Word embedding method and device and word search method

Abstract

A word embedding method and device and a word search method are provided, wherein the word embedding method includes: training a word embedding model based on characteristic information of chemical substances; and obtaining an embedding vector of a word representing the chemical substance from the word embedding model, wherein the word embedding model is configured to predict a context word of an input word.

Description

[0001] This application claims the benefit of Korean Patent Application No. 10-2019-0127032, filed on October 14, 2019, in the Korean Intellectual Property Office, the disclosure of which is incorporated herein in its entirety for all purposes. Technical Field

[0002] The following description relates to word embedding and word search methods and devices. Background Art

[0003] A large amount of knowledge has been published as text, such as papers and books. Such accumulated knowledge described in free text is in a form that users understand, and efforts are underway to extract structured knowledge from text using natural language processing (NLP) technology. Summary of the Invention

[0004] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

[0005] In one general aspect, a word embedding method is provided, comprising: training a word embedding model based on characteristic information of a chemical substance; and obtaining an embedding vector of a word representing the chemical substance from the word embedding model, wherein the word embedding model is configured to predict a context word of an input word.

[0006] The step of training the word embedding model may include: training the word embedding model based on any one or any combination of structural information, composition information, and physical property information of the chemical substance.

[0007] The step of training the word embedding model may include: training the word embedding model to output context words of words representing the chemical substance in response to structural information of the chemical substance being input into the word embedding model.

[0008] The structural information of the chemical substance is determined based on a format of one of a fingerprint, a simplified molecular linear input specification (SMILES), a graph, or an image.

[0009] The step of training the word embedding model may include: training the word embedding model to output context words of words representing the chemical substance from the word embedding model in response to component information of the chemical substance being input into the word embedding model.

[0010] The composition information of chemical substances can be obtained from words representing chemical substances.

[0011] Words representing chemical substances may be divided into letters or elements, and the letters or elements may be sequentially input into a word embedding model.

[0012] The step of training the word embedding model may include: training the word embedding model to output physical property information of the chemical substance from the word embedding model.

[0013] Physical property information may include information about any one or any combination of mass, volume, color, melting point, and boiling point of a chemical substance.

[0014] The word embedding method may include: inputting the embedding vector into a portion corresponding to a word representing a chemical substance in a word embedding matrix corresponding to a word embedding model.

[0015] The word embedding method may include determining whether a word having an embedding vector to be generated represents a chemical material.

[0016] In one general aspect, another word search method is provided, comprising: receiving characteristic information of a chemical substance or a word representing the chemical substance; and outputting a word representing a substance having similar characteristics to the chemical substance based on a word embedding matrix, wherein the word embedding matrix is ​​obtained from a word embedding model trained based on characteristic information of multiple chemical substances, and the word embedding model is configured to predict context words of the input word.

[0017] The characteristic information of the chemical substance may include any one or any combination of structural information, component information, and physical property information of the chemical substance.

[0018] A word embedding device includes: a processor configured to: train a word embedding model based on characteristic information of chemical substances, and obtain an embedding vector of a word representing the chemical substance from the word embedding model; and a word embedding model configured to predict a context word of an input word.

[0019] The processor may be configured to train a word embedding model based on any one or any combination of structural information, composition information, and physical property information of the chemical substance.

[0020] The processor may be configured to train the word embedding model to output context words of words representing the chemical substance in response to structural information of the chemical substance being input into the word embedding model.

[0021] The processor may be configured to train the word embedding model to output context words of words representing the chemical substance in response to component information of the chemical substance being input into the word embedding model.

[0022] The processor can be configured to: train a word embedding model to output physical property information of the chemical substance.

[0023] The processor may be configured to: input the embedding vector into a portion corresponding to the word representing the chemical substance in the word embedding matrix corresponding to the word embedding model.

[0024] In one general aspect, another word embedding device is provided, comprising: a processor configured to: divide a sentence into a plurality of words, determine whether a word among the plurality of words represents a chemical substance, divide the determined word into a plurality of terms smaller than the determined word, and sequentially input the determined word and any one or any combination of one or more terms among the plurality of terms into a word embedding matrix to output a context word, wherein the word embedding matrix is ​​obtained from a word embedding model trained based on characteristic information of the chemical substance, and the word embedding model is configured to predict the context word.

