A method and device for constructing a human-like solution engine

By building a human-like solution engine, combining and configuring state transition diagrams of various solution algorithms, problems with limited answer ranges in the existing technology are solved, more flexible and intelligent algebraic questions are realized, and answerable questions are expanded. scope.

CN118690005BActive Publication Date: 2024-12-03HUAZHONG NORMAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411191612.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-12-03
Estimated Expiration
2044-08-28

AI Technical Summary

Technical Problem

The existing machine-solution algorithms have limited questions and cannot effectively answer different types of algebraic questions, and existing algorithms often find it difficult to solve multiple types of questions in a unified manner.

Method used

By building a human-like solution engine, obtaining and combining state transfer diagrams of various solution algorithms, forming an integrated state transfer diagram, and functionally configuring it, imitating the human thinking process to gradually analyze and solve algebraic problems, and implementing multiple state transfer paths integration and optimization.

Benefits of technology

Expand the scope of answers, improve the performance and efficiency of answers, make the answer engine more flexible and intelligent, and be able to handle complex and changeable types of questions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118690005B_ABST
    Figure CN118690005B_ABST
Patent Text Reader

Abstract

The present invention provides a method and device for constructing a human-like solution engine, belonging to the field of artificial intelligence technology, the method comprising: obtaining a state transition diagram of each solution algorithm; merging and integrating the input nodes, output nodes, the same key state nodes, and the same state transition paths in all state transition diagrams to form an integrated state transition diagram; and configuring the integrated state transition diagram to construct a solution technology static diagram as a human-like solution engine. The method and device for constructing a human-like solution engine provided by the present invention imitate the human thinking process, and gradually parse and solve algebraic problems through a series of state transitions, so that the solution engine is more flexible and intelligent, can handle complex and changeable types of problems, and expand the scope of problems that can be solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for constructing a human-like answering engine. Background Art

[0002] Machine answering is a long-term active frontier research topic in the field of artificial intelligence. In recent years, driven by the joint efforts of technological progress in related fields and the demand for intelligent education, machine answering of algebraic questions has formed a new research peak. Algebraic questions are questions involving algebraic relationship calculations, including physical algebra problems, chemical algebra problems, arithmetic word problems, algebraic problems in mathematics, algebraic problems with graphics, plane geometry calculation problems, plane geometry calculation algebraic problems, function-based algebraic problems, etc. After more than 60 years of development, machine answering has produced rich research results. In terms of machine answering theory, the state transition theory has been developed, which can not only explain the existing answering algorithms, but also provide thinking guidance for the design of new answering algorithms. In terms of machine answering algorithms, more than 100 answering algorithms for algebraic questions have been developed. In order to introduce their main technologies, the state transition theory of machine answering can be used to divide these more than 100 algorithms into five main methods of answering state representation, namely, the answering method based on the double-box model, the answering method based on equations, the answering method based on expressions, the answering method based on Seq2X, and the answering method based on relationship groups. The following is a brief description of the theory and solution algorithms developed by machines for solving algebraic problems.

[0003] 1. State transition theory for machines solving algebraic problems

[0004] The state transition theory of machine solving algebraic problems models the process of solving algebraic problems as a cyclic representation process of transforming from one state to another state through the transition technology. The "state" is an abstract representation of the intermediate result state of the algorithm for solving algebraic problems, and the "transfer technology" is the method of realizing the transition from one state to another. At the level of algorithm program implementation, the state is an abstract data structure used to store a set of intermediate results in a consistent manner, and the transfer technology is a set of execution programs with standard interfaces. According to the state transition theory, the algorithm for machine solving algebraic problems can be represented as a "state transition diagram". The state transition diagram of machine solution is a graphical representation of the internal structure of the existing machine solution algorithm, which not only helps to analyze the existing machine solution methods and algorithms, but also helps to analyze the next evolution direction of machine solution methods and algorithms.

[0005] 2. Existing solution algorithms can be summarized into five categories:

[0006] (1) Solution method based on the dual-frame model. The core idea of ​​the solution method based on the dual-frame model is to use the knowledge representation framework to store the results of question understanding, and use the strategy framework as the driving force of knowledge reasoning to act on the knowledge representation framework, thereby generating the understanding process of the question and the solution results. According to the state transition theory, the state transition diagram of the solution method based on the dual-frame model generally uses two types of frameworks to construct the solution model for arithmetic text questions. These two types of frameworks are the knowledge representation framework and the reasoning strategy framework. Therefore, there are two main state nodes in the solution method based on the dual-frame model, namely the state node of the knowledge representation framework and the state node of the reasoning strategy framework. Combined with the problem input state node and the solution output state node, there are three main transfer technologies in the solution method based on the dual-frame model: the first transfer technology is to use semantic template matching to fill the knowledge contained in the problem text into the knowledge representation framework; the second transfer technology is to use the problem type matching to select the reasoning strategy framework of the problem; the third transfer technology is to use the reasoning strategy framework to reason and solve the problem to obtain the solution.

[0007] (2) Equation-based solution method. The core idea of ​​the equation-based solution method is to use equations as the equivalent representation state of the question, and to form a system of equations based on the equations, and to solve the answer to the question by solving the system of equations. Since solving the system of equations can rely on existing mature mathematical tools, the main stage of solving the problem in the equation-centered solution method is to understand the problem, that is, how to obtain the equations that represent the problem equivalently and form a system of equations. The system of equations used in the equation-centered method is a new state of problem equivalent representation proposed after the double-frame model. Compared with the double-frame model method, the problem equivalent state of the equation-centered method has been simplified, and its process of obtaining the system of equations reduces the reliance on natural language understanding of the problem text. According to the state transition theory, there are five states in the state transition diagram of the equation-based solution method, of which there are three main states, namely, feature text state node, equation template state node and equation group state node. Therefore, there are four transfer techniques in this type of solution algorithm: the first transfer technique is to use feature extraction method to obtain the feature text of the question; the second transfer technique is to calculate and match the feature text to the equation template; the third transfer technique is to use parameter mapping to transform the equation template into a system of equations; the fourth transfer technique is to solve the system of equations to obtain the answer to the question.

