Directed resource push method and system for Olympiad mathematics learning

Abstract

The embodiment of the present application provides a method and system for pushing directional resources applied to Olympiad mathematics learning. By obtaining the logical vector representing the target Olympiad mathematics learning questions, as well as multiple directional problem-solving thinking models and their corresponding problem-solving thinking guidance information, a deep understanding and multi-angle analysis of Olympiad mathematics questions are achieved. A problem-solving path feature vector containing the content of the answer to the question and the characteristics of the problem-solving thinking is generated according to the problem-solving thinking model, and then a problem-solving path result under a plurality of problem-solving thinking labels is generated, which effectively improves the diversity and flexibility of Olympiad mathematics problem-solving. At the same time, by obtaining the target problem-solving thinking label configured by the target user and pushing the corresponding problem-solving path result, personalized resource push for the user is achieved, which improves the efficiency of Olympiad mathematics learning and the user experience.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for pushing directional resources applied to Olympiad mathematics learning. Background Art

[0002] Mathematical Olympiad learning, as an important branch of mathematics education, has always attracted much attention due to the complexity of its questions and the diversity of its problem-solving ideas. Traditional Mathematical Olympiad learning mainly relies on paper textbooks and teachers' explanations, and students need to spend a lot of time and energy to understand and master various problem-solving ideas. However, this learning method has many limitations, such as inconvenient resource acquisition, single problem-solving ideas, and low learning efficiency.

[0003] With the rapid development of artificial intelligence technology, its application in the field of education is becoming increasingly widespread. In response to the difficulties in Olympiad learning, relevant Olympiad learning assistance systems have begun to try to use artificial intelligence technology to assist in problem solving and learning. However, most of the existing Olympiad learning assistance systems simply provide questions and answers, or analyze the questions to a certain extent, but often lack targeted guidance on problem-solving ideas and cannot meet the diverse learning needs of students.

[0004] In addition, the answers to Mathematical Olympiad questions often involve multiple solutions, and different students may have different preferences and understanding abilities for different solutions. Therefore, how to provide customized guidance on problem-solving ideas based on students' learning characteristics and needs has become an urgent problem to be solved in the Mathematical Olympiad learning assistance system. Summary of the invention

[0005] In order to at least overcome the above-mentioned deficiencies in the prior art, the purpose of the embodiments of the present application is to provide a method and system for pushing targeted resources for Olympiad mathematics learning.

[0006] According to one aspect of the present application, a method for directional resource push applied to Mathematical Olympiad learning is provided, the method comprising:

[0007] Acquire a question characterization logic vector of a target Mathematical Olympiad learning question, wherein the question characterization logic vector reflects the question comprehension logic of the target Mathematical Olympiad learning question;

[0008] Obtaining X directional problem-solving idea models and Y problem-solving idea guidance information corresponding to the X directional problem-solving idea models, wherein one piece of the problem-solving idea guidance information represents a problem-solving idea logic of a problem-solving idea label, and one directional problem-solving idea model corresponds to at least one piece of the problem-solving idea guidance information, X and Y are both positive integers, and X is not greater than Y;

[0009] According to the X directional problem-solving idea models, under the restriction of the problem-representation logic vector and the corresponding problem-solving idea guidance information, a problem-solving path feature vector corresponding to each problem-solving idea guidance information is generated, wherein one problem-solving path feature vector includes a solution content feature vector of the target Olympiad mathematics learning problem and a problem-solving idea feature vector of a problem-solving idea label guided by the corresponding problem-solving idea guidance information, wherein the solution content feature vector is generated by artificial intelligence problem-solving learning of the problem-understanding logic expressed by the problem-representation logic vector, and the problem-solving idea feature vector is generated by artificial intelligence problem-solving learning of the problem-solving idea logic of the problem-solving idea label guided by the corresponding problem-solving idea guidance information;

[0010] Generate problem-solving results for the problem-solving path feature vectors corresponding to the Y problem-solving idea guidance information, respectively, to obtain the problem-solving path results of the target Olympiad mathematics learning problem under the Y problem-solving idea labels;

[0011] The target problem-solving idea label configured by the target user is obtained, and after extracting the target problem-solving path result under the target problem-solving idea label from the problem-solving path results under Y problem-solving idea labels of the target Olympiad mathematics learning problem, the target problem-solving path result is pushed to the target user as a directional resource.