[0025] The word embedding device may include: a non-transitory computer-readable storage medium storing structural information, composition information, and physical property information of a chemical substance; and a processor configured to: retrieve any one or any combination of the structural information, composition information, and physical property information of the chemical substance from the non-transitory computer-readable storage medium, and input any one or any combination of the structural information, composition information, and physical property information of the chemical substance into a word embedding matrix.

[0026] Context words may represent substances that have similar properties to chemical substances.

[0027] Other features and aspects will be apparent from the following detailed description, drawings, and claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 Shows an example of a word embedding model.

[0029] Figure 2 An example of a process of generating an embedding vector is shown.

[0030] Figure 3 Shows an example of the process of training a word embedding model.

[0031] Figure 4 and Figure 5 An example of the operation of a word embedding device is shown.

[0032] Figure 6 An example of a process of performing post-processing on a word embedding matrix is ​​shown.

[0033] Figure 7 Shows an example of a word embedding method.

[0034] Figure 8 An example of a word search method is shown.

[0035] Figure 9 An example of an electronic device is shown.

[0036] Throughout the drawings and detailed description, unless otherwise described or provided, the same drawing reference numerals will be understood to refer to the same elements, features, and structures. The drawings may not be to scale, and the relative sizes, proportions, and depictions of the elements in the drawings may be exaggerated for clarity, illustration, and convenience. DETAILED DESCRIPTION

[0037] The following detailed description is provided to help the reader gain a comprehensive understanding of the methods, devices, and / or systems described herein. However, various changes, modifications, and equivalents of the methods, devices, and / or systems described herein will be clear after understanding the disclosure of the present application. For example, the order of operations described herein is merely an example and is not limited to the order set forth herein, but may be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a particular order. In addition, descriptions of features known in the art may be omitted for clarity and conciseness.

[0038] The features described herein can be implemented in different forms and are not to be construed as limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein that will be apparent after understanding the disclosure of this application.

[0039] The following structural or functional descriptions of the examples disclosed in this disclosure are intended only for the purpose of describing the examples, and the examples can be implemented in various forms. The examples are not intended to be limiting, but rather to encompass various modifications, equivalents, and alternatives within the scope of the claims.

[0040] Although the terms "first" or "second" are used to explain various components, the components are not limited by these terms. These terms should only be used to distinguish one component from another. For example, within the scope of the rights according to the concept of the present disclosure, a "first" component may be referred to as a "second" component, or similarly, a "second" component may be referred to as a "first" component.

[0041] It will be understood that when a component is referred to as being “connected to” another component, it can be directly connected or coupled to the other component or intervening components may be present.

[0042] Unless the context clearly indicates otherwise, as used herein, the singular is intended to include the plural. It should also be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of stated features, integers, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0043] Hereinafter, examples will be described in detail with reference to the accompanying drawings, and like reference numerals refer to like elements throughout.

[0044] Figure 1 An example of a word embedding model 100 is shown.

[0045] Reference Figure 1 , the word embedding model 100 includes an input layer 110 , a hidden layer 120 and an output layer 130 .

[0046] The word embedding model 100 is a model for expressing words in the form of vectors, and may be a neural network based on Word2Vec that discovers hidden knowledge in a text in an unsupervised manner even in the absence of domain knowledge. Word2Vec may be a scheme for obtaining a vector representation or embedding vector of a word based on words that appear around the relevant word based on a high probability that the relevant word appears in the same sentence. For example, when a word embedding model 100 that uses a word as input to predict words that appear around the word is trained, the hidden representation of the word embedding model 100 may be obtained as an embedding vector of the word. As described above, the word embedding model 100 may be referred to as Word2Vec for converting words into vectors, and in particular, also referred to as a “skip-gram model”.

[0047] The word embedding model 100 is trained so that in response to the t-th word in a sentence being input to the input layer 110, the “n” words (e.g., Figure 1 The two words in ( t ) are output from the output layer 130, and the embedding vector of the t-th word is obtained in the hidden layer 120 of the trained word embedding model 100. Since the words appearing around similar words are similar to each other, similar embedding vectors can be obtained. In addition, the embedding vector may include semantic information of the corresponding word. In addition, w(t) represents a one-hot vector that is a word ID representing the t-th word.

