, 3134, 3135) may include a molecular structure image of the molecular structure or a molecular graph for the molecular structure.

[0522] In one embodiment, the answer generation system (100) may provide a visualization of the first compound as an answer (3111) to the user query (3110) based on the user query (3110) “Please provide a visualization of the first compound (e.g., ethanol)” received from the user terminal (10). In this case, the answer generation system (100) may assign a first label (M1) to the molecular structure (3131) of the first compound based on the fact that the molecular structure of the first compound is included in the answer (3111), and update the molecular structure (3131) of the first compound assigned the first label (M1) in the third region (3130).

[0523] In another embodiment, the answer generation system (100) may provide a visualized molecular structure of a compound similar to the molecular structure of the first compound as an answer (3121) to the user query (3120) based on receiving a user query (3120) from the user terminal (10) requesting to provide a visualized molecular structure of a compound similar to the molecular structure of the first compound. In this case, the answer generation system (100) may assign a second label (M2) to the molecular structure (3132) of the new compound based on the fact that the molecular structure of the new compound (or the second compound) is included in the answer (3121), and update the molecular structure (3132) of the new compound assigned the second label (M2) in the third region (3130).

[0524] Furthermore, as illustrated in FIG. 32, the answer generation system (100) may generate Python code corresponding to the user's query and provide an answer (3211) including the Python code, based on a user query (3210) received through a service page (3200) that includes the content, “Please check if there is experimental information on the boiling points of the compounds, and if not, please provide a Python code for estimating the boiling points of the two compounds.”

[0525] That is, users can be provided with Python code to query experimental boiling point information of a compound or to obtain an estimated value when experimental data is not available.

[0526] As discussed above, according to the method and system for generating answers for a super-large foundation model according to the present invention, an answer suitable for a user's query is generated and provided based on data extracted from a document, thereby suggesting an optimal research method to the user, thereby minimizing the risk of research failure.

[0527] Furthermore, the method and system for generating answers using a super-large foundation model according to the present invention can provide answers to user queries using data extracted from documents or generated from a pre-trained predictive model. This allows users to quickly and accurately receive the information they need, while reducing the time and costs required for research and / or development.

[0528] Furthermore, the method and system for generating answers using a super-large foundation model according to the present invention can generate answers to user queries using predicted results from a pre-trained predictive model and provide the generated answers to the user. This allows users to shorten the time required for research and / or development and reduce the number of trial and errors in research and / or development.

[0529] Furthermore, according to the method and system for generating answers for a super-large foundation model according to the present invention, the extracted molecular structure and related data are provided in a visualized manner through a user interface, thereby enabling the user to intuitively recognize the information required and understand it more quickly, thereby increasing the accuracy and efficiency of research.

[0530] Meanwhile, the present invention discussed above can be implemented as a program that is executed by one or more processes on a computer and can be stored on a medium (or recording medium) that can be read by the computer.

[0531] Furthermore, the present invention discussed above can be implemented as computer-readable code or instructions on a program-recorded medium. In other words, the present invention can be provided in the form of a program.

[0532] Meanwhile, computer-readable media include all types of recording devices that store data that can be read by a computer system. Examples of computer-readable media include hard disk drives (HDDs), solid-state disk drives (SSDs), silicon disk drives (SDDs), ROMs, RAMs, CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices.

[0533] Furthermore, the computer-readable medium may include a storage device and may be a server or cloud storage device accessible by an electronic device via communication. In this case, the computer may download the program according to the present invention from the server or cloud storage device via wired or wireless communication.

[0534] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, i.e., a CPU (Central Processing Unit), and there is no particular limitation on its type.

[0535] Meanwhile, the detailed description above should not be construed as limiting in any respect and should be considered illustrative. The scope of the present invention should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the present invention are intended to be included within the scope of the present invention.