[0012] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in various optional implementations of the above three aspects.

[0013] In the technical solutions provided by some embodiments of the present application, the embodiments of the present application achieve in-depth understanding and multi-angle analysis of Olympiad mathematics questions by obtaining the logical vector representing the target Olympiad mathematics learning questions, as well as multiple directional problem-solving thinking models and their corresponding problem-solving thinking guidance information, and generate a problem-solving path feature vector containing the content of the answer to the question and the characteristics of the problem-solving thinking according to the problem-solving thinking model, and then generate the problem-solving path results under multiple problem-solving thinking labels, effectively improving the diversity and flexibility of Olympiad mathematics problem-solving. At the same time, by obtaining the target problem-solving thinking label configured by the target user and pushing the corresponding problem-solving path results, personalized resource push for users is achieved, which improves the efficiency of Olympiad mathematics learning and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A flow chart of a method for pushing targeted resources for Mathematical Olympiad learning provided in an embodiment of the present application;

[0015] Figure 2 A schematic block diagram of the architecture of a targeted resource push system for Olympiad mathematics learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] Figure 1 This is a flow chart of a method for pushing targeted resources applied to Mathematical Olympiad learning provided by an embodiment of the present application. The method for pushing targeted resources applied to Mathematical Olympiad learning is introduced in detail below.

[0017] Step S110, obtaining a question characterization logic vector of a target Olympiad mathematics learning question, wherein the question characterization logic vector reflects the question comprehension logic of the target Olympiad mathematics learning question.

[0018] In detail, the question representation logic vector is a vector that reflects the logic of understanding the target Mathematical Olympiad learning question. It is generated by segmenting the question text, semantically analyzing it, and constructing a set of logic units and a logic diagram, and finally mapping it to the vector space. This question representation logic vector completely contains various mathematical relationships, conditions, and solution goals in the question.

[0019] For example, in this embodiment, after receiving the target Mathematical Olympiad learning topic, the server starts to construct the process of logical vector representation of the topic. First, the topic text is processed using a pre-built special vocabulary in the field of Mathematical Olympiad. This vocabulary contains various mathematical terms, concept names and operation symbols in Mathematical Olympiad. For example, for a Mathematical Olympiad topic, the server will accurately segment the entire topic text according to the vocabulary in the vocabulary, and divide it into basic units, such as basic mathematical elements such as numbers and mathematical symbols. For each basic mathematical element, the server analyzes it according to the semantics in the Mathematical Olympiad knowledge system. For example, for a number representing a quantity, it may have different semantics under different Mathematical Olympiad concepts, such as in a sequence problem, it may represent the number of items or the value of a certain item in the sequence. The server constructs a set of logical units based on these semantic analysis results, and each logical unit is a logical structure of related basic mathematical elements combined according to a set logical relationship. For example, in a Mathematical Olympiad problem about a journey problem, the basic mathematical elements of speed, time and distance may form a logical unit to represent the relationship between the three. Then, the logical relationships of each logical unit in the logical unit set are sorted out to construct a logical diagram. In this logic diagram, nodes represent logic units, edges represent logical relationships between logic units, and there are start nodes (representing the initial condition logic units in the question) and end nodes (representing the logic units corresponding to the content that the question requires to be solved or proved). Next, the server defines a vector space, whose dimensions are determined based on the reference logic types in the Mathematical Olympiad knowledge system (such as numerical calculation logic dimensions, geometric relationship logic dimensions, logical reasoning logic dimensions, conditional judgment logic dimensions, etc.) and the characteristics of the question. Finally, each logic unit in the logic diagram is mapped to the corresponding dimension in the vector space, and the mapping results on each dimension are synthesized to generate the question representation logic vector of the target Mathematical Olympiad learning question. This vector fully reflects the logic of question understanding, including various mathematical relationships, conditions, and solution goals in the question.