[0048] The t-th word input to the input layer 110 is referred to as a "target word," and the "n" words on each of the right and left sides of the t-th word output from the output layer 130 are referred to as "context words." In addition, the hidden layer 120 includes at least one layer and is referred to as a "projection layer."

[0049] For example, when the target word is a word representing a chemical substance, the word embedding model 100 may be trained based on the characteristics of the chemical substance to determine an embedding vector such that the characteristics of the chemical substance are reflected, which will be further described below.

[0050] In one example, the word embedding model 100 can be implemented as an artificial neural network including a two-dimensional convolutional neural network (CNN) and a pre-trained spatial pyramid pooling network. In one example, the CNN can be a deep neural network (DNN). In one example, the DNN can include a region generation network (RPN), a classification network, a reinforcement learning network, a fully connected network (FCN), a deep convolutional network (DCN), a long short-term memory (LSTM) network, and a gated recurrent unit (GRU). In one example, the CNN includes a plurality of layers, each layer including a plurality of nodes. In addition, the CNN includes connection weights connecting a plurality of nodes included in each layer of the plurality of layers to a node included in another layer of the CNN.

[0051] In one example, a CNN may receive a target word. In such an example, a convolution operation is performed on the target word and a kernel, and as a result, a feature map is output. The output feature map is used as the input feature map and the kernel is convolved again, and a new feature map is output. When the convolution operation is repeatedly performed in this manner, a recognition result regarding the features of the target word can be ultimately outputted through the output layer of the CNN.

[0052] Figure 2 An example of a process of generating an embedding vector is shown.

[0053] Figure 2 Various elements that may be used to determine the embedding vector 250 are shown.

[0054] When a common word is not associated with a chemical substance, context information 210 shown in the literature may be used to determine an embedding vector 250. For example, the embedding vector 250 may be determined using a word embedding model trained based on a target word and context words.

[0055] For example, when a word represents a chemical substance, the embedding vector 250 may be determined based on the characteristics of the chemical substance. In this example, the characteristics of the chemical substance include composition information 220, structure information 230, and physical property information 240. The composition information 220 is contained in the chemical substance word, so the vocabulary information is encoded and used as the composition information 220. The structure information 230 indicates the chemical structure or molecular structure of the chemical substance and may be determined based on the format of any one of a fingerprint, a simplified molecular linear input specification (SMILES), a graph, and an image. The physical property information 240 includes information about any one or any combination of the mass, volume, color, melting point, and boiling point of the chemical substance. It may be determined by reflecting any one or any combination of the context information 210, the composition information 220, the structure information 230, and the physical property information 240 described above. Figure 2Embedding vector 250. Furthermore, context words or chemical substance words with similar characteristics also have similar physical properties, and thus similar embedding vectors can be obtained. Chemical substances with similar characteristics can be retrieved based on the cosine similarity between the embedding vectors. When the embedding vectors are represented in a two-dimensional (2D) plane, the embedding vectors of chemical substance words with similar characteristics can be located adjacent to each other.

[0056] Figure 3 Shows an example of the process of training a word embedding model.

[0057] Figure 3 A word embedding model is shown for determining embedding vectors for words representing chemical substances. Words appearing in literature can be, for example, common words unrelated to chemical substances, or words representing chemical substances. In one example, a chemical entity recognizer can determine whether a word represents a chemical substance. For example, when the word being modeled represents a chemical substance, an embedding vector can be obtained using the operations described below.

[0058] As described above, the composition information is contained in the words representing the chemical substances. When the words representing the chemical substances are extracted from the literature, the composition information can be obtained by performing lexicon-level encoding. The obtained composition information is input to the input layer 310.

[0059] In addition, the structural information of the chemical substance determined by the chemical structure coding may be acquired from a database (DB), and the acquired structural information may be input to the input layer 310 .

[0060] For a word representing a chemical substance, component information and / or structure information of the chemical substance may be input to the input layer 310 of the word embedding model. A word ID indicating a word may not be input to the input layer 310.

[0061] In addition, physical property information of chemical substances can be obtained from the database. The word embedding model can be trained so that the obtained physical property information can be output from the output layer 330.