[0008] (3) Expression-based solution method. The expression-based solution method converts the question text into an answer expression composed of numbers and operators for solution. According to the state transition theory, there are three main states in the state transition diagram of the expression-based solution method: quantity operation object, expression tree and expression. Therefore, there are four transition techniques in this state transition diagram: the first transition technique is to classify and identify the quantity in the question text to obtain the quantity operation object of parameter operation; the second transfer technique is to use rules to infer the quantity operation object to obtain the expression tree corresponding to the question text; the third transfer technique is to convert the expression tree into an expression; the fourth transfer technique is to solve the expression to obtain the answer to the question.

[0009] (4) Seq2X-based solution method. The Seq2X-based solution method (Sequence-to-X, where X can be Seq, Tree, and DAG, corresponding to sequence, tree, and directed acyclic graph, respectively) uses the trained Seq2X model to transfer the question input state into the answer expression for calculation and solution. The underlying assumption of the Seq2X solution method is that a question can be equivalent to an answer expression for representation. This method draws on the "encoding-decoding" paradigm in the field of natural language processing, and its process is mainly to calculate vectors. According to the state transition theory, there are five state nodes in the state transition diagram of the Seq2X-based solution algorithm. In addition to the input question text state node and the output question answer state node, there are three key state nodes: text vector sequence state node, expression vector sequence state node, and expression state node. Therefore, there are four transfer techniques in this type of solution algorithm to connect the five state nodes: the first transfer technique is to encode the given question text into a text vector sequence; the second transfer technique is to convert the vector sequence of the question text into an expression latent vector; the third transfer technique is to convert the expression latent vector into an answer expression; the fourth transfer technique is to calculate and solve the answer expression to obtain the answer to the question. The state transition diagram of the Seq2X-based solution method shows that although the Seq2X-based solution algorithm has integrated many new deep learning technologies and methods in subsequent research, it mainly introduces new technical methods for the actions in the state transition diagram to construct a new solution algorithm.

[0010] (5) Solution method based on relation groups. The solution method based on relation groups converts the input state of the question into a relation group and then into a system of equations for solution. The underlying assumption of this method is that an arithmetic text problem can be equivalently expressed as a system of equations for solution. The core of the solution method based on relation groups is to use relations as the granularity for understanding the question, rather than isolated numbers. The main process of understanding the question is to obtain the relation group required for solving the question. There is a common state in the algorithm of the solution method based on relation groups, namely the relation group. Therefore, there are three main states in the state transition diagram of the solution method based on relation groups: text annotation sequence, relation group and equation group. The state output by this method is the solution to the question. There are four key transition technologies in the state transition diagram of the solution method based on relation groups: the first transition technology is to perform part-of-speech tagging on the question text to obtain a text annotation sequence; the second transition technology is to extract relations from the annotation sequence of the question text to obtain a relation group; the third transition technology is to convert the relation group into an equation group; and the fourth transition technology is to solve the equation group to obtain the solution to the question. The solution method based on relation groups subsequently introduced vector calculation, which resulted in vectorized representation states and transition technologies.

[0011] From the above review of the theories and algorithms developed by machines to solve algebraic problems, we can see that the research on machine solving algebraic problems has achieved rich results, but the relevant research results have not yet met the requirements of intelligent education applications, and there is still a lot of room for development. Although more than 100 algorithms have been proposed in the field of machine solving algebraic problems, a common situation is that the solution algorithms developed in a paper often hope to use one algorithm to uniformly solve different types of problems. This solution method has been proven to be unfeasible, and the range of problems that the algorithm can solve is very limited. Summary of the invention

[0012] The present invention provides a method and device for constructing a human-like answering engine, so as to solve the defect that the scope of questions that can be answered in the prior art is limited.

[0013] In a first aspect, the present invention provides a method for constructing a human-like solution engine, comprising: obtaining a state transition diagram for each solution algorithm; the state transition diagram comprises input nodes, key state nodes, output nodes, and state transition paths between nodes, the input nodes correspond to the input states of the solution algorithm, the key state nodes correspond to the key states of the solution algorithm, the output nodes correspond to the output states of the solution algorithm, and the state transition paths correspond to the transition relationships between the states; all the input nodes, output nodes, and the same key state nodes and the same state transition paths in the state transition diagram are merged and integrated to form an integrated state transition diagram; the integrated The state transition diagram is functionally configured to construct a static diagram of the solution technology as a human-like solution engine; wherein the method for functionally configuring the integrated state transition diagram includes: configuring the input node in the integrated state transition diagram to receive algebraic questions and perform content analysis on the algebraic question type, and determine the next state transition path for state transfer; configuring the key state node in the integrated state transition diagram to determine the next state transition path for state transfer according to the state content generated by the state transfer received by the current node; configuring the output node in the integrated state transition diagram to output the solution content of the algebraic question according to the state content received from the state transfer node input.

[0014] According to a method for building a human-simulated answering engine provided by the present invention, for input nodes and key state nodes, the next state transition path is determined for state transition, including: when a node has multiple state transition paths, at least one target state transition path is determined from the multiple state transition paths according to the currently received state content for state transition; when a node has only one state transition path, the only state transition path is used for state transition.

[0015] According to a method for constructing a human-simulated solution engine provided by the present invention, when the key state node receives state contents transferred by multiple state transfer paths, data integration processing is performed on all state contents.