[0020] Step S120, obtaining X directional problem-solving idea models and Y problem-solving idea guidance information corresponding to the X directional problem-solving idea models, wherein one piece of the problem-solving idea guidance information represents a problem-solving idea logic of a problem-solving idea label, and one directional problem-solving idea model corresponds to at least one piece of the problem-solving idea guidance information, and both X and Y are positive integers, and X is not greater than Y.

[0021] In detail, the directional problem-solving thinking model is a problem-solving model specially constructed for Olympiad mathematics learning questions. Each directional problem-solving thinking model has a unique problem-solving logic structure, which is used to process Olympiad mathematics questions of different types or with specific characteristics. These directional problem-solving thinking models are designed to help the server understand and answer Olympiad mathematics questions. The problem-solving thinking guidance information is information that characterizes the problem-solving thinking logic of a problem-solving thinking label. Each directional problem-solving thinking model corresponds to at least one problem-solving thinking guidance information, which is used to guide the server to solve the problem according to a specific problem-solving idea. This information helps the server clarify different problem-solving directions.

[0022] For example, in this embodiment, the server obtains pre-established directional problem-solving thinking models and problem-solving thinking guidance information from the storage system. These directional problem-solving thinking models and guidance information are specially constructed for Mathematical Olympiad learning questions. The server finds X directional problem-solving thinking models according to internal indexes or identifiers, and these models are designed to handle Mathematical Olympiad questions of different types or with specific characteristics. Each directional problem-solving thinking model has its unique problem-solving logical structure. At the same time, each directional problem-solving thinking model corresponds to at least one problem-solving thinking guidance information. For example, for a directional problem-solving thinking model about the summation of a sequence, there may be a problem-solving thinking guidance information that guides the problem-solving according to the thinking label of the summation formula of an arithmetic sequence, and another problem-solving thinking guidance information may be a thinking label for deriving the summation according to the general formula of the sequence. These problem-solving thinking guidance information characterizes the problem-solving thinking logic of different problem-solving thinking labels, helping the server to clarify different problem-solving directions. The values ​​of X and Y are determined according to the number of existing directional problem-solving idea models and guidance information in the server, and X is not greater than Y, which means that each directional problem-solving idea model may correspond to more than one guidance information to cover more different problem-solving idea situations.

[0023] Step S130, generating a problem-solving path feature vector corresponding to each problem-solving idea guidance information according to the X directional problem-solving idea models under the restriction of the problem-representation logic vector and the corresponding problem-solving idea guidance information, wherein one problem-solving path feature vector includes a solution content feature vector of the target Olympiad mathematics learning problem and a problem-solving idea feature vector of a problem-solving idea label guided by the corresponding problem-solving idea guidance information, wherein the solution content feature vector is generated by performing artificial intelligence problem-solving learning on the problem-understanding logic expressed by the problem-representation logic vector, and the problem-solving idea feature vector is generated by performing artificial intelligence problem-solving learning on the problem-solving idea logic of the problem-solving idea label guided by the corresponding problem-solving idea guidance information.

[0024] In detail, the problem-solving path feature vector is a vector that includes the solution content feature vector of the target Olympiad mathematics learning question and the problem-solving idea feature vector of the problem-solving idea label guided by the corresponding problem-solving idea guidance information. It is generated by a directional problem-solving idea model under the constraints of the question representation logical vector and the corresponding problem-solving idea guidance information, and reflects the specific content of the answer to the question and the logical structure of the problem-solving idea.

[0025] The solution content feature vector is a vector generated by artificial intelligence problem-solving learning of the problem understanding logic expressed by the problem representation logic vector, represents the specific solution content of the target Olympiad learning problem, and is generated through the problem feature deduction branch in the directional problem-solving thinking model. The solution idea feature vector is a vector generated by artificial intelligence problem-solving learning of the problem-solving idea logic of the problem-solving idea label guided by the corresponding problem-solving idea guidance information, reflects the problem-solving logic under a specific problem-solving idea label, and is generated through the idea feature deduction branch in the directional problem-solving thinking model.