[0062] In one example, a word embedding model can be trained to output context words representing a chemical substance in response to input of compositional information and / or structural information of a chemical substance into the word embedding model. In another example, a word embedding model can be trained to output context words representing a chemical substance and physical property information of the chemical substance in response to input of compositional information and / or structural information of the chemical substance into the word embedding model. The word embedding model can predict both the physical property information of the chemical substance and the context words, and can also be trained using multi-task learning.

[0063] As described above, the composition information and structure information, which are inherent characteristics of chemical substances, can be used as training input data of the word embedding model, and the physical property information can be used as training output data to perform training, thereby obtaining an embedding vector that reflects any one or any combination of the composition information, structure information, and physical property information of the chemical substance.

[0064] Figure 4 and Figure 5 An example of the operation of a word embedding device is shown.

[0065] Figure 4 4 is a diagram illustrating an example of the operation of a word embedding device. The word-level tokenizer 410 divides sentences appearing in a document into words. The chemical entity identifier 420 determines whether a word having an embedding vector to be generated represents a chemical substance. The dictionary-level tokenizer 430 divides a word into terms smaller than a word. For example, the dictionary-level tokenizer 430 can segment words based on letters or elements. In one example, the word "CuGaTe2" can be divided into letters "C", "u", "G", "a", "T", "e", and "2", or into elements "Cu", "Ga", "Te", and "2". The letters or elements can be sequentially input into the word embedding model 440. The word embedding model 440 is the model to be finally trained, and the DB 450 stores the structural information and physical property information of the chemical substance. Depending on the situation, the DB 450 may be located outside the word embedding device and may be connected to the word embedding device via a wired network and / or a wireless network, or may be included in the word embedding device. The model post-processor 460 performs post-modification on the trained word embedding model 440. Since an embedding vector is obtained for each word ID of a common word that is not related to a chemical substance, but the word ID of a word representing a chemical substance is not used in training, a post-processing process may be performed. Figure 6 The post-processing process is further described.

[0066] Figure 5 is a diagram illustrating an example of the operation of a word embedding device. Although the order of some operations may be changed or some operations may be omitted without departing from the spirit and scope of the illustrative examples described, Figure 5 The operations in can be performed in the order and manner as shown in the figure. Figure 5 Many of the operations shown in FIG. 1 may be performed in parallel or simultaneously. Figure 5 One or more blocks and combinations of blocks may be implemented by computers and devices (such as processors) based on dedicated hardware that perform the specified functions, or a combination of dedicated hardware and computer instructions. Figure 5 In addition to the description, Figures 1 to 4 The description also applies to Figure 5Therefore, the above description may not be repeated here.

[0067] In operation 510, the word embedding device extracts a target word and context words with an embedding vector to be generated from the document. In operation 520, the word embedding device determines whether the target word represents a chemical substance. In one example, when the target word is a common word unrelated to a chemical substance, operation 550 may be performed. In another example, when the target word represents a chemical substance, operation 530 may be performed. In operation 530, the word embedding device obtains the composition information of the chemical substance by dividing the target word into terms. In operation 540, the word embedding device obtains the structural information and physical property information of the chemical substance from the database. In operation 550, the word embedding device extracts the word ID of the target word and / or the context word.

[0068] In operation 560, word embedding recognition determines training data based on any one or any combination of component information, structural information, physical property information, and word IDs of target words and / or context words. The component information, structural information, and word IDs of target words may be determined as training input data, and the physical property information and word IDs of context words may be determined as training output data. In operation 570, the word embedding device trains a word embedding model based on the determined training data. In operation 580, the word embedding device determines whether there is a next sentence in the document. In one example, when there is a next sentence in the document, operation 510 may be performed on the next sentence. In another example, when there is no next sentence in the document, operation 590 may be performed. In operation 590, the word embedding device performs post-processing on the word embedding model. The following will refer to Figure 6 Describes an example of post-processing.

[0069] Figure 6 Shows an example of performing post-processing on a word embedding matrix.