[0016] According to a method for building a human-like answering engine provided by the present invention, when a node has multiple state transition paths, the node is called a divergent transition node; according to the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transition, including: inputting the currently received state content into a pre-built first decision model of the current divergent transition node, and outputting a predicted path label to perform state transition; wherein the step of building the first decision model of any divergent transition node includes: obtaining a decision data set about the state transition path of any divergent transition node; the decision data set includes the state content received by any divergent transition node, and the corresponding path label; wherein the path label includes the label of the next state transition path and the node; based on the decision data set, analyzing the rule-driven mapping relationship between the state content and the path label, and building the first decision model of any divergent transition node.

[0017] According to a method for building a human-like answering engine provided by the present invention, when a node has multiple state transition paths, the node is called a divergent transition node; based on the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transfer, including: inputting the currently received state content into the second decision model of the current divergent transition node that has been pre-trained, and outputting the predicted path label to perform state transfer; wherein, the step of building the second decision model of any divergent transition node includes: obtaining a decision data set about the state transition path of any divergent transition node; the decision data set includes the state content received by any divergent transition node, and the corresponding path label; wherein the path label includes the label of the next state transfer path and node; using the decision data set to train a preset machine learning model, and using the trained preset machine learning model as the second decision model.

[0018] A method for constructing a human-simulated solution engine provided by the present invention further includes: acquiring necessary knowledge from a preset knowledge base when necessary knowledge is lacking during the state transfer process.

[0019] A method for constructing a human-simulated solution engine provided by the present invention further includes: when a new solution algorithm appears, updating the configuration of the human-simulated solution engine.

[0020] According to a method for constructing a human-like answering engine provided by the present invention, the processing method for data integration processing of all status contents includes a combination of one or more of the following processing methods: information merging, content cleaning, and priority sorting.

[0021] In a second aspect, the present invention further provides a device for constructing a human-like solution engine, comprising:

[0022] A state transition diagram acquisition module is used to acquire a state transition diagram of each solution algorithm; the state transition diagram includes an input node, a key state node, an output node, and a state transition path between nodes, the input node corresponds to the input state of the solution algorithm, the key state node corresponds to the key state of the solution algorithm, the output node corresponds to the output state of the solution algorithm, and the state transition path corresponds to the transfer relationship between the states;

[0023] An integrated state transition diagram construction module is used to merge and integrate the input nodes, output nodes, the same key state nodes, and the same state transition paths in all state transition diagrams to form an integrated state transition diagram;

[0024] The human-like solution engine construction module is used to perform functional configuration on the integrated state transition diagram to construct a solution technology static diagram as a human-like solution engine; wherein the functional configuration of the integrated state transition diagram includes:

[0025] An input node in the integrated state transition diagram is configured to receive an algebraic question and parse the content of the algebraic question to determine the next state transition path for state transition; a key state node in the integrated state transition diagram is configured to determine the next state transition path for state transition according to the state content generated by the state transition received by the current node; and an output node in the integrated state transition diagram is configured to output the solution content of the algebraic question according to the state content received from the state transition node input.

[0026] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for constructing a human-simulated answering engine as described in any one of the above are implemented.

[0027] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described methods for constructing a human-simulated answering engine.

[0028] The method and device for constructing a human-like solution engine provided by the present invention can further improve the performance and efficiency of problem solving by integrating the state transition diagrams of multiple algorithms, constructing an integrated state transition diagram, and functionally configuring the integrated state transition diagram to generate a solution technology static diagram as a human-like solution engine; further, the present invention imitates the human thinking process, and gradually parses and solves algebraic problems through a series of state transitions, making the solution engine more flexible and intelligent, capable of handling complex and changeable problem types, and expanding the scope of problem that can be solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 It is a flowchart of a method for building a human-like answering engine provided by an embodiment of the present invention;

[0031] Figure 2 It is a schematic diagram of constructing a static diagram of a solution technology using two algorithm state transition diagrams provided by an embodiment of the present invention;

[0032] Figure 3 It is a static diagram of the solution technology provided by the embodiment of the present invention;

[0033] Figure 4 It is a schematic diagram of the process of solving algebraic problems using a human-like solving engine provided by an embodiment of the present invention;

[0034] Figure 5 It is a structural schematic diagram of a device for building a human-like answering engine provided by an embodiment of the present invention;

[0035] Figure 6 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] It should be noted that, in the description of the embodiments of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0038] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0039] Combine the following Figure 1-Figure 6 The method and device for constructing a human-like answering engine provided by an embodiment of the present invention are described.

[0040] Figure 1 FIG. 1 is a flow chart of a method for constructing a human-like answering engine provided by an embodiment of the present invention. Figure 1 As shown, including but not limited to the following steps:

[0041] Step 101: Obtain a state transition diagram of each solution algorithm.

[0042] The state transition diagram in the present invention may be pre-constructed, and the state transition diagram of each solution algorithm is constructed according to the input state, key state, output state and the transition relationship between states constructed based on the transition technology of each solution algorithm to characterize the solution process of each solution algorithm. The state transition diagram includes input nodes, key state nodes, output nodes and state transition paths between nodes, the input nodes correspond to the input states of the solution algorithm, the key state nodes correspond to the key states of the solution algorithm one-to-one, the output nodes correspond to the output states of the solution algorithm, and the state transition paths correspond to the transition relationships between states one-to-one.

[0043] Specifically, referring to the state transition theory introduced in the above background technology, the present invention models the process of solving algebraic problems as a cyclic representation process of transforming from one state to another state through a transition technology, wherein the "state" is an abstract representation of the intermediate result state of the algebraic problem solving algorithm, and the "transfer technology" is a method for realizing the transition from one state to another. At the level of algorithm program implementation, the state is an abstract data structure used to store a set of intermediate results in a consistent manner, and the transfer technology is a set of execution programs with a standard interface. According to the state transition theory, the algorithm for machine solving algebraic problems can be expressed as a "state transition diagram", and the state transition diagram of the machine solution is a graphical representation of the internal structure of the existing machine solution algorithm.