[0026] For example, in this embodiment, for each directional problem-solving thinking model and its corresponding problem-solving thinking guidance information, the server operates based on the question characterization logic vector. Take one of the directional problem-solving thinking models (set as the target directional problem-solving thinking model) and its corresponding problem-solving thinking guidance information (set as the target problem-solving thinking guidance information) as an example. First, based on the target directional problem-solving thinking model, the question-limited feature vector of the target Olympiad learning question is generated under the restriction of the question characterization logic vector. In this process, the target directional problem-solving thinking model includes multiple problem-solving step networks (assuming Z) connected in sequence, and each problem-solving step network includes a question feature derivation branch and a thinking feature derivation branch connected in parallel. For the a-th problem-solving step network (1≤a≤Z), if a=1, its input problem-solving idea feature is the initial problem-solving idea feature of the target-oriented problem-solving idea model; if a>1, the input problem-solving idea feature is the derived feature vector of the a-1-th problem-solving step network (the derived feature vector is obtained by interactively fusing the feature vector output by the problem feature derivation branch in the a-1-th problem-solving step network and the feature vector output by the idea feature derivation branch). According to the problem feature derivation branch in the a-th problem-solving step network, under the restriction of the problem representation logic vector (this process is to interactively fuse the problem representation logic vector and the input problem-solving idea feature), the input problem-solving idea feature is derived and processed to generate the problem restriction sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem. Finally, the problem restriction sub-feature vector of the Z-th problem-solving step network for the target Olympiad learning problem is output as a problem restriction feature vector, which is used to reflect the solution content feature vector of the target Olympiad learning problem.

[0027] At the same time, based on the target-oriented problem-solving idea model, the problem-solving idea limiting feature vector of the target Olympiad learning problem is generated under the restriction of the target problem-solving idea guidance information. Similarly, for the a-th problem-solving step network, based on the idea feature derivation branch, under the restriction of the target problem-solving idea guidance information (this process is to interactively integrate the target problem-solving idea guidance information and the input problem-solving idea features), the input problem-solving idea features are deduced and processed to generate the initial problem-solving idea limiting feature vector of the a-th problem-solving step network for the target Olympiad learning problem. Then, the derivation adjustment unit in the idea feature derivation branch of the a-th problem-solving step network will adjust the initial problem-solving idea limiting feature vector (the derivation adjustment unit is in different states, such as state switching state or derivation granularity adjustment state, and there will be different processing methods), and generate the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem. Finally, the problem-solving idea limiting sub-feature vector of the Zth problem-solving step network for the target Olympiad learning problem is output as a problem-solving idea limiting feature vector. This problem-solving idea limiting feature vector is used to reflect the problem-solving idea feature vector of the problem-solving idea label guided by the target problem-solving idea guidance information.

[0028] Finally, the problem-limited feature vector and the problem-solving idea-limited feature vector are interactively fused to generate the problem-solving path feature vector corresponding to the target problem-solving idea guidance information. Assume that the problem-limited feature vector includes Q problem logic components on Q knowledge logic dimensions, and the problem-solving idea-limited feature vector includes Q idea logic components on Q knowledge logic dimensions. First, the logical components of the questions and logical components of the ideas on the same knowledge logic dimension are merged and operated respectively to generate the merged operation components of the target-oriented problem-solving idea model on each knowledge logic dimension; then, the logical components of the questions and logical components of the ideas on the same knowledge logic dimension are preliminarily connected to generate the connection components of the target-oriented problem-solving idea model on each knowledge logic dimension; finally, the merged operation components and connection components of the target-oriented problem-solving idea model on Q knowledge logic dimensions are fused (including performing advanced connection operations on the merged operation components and connection components on the same knowledge logic dimension to generate advanced connection components, using the Olympiad logic operation components containing Q logical operators to perform logical operations on the advanced connection components, and splicing the operation result features according to the Olympiad logic operation components) to generate the problem-solving path feature vector corresponding to the target problem-solving idea guidance information. This problem-solving path feature vector contains the solution content feature vector of the target Olympiad learning question and the problem-solving idea feature vector of the problem-solving idea label guided by the corresponding problem-solving idea guidance information.