Claims

1. In a method in which memory and at least one processor are performed cooperatively, A step of specifying a plurality of molecular structures to be predicted based on user input received through a service page; A step of storing information on the plurality of molecular structures in the above memory; A step of obtaining a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures stored in the memory; A step of receiving a user query for chemical reaction prediction related to the plurality of molecular structures through the above service page; A step of processing the molecular graph for the plurality of molecular structures as input to a chemical reaction prediction model so as to predict a chemical reaction corresponding to the user query; In the above chemical reaction prediction model, a step of performing chemical reaction prediction for the plurality of molecular structures; A step of obtaining a result of chemical reaction prediction for the plurality of structures from the chemical reaction prediction model; and An answer generation method comprising a step of generating an answer to the user query using the result of the chemical reaction prediction.

2. In paragraph 1, The above chemical reaction prediction model is, A first module that receives molecular structures as input and predicts graph-based chemical reactions; A second module for analyzing information related to prediction of chemical reactions associated with the plurality of molecular structures from text data, In the step of performing the above chemical reaction prediction, An answer generation method characterized in that the result according to the chemical reaction prediction is generated by using the prediction result of the first module and the analysis result of the second module.

3. In paragraph 2, A method for generating an answer, characterized in that in the step of performing the above chemical reaction prediction, the result predicted through the first module is verified using the output data analyzed in the second module.

4. In paragraph 1, The above plurality of molecular structures include a first molecular structure and a second molecular structure, In the step of obtaining a molecular graph for the above multiple molecular structures, By converting the atoms constituting the first molecular structure into nodes and converting the bonding relationships between the atoms constituting the first molecular structure into edges, a first molecular graph including nodes and edges corresponding to the first molecular structure is obtained, An answer generation method characterized in that the second molecular graph including nodes and edges corresponding to the second molecular structure is obtained by converting atoms constituting the second molecular structure into nodes and converting bonding relationships between atoms constituting the second molecular structure into edges.

5. In paragraph 4, A step of performing labeling so that a different label is assigned to each of the above plurality of molecular structures; and An answer generation method characterized by further comprising a step of storing, in the memory, information about the first molecular graph obtained through the process of obtaining the molecular graph and the label assigned to the first molecular graph, and information about the second molecular graph obtained through the process of obtaining the molecular graph and the label assigned to the second molecular graph.

6. In paragraph 5, When the user query including the plurality of molecular structures with the different labels is received, in the step of generating the answer, Processing the plurality of molecular structures assigned different labels as inputs to the chemical reaction prediction model, Using the result of the chemical reaction prediction obtained from the chemical reaction prediction model, an answer to the user query is generated, A method for generating an answer, characterized in that, if a specific molecular structure generated through the chemical reaction prediction model is included in the answer to the user query, a label is assigned to the specific molecular structure.

7. In paragraph 5, Further comprising a step of providing a plurality of graphic objects corresponding to each of the plurality of molecular structures to which the different labels are assigned, in an area of ​​the service page where the user query is received, A method for generating an answer, characterized in that each of the plurality of graphic objects includes the first molecular graph and the second molecular graph.

8. In paragraph 7, A method for generating an answer, characterized in that, based on receiving a user input for selecting one of the plurality of graphic objects, detailed information corresponding to the graphic object selected according to the user input is provided on the service page.

9. In paragraph 4, In the step of performing chemical reaction prediction for the above multiple molecular structures, Using information about the above plurality of molecular structures, an embedding vector corresponding to the molecular graph is obtained, Performing an attention operation related to the interaction between atoms of the above plurality of molecular structures, and updating the embedding vector based on the operation, Using the updated embedding vector, bond prediction and atom prediction are performed as chemical reactions of the plurality of molecular structures. An answer generation method characterized in that the result of the chemical reaction prediction is obtained by using the result of the above combination prediction and the result of the above atomic prediction from the chemical reaction of the plurality of molecular structures.

10. In paragraph 9, The above updated embedding vector is, An answer generation method characterized in that the attention score calculated based on the result of the above attention operation is updated by adding different biases according to the type of connection between nodes constituting the first molecular graph and the second molecular graph.

11. In paragraph 9, The above combined prediction is performed by performing a dot-product operation using the updated embedding vector, The above atomic prediction predicts the atomic properties of the atoms corresponding to the updated embedding vector by using the atomic property probability distribution of each of the atoms corresponding to the updated embedding vector, A method for generating an answer, characterized in that the above atomic properties include charge states of the atoms that can change during a chemical reaction process of the plurality of molecular structures.