[0044] The state transition diagram in the present invention includes input nodes, key state nodes, output nodes and state transition paths between nodes.

[0045] Among them, the input node corresponds to the input state, indicating where the algorithm starts and usually receiving input data (for example, algebraic problems); the key state nodes correspond one-to-one to the key states, representing important milestones in the solution process; the output node corresponds to the output state, indicating where the algorithm ends and outputs the final solution result; the state transition path corresponds one-to-one to the transition relationship, which is the edge connecting different nodes, indicating the transition process from one state to another.

[0046] In the state transition diagram, new solution methods and solutions can be defined: the solution method describes the machine solution from an abstract level, which only considers the connection path from the question to the understanding state, and does not consider the technical methods on the connecting edges between nodes; the solution algorithm considers the specific path from the starting state to the end state, including the technical methods on the connecting edges between nodes.

[0047] Referring to the content introduced in the above background technology, in terms of machine solution algorithms, more than 100 solution algorithms for algebraic problems have been developed. In order to introduce its main technologies, the state transition theory of machine solution can be used to divide these more than 100 algorithms into five main solution state representation methods, namely, the solution method based on the double-box model, the solution method based on equations, the solution method based on expressions, the solution method based on Seq2X, and the solution method based on relationship groups. I will not go into details here.

[0048] The present invention analyzes the existing solution algorithms based on the state transition theory, thereby constructing a solution technology static diagram, which is further explained below.

[0049] Step 102: All input nodes, output nodes, identical key state nodes, and identical state transition paths in the state transition graph are merged and integrated to form an integrated state transition graph.

[0050] By merging the state transition graphs of all individual algorithms into an integrated state transition graph and merging repeated key state nodes and state transition paths, redundancy is reduced and efficiency is improved.

[0051] It is understandable that different key state nodes and state transition paths should be retained so that the algorithm paths of all state transition diagrams are retained in the integrated state transition diagram constructed by fusing all state transition diagrams.

[0052] Furthermore, it should be noted that the integrated state transition diagram formed by the present invention may also include other new solution paths, thereby increasing the scope of solutions to algebraic problems.

[0053] Step 103: Functionally configure the integrated state transition diagram to construct a static diagram of the solution technology as a human-like solution engine; wherein the method for functionally configuring the integrated state transition diagram includes: configuring an input node in the integrated state transition diagram to receive an algebraic question and perform content analysis on the algebraic question, and determine the next state transition path for state transfer; configuring a key state node in the integrated state transition diagram to determine the next state transition path for state transfer according to the state content generated by the state transfer received by the current node; configuring an output node in the integrated state transition diagram to output the solution content of the algebraic question according to the state content received as input by the state transfer node.

[0054] It should be noted that the next state transition path may include multiple state transition paths.

[0055] The process of functionally configuring the integrated state transition diagram is essentially designing and optimizing the algorithmic logic for the machine to solve algebraic problems. The specific steps include:

[0056] (1) Configure input nodes

[0057] The input node is the entry point of the entire solution process, responsible for receiving algebraic questions and performing preliminary content analysis. This step mainly involves:

[0058] Input format parsing: Identify and understand the structure and format of input algebraic problems, for example, identify types such as equations, inequalities, expressions, graphs, and their components such as variables, constants, operators, etc.

[0059] State initialization: Create an initial state based on the parsed question content. This state contains all the necessary information for the question and is used in the subsequent state transfer process.

[0060] (2) Configure key status nodes

[0061] Key state nodes represent important decision points in the algorithm. Each node corresponds to a specific state. At these nodes, the algorithm will decide the next operation based on the content of the current state. The configuration process includes:

[0062] Status assessment: Analyze the information of the current status to determine whether a certain condition is met or a certain goal is achieved.

[0063] Transition path selection: Based on the results of the state evaluation, the appropriate transition path is selected, that is, the state node to be transferred to next is determined. This may involve complex logic of the algorithm. For example, different mathematical rules or algorithms may need to be applied for specific types of algebraic problems.

[0064] (3) Configure output nodes

[0065] The output node is the end point of the algorithm and is used to output the final solution based on the received state content. The configuration process mainly includes:

[0066] Result generation: Generate the final answer based on the current state of the content. This may include operations such as calculation, simplification, and formatting.

[0067] Output formatting: Converting results into a user-friendly format, for example, converting mathematical expressions into an easily readable form.

[0068] (4) Other configuration strategies

[0069] Throughout the configuration process, the focus is on ensuring that the algorithm can correctly parse the questions, effectively transfer states, and generate accurate answers. In addition, the optimization configuration may also include:

[0070] Error handling: Add an exception handling mechanism to ensure that the algorithm can give reasonable feedback when encountering unsolvable questions or error states.

[0071] Extensible design: When a new solution algorithm appears, the present invention can configure and update the human-simulated solution engine according to the new solution algorithm, so that the algorithm can flexibly adapt to algebraic questions of different types and difficulties, and expand the function of the algorithm by adding new state nodes and transfer paths.

[0072] Through the above configuration, the solution technology static graph can become a powerful and flexible tool for automatically solving a wide range of algebraic problems. The present invention imitates the human thinking process and gradually parses and solves algebraic problems through a series of state transitions, making the solution engine more flexible and intelligent, capable of handling complex and changeable problem types, and expanding the range of solvable problems.