[0029] Step S140, generating problem-solving results for the problem-solving path feature vectors corresponding to the Y problem-solving idea guidance information respectively, and obtaining the problem-solving path results of the target Olympiad mathematics learning problem under Y problem-solving idea labels.

[0030] In detail, the problem-solving path result is the problem-solving result obtained after generating the problem-solving result for the problem-solving path feature vector corresponding to each problem-solving idea guidance information, and represents the solution to the target Olympiad mathematics learning problem under a specific problem-solving idea label.

[0031] For example, in this embodiment, the server processes the problem-solving path feature vector corresponding to each problem-solving idea guidance information. For each problem-solving path feature vector, it is first split into a solution content feature vector and a problem-solving idea feature vector. For the solution content feature vector, based on the pre-constructed Olympiad knowledge classification system, a semantic analysis algorithm is used to identify different types of knowledge elements contained in the problem-solving path feature vector. For example, if it is an Olympiad problem about geometry, knowledge elements such as triangles, angles, and side lengths may be identified. For the problem-solving idea feature vector, the logical elements representing the sequence of problem-solving steps, key logical judgment points, and special problem-solving method identifiers are parsed out. Thus, the resolved solution content knowledge element set and the problem-solving idea logical element set are obtained.

[0032] Then, the knowledge domain involved in the target Olympiad learning questions is determined based on the knowledge element set of the answer content, and the basic concepts, theorems and formulas related to the knowledge domain are extracted from the Olympiad knowledge classification system to construct the corresponding preliminary knowledge framework. At the same time, the corresponding problem-solving process framework is determined based on the logical element set of the problem-solving idea, and the preliminary knowledge framework and the problem-solving process framework are integrated to form a directional problem-solving framework for each problem-solving path feature vector.

[0033] Under the guidance of the preliminary knowledge framework, for each element in the knowledge element set of the answer content, operate according to the order and logic in the problem-solving process framework. Each step of operation is based on the operation result of the previous step and the relevant elements in the problem-solving process framework, and the problem-solving process is gradually promoted to obtain the problem-solving operation process record corresponding to each problem-solving path feature vector. This record includes the operation content, operation basis, operation input and operation output of each step.

[0034] According to the last operation output in the problem-solving operation process record, combined with the preset question requirements of the target Olympiad learning questions, determine whether the problem-solving operation process record has obtained a complete problem-solving result. If it is a numerical solution question, check whether the numerical value of the problem-solving result meets all the conditions in the question; if it is a proof question, check whether the proof process of the problem-solving result is rigorous and complete. If it is detected that the problem-solving operation process record has obtained a complete problem-solving result, the preliminary problem-solving result of each problem-solving path feature vector is output; if it is detected that the problem-solving operation process record has not obtained a complete problem-solving result, the problem-solving operation process record is traced back to find the operation steps that may cause problems, and a preliminary result is generated to mark the problem point or error point. For the preliminary results of marking problem points or error points, adjustments are made on the basis of the problem-solving process framework according to the marked problem points or error points to generate the final problem-solving results corresponding to each problem-solving path feature vector after improvement and optimization. These final problem-solving results are the problem-solving path results of the target Olympiad learning questions under Y problem-solving idea labels.

[0035] Step S150, obtain the target problem-solving idea label configured by the target user, and extract the target problem-solving path result under the target problem-solving idea label from the problem-solving path results under Y problem-solving idea labels of the target Olympiad mathematics learning problem, and then push the target problem-solving path result to the target user as a targeted resource.

[0036] The target problem-solving idea label is a specific problem-solving idea label selected by the target user according to his / her own learning needs or problem-solving preferences. The server will extract the corresponding target problem-solving path result from the problem-solving path result according to this label. The target problem-solving path result is the problem-solving result of the target Olympiad learning problem obtained under the target problem-solving idea label. It is extracted from the problem-solving path result by the server according to the user's configuration information and pushed to the target user as a directional resource.