[0070] Figure 6 An m×n word embedding matrix 600 is shown. In the word embedding matrix 600, one axis represents "m" words and the other axis represents "n" dimensions of the vector. Each row of the word embedding matrix 600 represents an embedding vector for a word and may correspond to a word ID of the corresponding word. An n-dimensional embedding vector is determined for each word ID. Since the word ID of the word representing the chemical substance is not input to the word embedding model during training, the embedding vector is not input to the row corresponding to the word ID by only training the word embedding model. Figure 6 In the word embedding matrix 600, the white parts indicate rows without input embedding vectors.

[0071] Therefore, through post-processing, the embedding vectors of the chemical substance words obtained during training can be input to the portion corresponding to the chemical substance words in the word embedding matrix 600. By performing the above post-processing, the word embedding matrix 600 without a blank portion can be completed.

[0072] Figure 7 An example of a word embedding method is shown. Although the order of some operations may be changed or some operations may be omitted without departing from the spirit and scope of the illustrative examples described, Figure 7 The operations in can be performed in the order and manner as shown in the figure. Figure 7 Many of the operations shown in FIG. 1 may be performed in parallel or simultaneously. Figure 7 One or more blocks and combinations of blocks may be implemented by computers and devices (such as processors) based on dedicated hardware that perform the specified functions, or a combination of dedicated hardware and computer instructions. Figure 7 In addition to the description, Figures 1 to 6 The description also applies to Figure 7 Therefore, the above description may not be repeated here.

[0073] Figure 7 The word embedding method is executed by a processor such as a word embedding device.

[0074] In operation 710, a word embedding device trains a word embedding model based on characteristic information of a chemical substance. The word embedding model predicts context words of an input word.

[0075] The word embedding device trains a word embedding model based on any one or any combination of structural information, composition information, and physical property information of a chemical substance. In one example, the word embedding device may train the word embedding model so that, in response to inputting structural information of a chemical substance into the word embedding model, context words representing words of the chemical substance may be output from the word embedding model. In another example, the word embedding device may train the word embedding model so that, in response to inputting composition information of a chemical substance into the word embedding model, context words representing words of the chemical substance may be output from the word embedding model. In another example, the word embedding device may train the word embedding model so that physical property information of the chemical substance may be output from the word embedding model.

[0076] In operation 720 , the word embedding device obtains an embedding vector of a word representing a chemical substance from a word embedding model.

[0077] Furthermore, the word embedding device inputs the embedding vector into a portion corresponding to the word representing the chemical substance in the word embedding matrix corresponding to the word embedding model.

[0078] In one example, before operation 710 , the word embedding device may determine whether a word having an embedding vector to be generated represents a chemical substance.

[0079] Figure 8 An example of a word search method is shown. Although the order of some operations may be changed or some operations may be omitted without departing from the spirit and scope of the illustrative examples described, Figure 8 The operations in can be performed in the order and manner as shown in the figure. Figure 8 Many of the operations shown in FIG. 1 may be performed in parallel or simultaneously. Figure 8 One or more blocks and combinations of blocks may be implemented by computers and devices (such as processors) based on dedicated hardware that perform the specified functions, or a combination of dedicated hardware and computer instructions. Figure 8 In addition to the description, Figures 1 to 7 The description also applies to Figure 8 Therefore, the above description may not be repeated here.

[0080] Figure 8 The word search method is executed by, for example, a processor of a word search device.

[0081] In operation 810, the word search apparatus receives characteristic information of a chemical substance to be searched or a word representing a chemical substance. The characteristic information of the chemical substance may include any one or any combination of structural information, component information, and physical property information of the chemical substance.

[0082] In operation 820, the word search device outputs words representing substances having properties similar to those of the chemical substance based on a word embedding matrix. The word embedding matrix may be obtained from a word embedding model trained based on property information of multiple chemical substances. The word embedding model may predict context words for the input word.

[0083] Figure 9 An example of an electronic device 900 is shown.

[0084] Reference Figure 9 , the electronic device 900 includes a memory 910 , a processor 920 , and an input / output interface 930 . The memory 910 , the processor 920 , and the input / output interface 930 communicate with each other via a bus 940 .