[0073] Based on the contents of the above embodiments, as an optional embodiment, the method for constructing a human-like answering engine provided by the present invention determines the next state transition path for state transition for input nodes and key state nodes, including: when a node has multiple state transition paths, at least one target state transition path is determined from multiple state transition paths according to the currently received state content for state transition; when a node has only one state transition path, the only state transition path is used for state transition. When an input node and a key state node have multiple state transition paths, the present invention may refer to the input node and the key state node as a divergent transition node. It can be understood that the divergent transition node in the present invention generally refers to an input node or a key state node having multiple state transition paths.

[0074] The core of the method of building a human-like problem-solving engine is to imitate the thinking process of humans in solving problems to achieve a gradual understanding and solution of the problem. This method particularly emphasizes the ability to dynamically select the transfer path based on the received state content at the divergent outgoing node, thereby achieving a more intelligent and flexible problem-solving strategy. Specifically, this method includes the following two key scenarios.

[0075] Multi-path selection scenario: When there are multiple possible state transition paths for the divergent outgoing node connection, the engine needs to be able to intelligently select at least one of the most appropriate target state transition paths based on the current state content received. This means that the engine must not only be able to understand the current state, but also be able to predict the possible results of different paths, so as to make the best choice. This ability is similar to the process of human decision-making based on experience and knowledge when facing complex problems.

[0076] Single path determination scenario: On the contrary, when there is only a single state transfer path for the divergent transition node, the engine does not need to make complex path selection, but directly transfers the state along the path. This usually happens when the solution path to the problem is relatively clear or there is only one reasonable operation at a certain stage.

[0077] Whether it is multi-path selection or single-path determination, it is the result of a deep understanding of the state content. The engine can navigate in the state space and find the best path from problem to solution by gradually analyzing the problem. This approach not only improves the efficiency of problem solving, but also enhances the adaptability and robustness of the engine, enabling it to cope with mathematical problems of various complexities and types, and even problems in other fields.

[0078] Optionally, the method for constructing a human-like answering engine provided by the present invention further includes: when the key state node receives state contents transferred by multiple state transition paths, data integration processing is performed on all state contents, and such a key state node may be referred to as a convergence node. It should be noted that a key state node may be both a convergence node and a divergent transfer-out node.

[0079] In the process of building a human-like problem-solving engine, in addition to the selection and determination of the state transfer path mentioned above, when the key state node receives the state content transferred from multiple state transfer paths, it is also a crucial step to effectively integrate the data of these state contents. This process aims to integrate the information from different paths to form a more comprehensive and deeper understanding to support subsequent decision-making and problem solving.

[0080] By integrating the status contents on different paths, the engine can obtain information of more dimensions, thereby making more accurate and comprehensive judgments, which helps to avoid wrong decisions caused by one-sided information.

[0081] Furthermore, the status contents on different paths often reflect different aspects or perspectives of the problem. Integrating this information can help the engine understand the essence of the problem more deeply and improve the accuracy and depth of the answer.

[0082] Optionally, the processing method of the present invention for data integration processing of all status contents includes a combination of one or more of the following processing methods: information merging, content cleaning, and priority sorting.

[0083] Information merging refers to integrating the state content collected from multiple state transition paths into a unified whole. This process usually includes data fusion, which is to merge similar or related information, reduce duplication, and extract the most core content.

[0084] Content cleaning aims to improve data quality, conduct preliminary cleaning and preprocessing on the collected status content, remove irrelevant or redundant information, and ensure the quality and consistency of the data.

[0085] Prioritization is to determine which status content is more important or urgent for answering the current problem, and to sort the status content in order of priority.

[0086] Based on the contents of the above embodiments, as an optional embodiment, the method for constructing a human-like answering engine provided by the present invention, for a divergent transition node, determines at least one target state transition path from multiple transition paths according to the currently received state content to perform state transition, including: inputting the currently received state content into a pre-constructed first decision model of the current divergent transition node, and outputting a predicted path label to perform state transition.

[0087] Among them, the step of constructing the first decision model of any divergent transition node includes: obtaining a decision data set of any divergent transition node about the state transition path; the decision data set includes the state content received by any divergent transition node, and the corresponding path label; wherein the path label includes the label of the next state transition path and node; based on the decision data set, analyzing the rule-driven mapping relationship between the state content and the path label, and constructing the first decision model of any divergent transition node.

[0088] Specifically, when judging that the data type of the input state content is a data type with a clear structure and a recognizable pattern (for example, keyword text, syntactic semantic text), it is more advantageous to adopt a rule-driven first decision model.

[0089] Taking keyword text as an example, we identify keywords or phrases closely related to path selection from the decision data set, analyze the frequency and location of keyword occurrence and the correlation between them, and summarize the keyword combinations or patterns corresponding to specific path labels.

[0090] Taking syntactic semantic text as an example, analyze the syntactic structure in the state content, identify basic components such as subject, predicate, object, and part of speech change patterns, and understand the basic framework of the sentence; label the semantic roles of each component in the sentence to understand their role and meaning in the sentence; based on the semantic roles and syntactic structure, identify the semantic patterns that match the specific path labels.

[0091] Based on the above analysis, a series of decision rules are defined, which clearly indicate which path label should be selected for state transfer when the input meets certain conditions (i.e., based on rule-driven mapping relationships). For example, "if the input contains keywords A and B, and the syntactic structure conforms to pattern X, then path Y is selected."

[0092] When the first decision model receives new state content, it first performs keyword and syntactic semantic analysis to extract key information; matches the extracted information with predefined decision rules to find out whether there are rules that meet the conditions; if a matching rule is found, the corresponding path label is output according to the rule instructions to guide the state transfer.

[0093] Based on the contents of the above embodiments, as an optional embodiment, the method for constructing a human-like answering engine provided by the present invention, for a divergent transition node, determines at least one target state transition path from multiple transition paths for state transition according to the currently received state content, including: inputting the currently received state content into a pre-trained second decision model of the current divergent transition node, and outputting a predicted path label for state transition.