[0037] For example, in the present embodiment, the server receives configuration information sent by the target user, which includes a target problem-solving idea label. For example, the user may select a specific problem-solving idea label according to his or her own learning needs or problem-solving preferences, such as "sequence general formula method" (here is just an example of a type of problem-solving idea label). After receiving this target problem-solving idea label, the server searches for the corresponding target problem-solving path result from the problem-solving path results of the target Olympiad learning questions under Y kinds of problem-solving idea labels obtained before. After finding it, this target problem-solving path result is used as a directional resource and pushed to the target user through network communication and other methods. In this way, the user can get the solution result of the Olympiad problem under the problem-solving idea he or she wants, which helps the user to learn Olympiad mathematics and improve problem-solving ability.

[0038] Based on the above steps, the embodiment of the present application achieves a deep understanding and multi-angle analysis of the Olympiad mathematics questions by obtaining the logical vector representing the target Olympiad mathematics learning questions, as well as multiple directional problem-solving thinking models and their corresponding problem-solving thinking guidance information, and generates a problem-solving path feature vector containing the content of the question answer and the characteristics of the problem-solving thinking according to the problem-solving thinking model, and then generates the problem-solving path results under multiple problem-solving thinking labels, effectively improving the diversity and flexibility of Olympiad mathematics problem-solving. At the same time, by obtaining the target problem-solving thinking label configured by the target user and pushing the corresponding problem-solving path results, personalized resource push for users is achieved, which improves the efficiency of Olympiad mathematics learning and user experience.

[0039] In a possible implementation, any one of the X directional problem-solving thinking models is a target-oriented problem-solving thinking model, and any one of the problem-solving thinking guidance information corresponding to the target-oriented problem-solving thinking model is the target problem-solving thinking guidance information. Step S130 includes:

[0040] Step S131, generating a question-restricted feature vector of the target Olympiad mathematics learning question according to the target-oriented problem-solving thinking model under the restriction condition of the question representation logical vector, wherein the question-restricted feature vector is used to reflect the solution content feature vector of the target Olympiad mathematics learning question.

[0041] In this embodiment, the goal-oriented problem-solving thinking model may include Z problem-solving step networks (Z is a positive integer) connected in sequence for feature derivation. For example, assuming Z=3, these three problem-solving step networks can be understood as three processing links.

[0042] Each problem-solving step network includes a parallel connection of a question feature derivation branch and a thought feature derivation branch. The target-oriented problem-solving thought model has a first initial problem-solving thought feature to be optimized.

[0043] When processing the first problem-solving step network (a=1), its input problem-solving idea feature is the first initial problem-solving idea feature. When processing the second problem-solving step network (a=2), its input problem-solving idea feature is the derived feature vector of the first problem-solving step network. This derived feature vector is obtained by interactively fusing the feature vector output by the problem feature derivation branch in the first problem-solving step network and the feature vector output by the idea feature derivation branch.

[0044] Next, for each problem-solving step network’s question feature derivation branch, the input problem-solving idea features are derived under the constraints of the question representation logic vector. This process is actually an interactive fusion of the question representation logic vector and the input problem-solving idea features.

[0045] For example, the server receives the problem representation logic vector of the target Mathematical Olympiad learning problem, which contains various mathematical relationships, conditions, and solution goals in the problem. Assume that the problem representation logic vector represents a condition about a quantitative relationship in a certain dimension, and the input problem-solving idea feature of the current problem-solving step network contains a strategy for processing quantitative relationships. Then, during the interactive fusion process, the server will adjust and refine the processing strategy according to this quantitative relationship condition, thereby generating a problem restriction sub-feature vector for this problem-solving step network for the target Mathematical Olympiad learning problem.