[0085] In one example, the electronic device 900 may be, for example, a word embedding device or a word search device. In one example, the electronic device 900 may be implemented as various devices supporting word embedding or word search (e.g., a smart phone, a mobile phone, a wearable smart device (such as a wristband, a watch, a pair of glasses, a glasses-like device, a bracelet, an anklet, a belt, a necklace, an earring, a hairband, a helmet, a device embedded in clothing, or an eyeglass display (EGD)), a computing device (e.g., a server, a laptop computer, a notebook computer, a small notebook computer, a netbook, an ultra mobile PC (UMPC), a tablet personal computer (tablet), a tablet phone, a mobile Internet device (MID), a personal digital assistant (PAD), an enterprise digital assistant (EDA), a portable laptop PC), an electronic A product (e.g., a robot, a digital camera, a digital video camera, a portable game console, an MP3 player, a portable / personal multimedia player (PMP), a handheld e-book, a global positioning system (GPS) navigator, a personal navigation device, a portable navigation device (PND), a handheld game console, an e-book, a television (TV), a high-definition television (HDTV), a smart TV, a smart home appliance, a smart home device or a security device for door control, a walking assistance device, a smart speaker, a robot, various Internet of Things (IoT) devices), or a self-service machine), and can be executed by an application, middleware, or operating system installed on a user device, or a program of a server that interacts with the corresponding application.

[0086] The memory 910 includes computer-readable instructions. The processor 920 performs the above operations by executing the instructions stored in the memory 910. The memory 910 may include, for example, a volatile memory or a non-volatile memory. The memory 910 includes a large-capacity storage medium (such as a hard disk) to store various data. Further details about the memory 910 are provided below.

[0087] The processor 920 is, for example, a device configured to execute instructions or programs, or to control the electronic device 900. The processor 920 includes, for example, a central processing unit (CPU), a processor core, a multi-core processor, a reconfigurable processor, a multiprocessor, an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a graphics processing unit (GPU), or any other type of multi-processor or single processor configuration. The electronic device 900 is connected to an external device via the input / output interface 930 and exchanges data. Further details about the processor 920 are provided below.

[0088] In one example, the electronic device 900 interacts with the user through the input / output interface 930. In one example, the electronic device 900 displays context words of the input words, physical property information of chemical substances, structural information of chemical substances, and words representing chemical substances on the input / output interface 930.

[0089] In one example, the input / output interface 930 may be a display that receives input from a user or provides output. In one example, the input / output interface 930 may be used as an input device and receive input from a user through traditional input methods (e.g., keyboard and mouse) and new input methods (e.g., touch input, voice input, and image input). Therefore, the input / output interface 930 may include, for example, a keyboard, a mouse, a touch screen, a microphone, and other devices that can detect input from a user and send the detected input to the data processing device.

[0090] In one example, the input / output interface 930 can function as an output device and provide the user with output of the electronic device 900 through visual, auditory, or tactile channels. The input / output interface 930 may include, for example, a display, a touch screen, a speaker, a vibration generator, and other devices that can provide output to the user.

[0091] However, the input / output interface 930 is not limited to the examples described above, and any other display operably connected to the electronic device 900, such as a computer monitor and an eyewear display (EGD), may be used without departing from the spirit and scope of the described illustrative examples. In one example, the input / output interface 930 is a physical structure including one or more hardware components that provide the ability to render a user interface, render a display, and / or receive user input.

[0092] For example, the chemical substances "LiMn2O2" and "LiNi 0.5 Mn 1.5 O4" both contain lithium ions and have similar structures. The "LiMn2O2" and "LiNi 0.5 Mn 1.5 The embedding vector of "LiNi O4" can reflect similar characteristics, so the embedding vector can be used to easily retrieve "LiNi 0.5 Mn 1.5 O4" as a chemical substance similar to "LiMn2O2". Therefore, word embedding devices and word search devices can be used for knowledge base construction in the field of materials, reasoning through text analysis, and knowledge discovery.

[0093] The electronic device 900 may process one or more of the operations described above.