[0094] Among them, the step of constructing a second decision model for any divergent transition node includes: obtaining a decision data set of any divergent transition node regarding the state transition path; the decision data set includes the state content received by any divergent transition node, and the corresponding path label; wherein the path label includes the label of the next state transition path and node; using the decision data set to train a preset machine learning model, and using the trained preset machine learning model as the second decision model.

[0095] It can be understood that the machine learning model is the core of the whole mechanism, and the choice of model can be any one of supervised learning, such as support vector machine (SVM), random forest (Random Forest), deep neural network (DNN), etc., depending on the characteristics of the data and the complexity of the problem. In the case that the data type of the input state data is not suitable for the first decision model, the second decision model based on machine learning is used to determine the path label for path transfer.

[0096] Based on the contents of the above embodiments, as an optional embodiment, the method for constructing a human-simulated answering engine provided by the present invention further includes: when necessary knowledge is lacking during the state transfer process, obtaining necessary knowledge from a preset knowledge base.

[0097] The knowledge base is an important part of the human-like answering engine. It stores a large amount of organized and classified domain knowledge, factual information, concept definitions, formula principles, etc., covering a wide range of fields from basic science to advanced theory. The construction and maintenance of the knowledge base is an ongoing process that needs to be constantly updated and expanded to adapt to the emergence of new knowledge and changes in problem scenarios.

[0098] In the present invention, external knowledge refers to knowledge that is necessary for the algorithm to solve a problem but is not directly described in the problem. This knowledge needs to be introduced from an external knowledge base, including formulas, theorems, common sense knowledge, etc. For example, to solve a problem about the area of ​​a triangle, the area formula of the triangle needs to be obtained from an external knowledge base.

[0099] The current key state node identifies whether the current state node needs additional external knowledge to continue to advance based on the analysis of the input state content. This identification process can be analyzed based on a variety of technical means such as keywords, semantic understanding, and question type analysis. Once it is identified that external knowledge needs to be added, the current key state node will automatically start the external knowledge addition method associated with the current state node.

[0100] In order to further illustrate the technical solution of the present invention, it is described below in conjunction with more specific embodiments.

[0101] Figure 2 is a schematic diagram of a static diagram of a solution technique using two algorithm state transition diagrams provided by an embodiment of the present invention, referring to Figure 2 ,Establishing the static graph of the solution technology can be divided into the following steps.

[0102] Step 1: Based on the existing solution algorithm A, construct a state transition diagram A to characterize the solution process of the existing solution algorithm A; based on the existing solution algorithm B, construct a state transition diagram B to characterize the solution process of the existing solution algorithm B.

[0103] The state transition diagram A includes: input node S 00 , output node S os And the key state node S 05 , S 03 and S 02 , and the state transition path T 15 ,T 14 ,T 04 and T 05 .

[0104] The state transition diagram B includes: input node S 00 , output node S os And the key state node S 10 , S 03 and S 02 , and the state transition path T 16 ,T 17 ,T 04 and T 06 .

[0105] Among them, the existing solution algorithm A comes from the paper: Yu X, Lyu X, Peng R, et al. Solving arithmetic word problems by synergizing syntax-semantics extractor for explicit relations and neural network miner for implicit relations[J]. Complex&Intelligent Systems, 2022, 9(1):697–717.

[0106] The existing solution algorithm B comes from the paper: Huang Litian, Yu Xinguo, Niu Lei, Feng Zihan. Solving algebraic problems with geometry diagrams using syntax-semantics diagram understanding[J], Computers, Materials&Continua, 2023.

[0107] It should be noted that constructing a state transition diagram according to an existing solution algorithm belongs to the content of the state transition theory introduced above. The present invention can realize the construction of the state transition diagram of most existing solution algorithms by using the above theory.

[0108] The present invention focuses on constructing and configuring an integrated state transition diagram based on multiple state transition diagrams to serve as a human-like solution engine. The construction of the state transition diagram belongs to the prior art and will not be described in detail in this embodiment.

[0109] Step 2: Connect the input node S in the state transition graph A and state transition B 00 , output node S os And the same key state node S 02 and S 03 Merge them separately and integrate the same state transition paths to form an integrated state transition graph;

[0110] Step 3: Functionally configure the integrated state transition diagram, including:

[0111] First, configure the divergent outgoing node of the integrated state transition diagram. 00 , is S 00 Establish decision maker D 00 , D 00 According to the content analysis of algebraic questions, the next path label of the current state node is set to {S 05 , T 15}, for the graph content of algebraic problems, set the next path label of the current state node to {S 10 , T 16}; For the divergent transfer node S 02 , is S 02 Establish decision maker D 02 , D 02 According to the state content generated by the state transfer received by the current node, set the next path label to {S os , T 05} and {S os , T 06}; Configure the remaining key status nodes S 03 , S 05 , S 10 , because the next step of these nodes has only one state transfer path to transfer to the next node, the only state transfer path is used for state transfer;

[0112] Second, configure the sink node S in the integrated state transition graph 03 , setting the data integration processing of the relational content obtained from the text and the relational content obtained from the graphic;

[0113] Third, configure the output node S in the integrated state transition graph os , according to the state content of the state transfer node input received, output the solution content of the algebraic problem.

[0114] The above solution technology static diagram indicates a simple example. In fact, the number of state transition diagrams contained in the solution technology static diagram is often greater than two. Figure 3 It is a static diagram of the solution technology provided by the embodiment of the present invention. Its construction process will not be described in detail here, but its structure will be briefly described.

[0115] S 00 represents an input node, which is used to receive and parse the algebraic problem, determine the next state transition path, and perform state transition; S 01 To S 08 Represents a key state node, which is used to determine the next state transfer path for state transfer based on the state content generated by the state transfer received by the current node.