12. In paragraph 11, A method for generating an answer, characterized in that the result of the above chemical reaction prediction corresponds to a final result stabilized through a diffusion feedback process.

13. In paragraph 7, The above service page is, A first region in which the plurality of graphic objects corresponding to each of the plurality of molecular structures to which the different labels are assigned and detailed information about the plurality of molecular structures are provided, A second area where answers to the above user inquiries are provided, and At least one of the third areas receiving the user query, Detailed information on the above multiple molecular structures is as follows: Each of the above multiple molecular structures includes at least one of a molecular structure image, a name, a physical property characteristic, and a string according to the SMILES notation, The above detailed information is extracted from the user input or obtained from at least one pre-learned prediction model, The above-mentioned pre-trained prediction model is, An answer generation method characterized by including at least one of a chemical reaction prediction reaction model that predicts a chemical reaction between molecular structures and a molecular property prediction model that predicts a property of a molecular structure.

14. In paragraph 4, A step of receiving an editing request for the plurality of molecular structures through a service page that provides an answer to the user's query; and A method for generating an answer, characterized in that it further comprises a step of providing an editing interface that provides an editing function for the plurality of molecular structures on the service page.

15. In paragraph 14, An answer generation method characterized in that the above editing interface provides a molecular graph corresponding to at least one of the first molecular structure and the second molecular structure obtained through the molecular graph transformation in an editable state.

16. In paragraph 15, Editing of the above multiple molecular structures, Deletion or change of position of at least one of the nodes corresponding to each of the atoms constituting each of the above plurality of molecular structures and the edges indicating the bonding relationship of the atoms, A method for generating an answer characterized by adding a new node corresponding to a new atom or adding a new edge that creates a new bonding relationship between the atoms.

17. In paragraph 16, Among the above multiple molecular structures, the molecular structure in which editing has been performed is stored in the memory, The above edited molecular structure is given a new label that specifies the above edited molecular structure. When a user query containing the above new label is input to the super-large foundation model, An answer generation method characterized in that the above-mentioned large foundation model generates an answer using an edited molecular structure corresponding to the new label.

18. In the answer generation system, The system comprises a memory and at least one processor, The above memory and the above processor cooperate to: Based on user input received through the service page, specify multiple molecular structures to be predicted, In the above memory, information about the plurality of molecular structures is stored, Based on the plurality of molecular structures stored in the memory, a molecular graph for the plurality of molecular structures is obtained by converting atoms into nodes and bonds between atoms into edges, Through the above service page, a user query is received for a chemical reaction prediction related to the plurality of molecular structures, Processing the molecular graphs for the plurality of molecular structures as input to a chemical reaction prediction model so that a chemical reaction corresponding to the user query is predicted, In the above chemical reaction prediction model, chemical reaction prediction is performed for the plurality of molecular structures, From the above chemical reaction prediction model, the results of chemical reaction prediction for the plurality of structures are obtained, An answer generation system characterized in that it generates an answer to the user query using the result of the chemical reaction prediction.

19. A program stored on a computer-readable medium and executed by one or more processes on an electronic device, The above program is, A step of specifying a plurality of molecular structures to be predicted based on user input received through a service page; A step of storing information on the plurality of molecular structures in the above memory; A step of obtaining a molecular graph for the plurality of molecular structures by converting atoms into nodes and bonds between atoms into edges based on the plurality of molecular structures stored in the memory; A step of receiving a user query for chemical reaction prediction related to the plurality of molecular structures through the above service page; A step of processing the molecular graph for the plurality of molecular structures as input to a chemical reaction prediction model so as to predict a chemical reaction corresponding to the user query; In the above chemical reaction prediction model, a step of performing chemical reaction prediction for the plurality of molecular structures; A step of obtaining a result of chemical reaction prediction for the plurality of structures from the chemical reaction prediction model; and A program stored on a computer-readable recording medium, characterized in that it includes commands for performing a step of generating an answer to the user query using the result of the chemical reaction prediction.

Images