[0046] After the Zth (here we assume Z=3) problem-solving step network generates the problem-restricted sub-feature vector for the target Mathematical Olympiad learning problem, the problem-restricted sub-feature vector is output as a problem-restricted feature vector. This problem-restricted feature vector reflects the solution content feature vector of the target Mathematical Olympiad learning problem. For example, this problem-restricted feature vector may contain features related to the solution content such as the key conditions and quantitative relationships of the problem obtained after being derived from multiple problem-solving step networks.

[0047] Step S132, generating a problem-solving idea limiting feature vector of the target Olympiad mathematics learning question based on the target-oriented problem-solving idea model under the restriction of the target problem-solving idea guiding information, wherein the problem-solving idea limiting feature vector is used to reflect the problem-solving idea feature vector of the problem-solving idea label guided by the target problem-solving idea guiding information.

[0048] Similarly, for each problem-solving step network, the input problem-solving idea features are derived under the constraints of the target problem-solving idea guidance information. This process is an interactive fusion of the target problem-solving idea guidance information and the input problem-solving idea features.

[0049] Assume that the target problem-solving guidance information is a specific problem-solving idea, such as guidance information for solving problems according to a special formula or method. When processing the first problem-solving step network (a=1), the input problem-solving idea feature is the first initial problem-solving idea feature, and the server adjusts and derives this initial feature according to the target problem-solving idea guidance information. For example, if the target problem-solving idea guidance information is about the summation method of a special series, and the initial problem-solving idea feature contains some basic analysis ideas about the series, then through interactive fusion, the initial problem-solving idea limiting feature vector of the first problem-solving step network for the target Olympiad learning problem will be generated. This vector preliminarily reflects the idea feature of solving problems according to the target problem-solving idea guidance information.

[0050] There is a derivation adjustment unit in the idea feature derivation branch of the a-th problem-solving step network. This derivation adjustment unit can adjust the granularity value of the feature vector output by the idea feature derivation branch of the a-th problem-solving step network when interacting and fusing with the feature vector output by the question feature derivation branch of the a-th problem-solving step network.

[0051] For example, assuming that the derivation and adjustment unit is in a state switching state, when it is in the enabled state, the initial problem-solving idea limiting feature vector is directly output as the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem; when it is in the disabled state, the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem is determined to be empty.

[0052] If the derivation adjustment unit is in the derivation granularity adjustment state and has a preset derivation granularity parameter, the initial problem-solving idea limiting feature vector is processed according to the derivation granularity parameter to generate the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem.

[0053] After the generation of the problem-solving idea limiting sub-feature vector of the target Olympiad learning problem for the Zth problem-solving step network is completed, this problem-solving idea limiting sub-feature vector is output as a problem-solving idea limiting feature vector. This problem-solving idea limiting feature vector is used to reflect the problem-solving idea feature vector of the problem-solving idea label guided by the target problem-solving idea guidance information. For example, this problem-solving idea limiting feature vector may include key features about the problem-solving idea that are gradually derived in each problem-solving step network according to the target problem-solving idea guidance information, such as the order of the problem-solving steps, the application of special problem-solving methods, etc.

[0054] Step S133, interactively merge the question-limited feature vector and the problem-solving idea-limited feature vector to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information.

[0055] Assume that the question-limited feature vector includes Q question logic components (Q is a positive integer) on Q knowledge logic dimensions, and the solution idea-limited feature vector includes Q idea logic components on Q knowledge logic dimensions.

[0056] For the question logic component and the idea logic component on the same knowledge logic dimension, firstly, merge operations are performed separately to generate the merged operation components of the goal-oriented problem-solving idea model on each knowledge logic dimension. For example, if in the numerical calculation logic dimension, the question logic component represents the numerical conditions of the question, and the idea logic component represents the numerical calculation strategy in the problem-solving idea, through the merge operation, a merged operation component that comprehensively considers the question conditions and the problem-solving idea is obtained.

[0057] Then, the logical components of the questions and the logical components of the train of thought on the same knowledge logic dimension are preliminarily connected to generate the connection components of the goal-oriented problem-solving train of thought model on each knowledge logic dimension. This connection processing is a logical connection operation that connects the logical components of the questions and the logical components of the train of thought more closely.