[0116] Specifically, D 00 To D 08 Indicates that S 01 To S 08 Corresponding to the decision maker, the present invention executes the above-mentioned decision step of "determining the next state transition path to perform state transition" by setting the decision maker.

[0117] S os Represents an output node, which is used to output the solution to the algebraic problem based on the state content input by the received state transfer node.

[0118] T 01 , T 02 , T 03 , ..., T 14 , representing 14 state transition technologies (state transition paths) in the present invention.

[0119] The following is an explanation using the solution process of a specific algebraic problem as an example. Figure 4 1 is a schematic diagram of a process of solving algebraic problems using a human-simulated solution engine provided by an embodiment of the present invention. After the problem is input into the human-simulated solution engine, the human-simulated solution engine autonomously performs the human-simulated solution process as follows:

[0120] Input the algebraic word problem content into the input node S of the static graph of the solution technique 00 , S 00 Receive the content of the question and input the received content into the decision maker D 00 , decision maker D 00 The decision-making process is performed through the decision model. Since the input question meets the typical characteristics of containing direct and implicit relations, the decision maker D 00 Calculate that the current node should be transferred to the vector text state, then output the next path label of the current node as {S04 , T 07}, then execute the transfer technology T 07 S 00 Transition to the next key state node S 04 ;

[0121] S 04 It is not a sink node and does not need to perform content aggregation (i.e., when a key state node receives state content transferred by multiple state transfer paths, it performs data integration on all state content). 04 To diverge out of the node, the received content is input into the decision maker D 04 , decision maker D 04 The decision-making process is executed through the decision model, and the next path label is output as {S 05 , T 12} and {S 06 , T 13}, then execute the transfer technology T 12 S 04 Transition to the next state node S 05 , perform transfer technology T 13 S 04 Transition to the next state node S 06 In this embodiment, the transfer technology T 13 During the execution process, the labels of the relationship between the distance formula and the cylinder surface area formula are identified. Then, the distance formula "L=vt" and the cylinder surface area formula "S=πdh" need to be added to S from the knowledge base. 06 ;

[0122] S 05 and S 06 are not aggregation nodes, and do not need to perform content aggregation. 05 and S 06 None of them are divergent transfer nodes, S 05 The next path is {S 03 , T 14}, then execute the transfer technology T 14 S 05 Transition to the next critical state node S 03 , S 06 The next path label is {S 03 , T 14}, then execute the transfer technology T 14 S 05 Transition to the next critical state node S 03 ;

[0123] S 03 It is a aggregation node and needs to perform content aggregation process. 03The received content is input into the content aggregator module. If the aggregator detects that it is a relational input, it will start the cleaning program and the merging program. The cleaning program deletes redundant relation groups according to the problem solving intention. The merging program merges relations of different forms into the same representation form and connects entities of the same type in the relation, thereby outputting a complete relation group to S. 03 .

[0124] S 03 Input the aggregated content into the decision maker D 03 , decision maker D 03 The decision model is used to execute the decision process and output the next path label of the current state node as {S 02 , T 04}, then execute the transfer technology T 04 S 03 Transition to the next state node S 02 ;

[0125] S 02 Input the received content into the decision maker D 02 , decision maker D 02 The decision model is used to execute the decision process and output the next path label of the current state node as {S os , T 06}, then execute the transfer technology T 06 S 02 Transition to the next state node S os .

[0126] Based on the above steps, we can solve algebraic problems.

[0127] Figure 5 is a schematic diagram of the structure of a device for constructing a human-like solution engine provided by an embodiment of the present invention, such as Figure 5 As shown, the device includes: a state transition diagram acquisition module 501, an integrated state transition diagram construction module 502 and a human-simulated solution engine construction module 503.

[0128] Among them, the state transition diagram acquisition module 501 is used to construct a state transition diagram of each solution algorithm according to the input state, key state, output state and the transfer relationship between states constructed based on the transfer technology of each solution algorithm, so as to characterize the solution process of each solution algorithm; the state transition diagram includes input nodes, key state nodes, output nodes and state transfer paths between nodes, the input nodes correspond to the input states, the key state nodes correspond to the key states one by one, the output nodes correspond to the output states, and the state transfer paths correspond to the transfer relationships one by one;

[0129] An integrated state transition diagram construction module 502 is used to merge and integrate the input nodes, output nodes, the same key state nodes, and the same state transition paths in all state transition diagrams to form an integrated state transition diagram;

[0130] The human-like solution engine construction module 503 is used to perform functional configuration on the integrated state transition diagram to construct a solution technology static diagram as a human-like solution engine; wherein the functional configuration of the integrated state transition diagram includes:

[0131] An input node in the integrated state transition diagram is configured to receive an algebraic question and parse the content of the algebraic question to determine the next state transition path for state transition; a key state node in the integrated state transition diagram is configured to determine the next state transition path for state transition according to the state content generated by the state transition received by the current node; and an output node in the integrated state transition diagram is configured to output the solution content of the algebraic question according to the state content received from the state transition node input.

[0132] It should be noted that the apparatus for constructing a human-simulated answering engine provided in the embodiment of the present invention can execute the method for constructing a human-simulated answering engine described in any of the above embodiments during specific operation, which will not be described in detail in this embodiment.

[0133] Figure 6 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 communicate with each other via the communication bus 540. The processor 610 may call the logic instructions in the memory 630 to execute the method for constructing a human-simulated solution engine.