[0094] Word embedding device, word level word segmenter 410, chemical entity identifier 420, dictionary level word segmenter 430, word embedding model 440, model post-processor 460 for post-modifying the word embedding model 440 of training, and other devices, units, modules, devices and other components are implemented by hardware components. Examples of hardware components that can be used to perform the operations described in this application include, where appropriate, controllers, sensors, generators, drivers, memories, comparators, arithmetic logic units, adders, subtractors, multipliers, dividers, integrators, and any other electronic components configured to perform the operations described in this application. In other examples, one or more of the hardware components performing the operations described in this application are implemented by computing hardware (e.g., by one or more processors or computers). A processor or computer can be implemented by one or more processing elements (such as logic gate arrays, controllers and arithmetic logic units, digital signal processors, microcomputers, programmable logic controllers, field programmable gate arrays, programmable logic arrays, microprocessors, or any other device or combination of devices configured to respond and execute instructions in a limited manner to achieve the desired result). In one example, a processor or computer includes or is connected to one or more memories storing instructions or software executed by a processor or computer. The hardware components implemented by a processor or computer can execute instructions or software (such as an operating system (OS) and one or more software applications running on the OS) for performing the operations described in this application. The hardware components can also access, manipulate, process, create and store data in response to the execution of instructions or software. For simplicity, the singular term "processor" or "computer" can be used in the description of the examples described in this application, but in other examples, multiple processors or computers can be used, or the processor or computer can include multiple processing elements or multiple types of processing elements or both. For example, a single hardware component or two or more hardware components can be implemented by a single processor, or two or more processors, or a processor and a controller. One or more hardware components can be implemented by one or more processors, or a processor and a controller, and one or more other hardware components can be implemented by one or more other processors, or another processor and another controller. One or more processors, or a processor and a controller can implement a single hardware component or two or more hardware components. The hardware components may have any one or more of different processing configurations, examples of which include: a single processor, independent processors, parallel processors, single instruction single data (SISD) multiprocessing, single instruction multiple data (SIMD) multiprocessing, multiple instruction single data (MISD) multiprocessing, and multiple instruction multiple data (MIMD) multiprocessing.

[0095] The method for performing the operations described in this application is performed by computing hardware (e.g., by one or more processors or computers), which is implemented as instructions or software as described above to perform the operations performed by the method described in this application. For example, a single operation or two or more operations may be performed by a single processor, or two or more processors, or a processor and a controller. One or more operations may be performed by one or more processors, or a processor and a controller, and one or more other operations may be performed by one or more other processors, or another processor and another controller. One or more processors, or a processor and a controller may perform a single operation or two or more operations.

[0096] The instructions or software for controlling a processor or computer to implement the hardware components and perform the methods described above are written as computer programs, code segments, instructions, or any combination thereof to individually or collectively instruct or configure the processor or computer to operate as a machine or special-purpose computer to perform the operations performed by the hardware components and methods described above. In one example, the instructions or software include at least one of an applet, a dynamic link library (DLL), middleware, firmware, a device driver, and an application that stores the word embedding method. In one example, the instructions or software include machine code (such as machine code generated by a compiler) that is directly executed by the processor or computer. In another example, the instructions or software include high-level code that is executed by the processor or computer using an interpreter. Ordinary programmers in this field can easily write instructions or software based on the block diagrams and flow charts shown in the accompanying drawings and the corresponding descriptions in the specification, which disclose algorithms for performing the operations performed by the hardware components and methods described above.

[0097] Instructions or software for controlling computing hardware (e.g., one or more processors or computers) to implement hardware components and perform the methods described above, and any associated data, data files, and data structures, may be recorded, stored, or fixed in or on one or more non-transitory computer-readable storage media. Examples of non-transitory computer-readable storage media include read-only memory (ROM), random-access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random-access memory (RAM), dynamic random-access memory (DRAM), static random-access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-RLT H, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk, and any other device, wherein any other device is configured to store instructions or software and any associated data, data files and data structures in a non-transitory manner and provide the instructions or software and any associated data, data files and data structures to a processor or computer so that the processor and computer can execute the instructions. In one example, the instructions or software and any associated data, data files and data structures are distributed on a networked computer system so that the instructions and software and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0098] Although the present disclosure includes specific examples, it will be clear after understanding the disclosure of the present application that various changes in form and detail can be made in these examples without departing from the spirit and scope of the claims and their equivalents. The examples described herein are to be considered merely descriptive and not for purposes of limitation. The description of features or aspects in each example should be considered applicable to similar features or aspects in other examples. Suitable results can be achieved if the described techniques are performed in a different order, and / or if the components in the described systems, architectures, devices, or circuits are combined in different ways, and / or replaced or supplemented by other components or their equivalents. Therefore, the scope of the disclosure is not limited by the specific embodiments, but by the claims and their equivalents, and all changes within the scope of the claims and their equivalents should be interpreted as included in the disclosure.