[0134] On the other hand, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the method for constructing a human-simulated answering engine provided in the above-mentioned embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for building a human-like solution engine, It is characterized in that include: Obtain a state transition diagram for each solution algorithm; the state transition diagram includes an input node, a key state node, an output node, and a state transition path between nodes, the input node corresponds to the input state of the solution algorithm, the key state node corresponds to the key state of the solution algorithm, the output node corresponds to the output state of the solution algorithm, and the state transition path corresponds to the transition relationship between the states; The input nodes, output nodes, the same key state nodes, and the same state transition paths in all state transition diagrams are merged and integrated to form an integrated state transition diagram; Functionally configuring the integrated state transition diagram to construct a solution technology static diagram as a human-like solution engine; wherein the method for functionally configuring the integrated state transition diagram includes: An input node in the integrated state transition diagram is configured to receive an algebraic question and parse the content of the algebraic question to determine the next state transition path for state transition; a key state node in the integrated state transition diagram is configured to determine the next state transition path for state transition according to the state content generated by the state transition received by the current node; an output node in the integrated state transition diagram is configured to output the solution content of the algebraic question according to the state content received from the state transition node input; Among them, for the input node and the key state node, determining the next state transfer path to perform state transfer includes: In the case where the node has multiple state transition paths, determining at least one target state transition path from the multiple state transition paths according to the currently received state content to perform state transition; When a node has only one state transfer path, the only state transfer path is used for state transfer; In the case where a node has multiple state transition paths, the node is referred to as a divergent transition node; and according to the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transition, including: Input the currently received state content into the first decision model of the current divergent transfer node constructed in advance, and output the predicted path label to perform state transfer; The step of constructing a first decision model for any divergent transfer-out node includes: Obtaining a decision data set of any divergent outgoing node on the state transition path; the decision data set includes the state content received by any divergent outgoing node and the corresponding path label; wherein the path label includes the label of the next state transition path and node; Based on the decision data set, analyzing the rule-driven mapping relationship between the state content and the path label to construct a first decision model; In the case where a node has multiple state transition paths, the node is referred to as a divergent transition node; and according to the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transition, including: Input the currently received state content into the second decision model of the current divergent transfer node that has been pre-trained, and output the predicted path label to perform state transfer; The step of constructing a second decision model for any divergent transfer-out node includes: Obtaining a decision data set of any divergent outgoing node on the state transition path; the decision data set includes the state content received by any divergent outgoing node and the corresponding path label; wherein the path label includes the label of the next state transition path and node; Using the decision data set to train a preset machine learning model, and using the trained preset machine learning model as the second decision model; Also includes: When the key state node receives state contents transferred by multiple state transfer paths, data integration processing is performed on all state contents.

2. The method for constructing a human-like solution engine according to claim 1, It is characterized in that Also includes: In the case of lack of necessary knowledge during state transfer, necessary knowledge is obtained from a pre-set knowledge base.

3. The method for constructing a human-like solution engine according to claim 1, It is characterized in that Also includes: When a new solution algorithm appears, the configuration of the human-simulated solution engine is updated.

4. The method for constructing a human-like solution engine according to claim 1, It is characterized in that The processing method for data integration processing of all status contents includes one or more combinations of the following processing methods: information merging, content cleaning, and prioritization.

5. A device for constructing a human-like solution engine, It is characterized in that include: A state transition diagram acquisition module, used to obtain a state transition diagram of each solution algorithm; The state transition diagram includes input nodes, key state nodes, output nodes, and state transition paths between nodes, the input nodes correspond to the input states of the solution algorithm, the key state nodes correspond to the key states of the solution algorithm, the output nodes correspond to the output states of the solution algorithm, and the state transition paths correspond to the transition relationships between the states. An integrated state transition diagram construction module is used to merge and integrate the input nodes, output nodes, the same key state nodes, and the same state transition paths in all state transition diagrams to form an integrated state transition diagram; The human-like solution engine construction module is used to perform functional configuration on the integrated state transition diagram to construct a solution technology static diagram as a human-like solution engine; wherein the functional configuration of the integrated state transition diagram includes: An input node in the integrated state transition diagram is configured to receive an algebraic question and parse the content of the algebraic question to determine the next state transition path for state transition; a key state node in the integrated state transition diagram is configured to determine the next state transition path for state transition according to the state content generated by the state transition received by the current node; an output node in the integrated state transition diagram is configured to output the solution content of the algebraic question according to the state content received from the state transition node input; Among them, for the input node and the key state node, determining the next state transfer path to perform state transfer includes: In the case where the node has multiple state transition paths, determining at least one target state transition path from the multiple state transition paths according to the currently received state content to perform state transition; When a node has only one state transfer path, the only state transfer path is used for state transfer; In the case where a node has multiple state transition paths, the node is referred to as a divergent transition node; and according to the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transition, including: Input the currently received state content into the first decision model of the current divergent transfer node constructed in advance, and output the predicted path label to perform state transfer; The step of constructing a first decision model for any divergent transfer-out node includes: Obtaining a decision data set of any divergent outgoing node on the state transition path; the decision data set includes the state content received by any divergent outgoing node and the corresponding path label; wherein the path label includes the label of the next state transition path and node; Based on the decision data set, analyzing the rule-driven mapping relationship between the state content and the path label to construct a first decision model; In the case where a node has multiple state transition paths, the node is referred to as a divergent transition node; and according to the currently received state content, at least one target state transition path is determined from the multiple transition paths to perform state transition, including: Input the currently received state content into the second decision model of the current divergent transfer node that has been pre-trained, and output the predicted path label to perform state transfer; The step of constructing a second decision model for any divergent transfer-out node includes: Obtaining a decision data set of any divergent outgoing node on the state transition path; the decision data set includes the state content received by any divergent outgoing node and the corresponding path label; wherein the path label includes the label of the next state transition path and node; Using the decision data set to train a preset machine learning model, and using the trained preset machine learning model as the second decision model; The key state node is configured to perform data integration processing on all state contents when the key state node receives state contents transferred by multiple state transfer paths.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, It is characterized in that When the processor executes the computer program, the steps of the method for constructing a human-simulated solution engine according to any one of claims 1 to 4 are implemented.