Claims

1. A word embedding method, comprising: Train word embedding models based on the characteristic information of chemical substances; and Get the embedding vectors of words representing chemical substances from the word embedding model, Among them, the word embedding model is configured to predict the context words of the input word, The step of training the word embedding model includes: training the word embedding model based on the composition information of the chemical substance, so that in response to the composition information of the chemical substance being input into the word embedding model, context words of the words representing the chemical substance are output from the word embedding model, Here, words representing chemical substances are divided into letters or elements, and the letters or elements are sequentially input into a word embedding model.

2. The word embedding method according to claim 1, wherein The step of training the word embedding model includes: training the word embedding model based on any one or any combination of structural information and physical property information of the chemical substance.

3. The word embedding method according to claim 2, wherein: The step of training the word embedding model includes: training the word embedding model so that, in response to structural information of the chemical substance being input into the word embedding model, context words of the words representing the chemical substance are output from the word embedding model.

4. The word embedding method according to claim 3, wherein: Structural information of the chemical substance is determined based on a format selected from among a fingerprint, a simplified molecular linear input specification, a graph, and an image.

5. The word embedding method according to claim 1, wherein: The component information of the chemical substance is acquired from the word representing the chemical substance.

6. The word embedding method according to claim 1 or 2, wherein: The word embedding model is trained to also output information about the physical properties of the chemical substances.

7. The word embedding method according to claim 6, wherein: Physical property information includes information about any one or any combination of mass, volume, color, melting point, and boiling point of a chemical substance.

8. The word embedding method according to claim 1, further comprising: The embedding vector is input into the portion corresponding to the word representing the chemical substance in the word embedding matrix corresponding to the word embedding model.

9. The word embedding method according to claim 1, further comprising: Determine whether the word with the embedding vector to be generated represents a chemical substance.

10. A non-transitory computer-readable storage medium storing instructions, wherein when the instructions are executed by a processor, the processor performs the word embedding method according to any one of claims 1 to 9.

11. A word search method comprising: receiving characteristic information of a chemical substance to be searched or a word representing the chemical substance; and Output words representing substances having properties similar to those of the chemical substance based on the word embedding matrix, wherein, The word embedding matrix is ​​obtained from a word embedding model trained based on the characteristic information of multiple chemical substances, and The word embedding model is configured to predict the context words of the input word, wherein the characteristic information of the chemical substance includes composition information of the chemical substance, and the word embedding model is trained so that in response to the composition information of the chemical substance being input into the word embedding model, a context word representing a word of the chemical substance is output from the word embedding model, Here, the words representing the chemical substances are divided into letters or elements, and the letters or elements are sequentially input into the word embedding model.

12. The word search method according to claim 11, wherein: The characteristic information of the chemical substance also includes any one or any combination of structural information and physical property information of the chemical substance.

13. A word embedding device, comprising: The processor is configured to: Train word embedding models based on the characteristic information of chemical substances; and Get embedding vectors for words representing chemical substances from a word embedding model, and word embedding models, configured to predict context words for an input word, The processor is configured to: train a word embedding model based on the composition information of the chemical substance, so that in response to the composition information of the chemical substance being input into the word embedding model, a context word representing the word of the chemical substance is output from the word embedding model, In this method, words representing chemical substances are divided into letters or elements, and the letters or elements are sequentially input into the word embedding model.

14. The word embedding device according to claim 13, wherein: The processor is configured to: further train a word embedding model based on any one or any combination of structural information and physical property information of the chemical substance.

15. The word embedding device according to claim 14, wherein: The processor is configured to: train the word embedding model so that in response to structural information of the chemical substance being input into the word embedding model, context words of the words representing the chemical substance are output from the word embedding model.

16. The word embedding device according to claim 13 or 14, wherein: The word embedding model is trained to also output information about the physical properties of the chemical substances.

17. The word embedding device according to claim 13, wherein: The processor is further configured to: input the embedding vector into a portion corresponding to the word representing the chemical substance in the word embedding matrix corresponding to the word embedding model.

Images