[0058] Then, the merge operation components and connection components of the goal-oriented problem-solving thinking model on the Q knowledge logic dimensions are fused. Specifically, the merge operation components and connection components of the goal-oriented problem-solving thinking model on the same knowledge logic dimension are firstly subjected to advanced connection operations to generate advanced connection components of the goal-oriented problem-solving thinking model on each knowledge logic dimension.

[0059] Then, the Mathematical Olympiad logic operation component is obtained, which contains Q logic operators. For example, for each knowledge logic dimension, there is a corresponding logic operator. These Q logic operators are used to perform logic operations on the advanced connection components of the goal-oriented problem-solving thinking model on the Q knowledge logic dimensions, and the operation result characteristics of the goal-oriented problem-solving thinking model on the Q knowledge logic dimensions are generated.

[0060] Finally, the operation result features of the target-oriented problem-solving model on Q knowledge logic dimensions are spliced ​​according to the Mathematical Olympiad logic operation component to generate the problem-solving path feature vector corresponding to the target problem-solving guidance information. This problem-solving path feature vector includes the solution content feature vector of the target Mathematical Olympiad learning question and the problem-solving feature vector of the problem-solving label guided by the corresponding problem-solving guidance information.

[0061] In a possible implementation, the goal-oriented problem-solving thinking model includes Z problem-solving step networks connected in sequence for feature derivation, and any one of the problem-solving step networks includes a question feature derivation branch and a thinking feature derivation branch connected in parallel, and Z is a positive integer.

[0062] Among them, the goal-oriented problem-solving idea model has a first initial problem-solving idea feature to be optimized, the Z problem-solving step networks include the a-th problem-solving step network, a is not greater than Z, and the a-th problem-solving step network has an input problem-solving idea feature.

[0063] If a is equal to 1, then the input problem-solving idea feature is the first initial problem-solving idea feature. If a is greater than 1, then the input problem-solving idea feature is the derived feature vector of the a-1th problem-solving step network of the target-oriented problem-solving idea model. The derived feature vector is obtained by interactively fusing the feature vector output by the question feature derivation branch in the a-1th problem-solving step network and the feature vector output by the idea feature derivation branch.

[0064] In this embodiment, the goal-oriented problem-solving thinking model is composed of Z problem-solving step networks connected in sequence for feature derivation, where Z is a positive integer. Each problem-solving step network includes a question feature derivation branch and a thinking feature derivation branch connected in parallel.

[0065] This goal-oriented problem-solving model has a first initial problem-solving feature to be optimized. When processing the first problem-solving step network (i.e. a=1), the input problem-solving feature of the problem-solving step network is the first initial problem-solving feature. This is the starting state of the entire problem-solving process, and this initial problem-solving feature is the basis for subsequent derivation.

[0066] When processing the a-th problem-solving step network (a>1 and a is not greater than Z), its input problem-solving idea feature is the derived feature vector of the a-1-th problem-solving step network of the goal-oriented problem-solving idea model.

[0067] Taking a=2 as an example, the process of obtaining the input problem-solving idea features of the second problem-solving step network is as follows:

[0068] First, the question feature derivation branch in the first problem-solving step network will derive the current problem-solving related information (such as the conditions and elements in the question) and output a feature vector. At the same time, the idea feature derivation branch in the first problem-solving step network will derive the information related to the problem-solving idea based on its own logic and rules, and also output a feature vector.

[0069] Then, the two feature vectors are interactively fused. This interactive fusion operation is performed according to pre-set rules. For example, it may be to perform specific mathematical operations (such as addition, multiplication, etc.) on the values ​​of the corresponding dimensions of the two vectors, or to combine them according to a certain logical relationship. After such interactive fusion, the result is the derived feature vector of the first problem-solving step network, and this derived feature vector becomes the input solution idea feature of the second problem-solving step network.

[0070] When a=3, in the same way as above, the question feature derivation branch and the idea feature derivation branch of the second problem-solving step network output feature vectors respectively, and after interactive fusion, the derivation feature vector of the second problem-solving step network is obtained, which is used as the input problem-solving idea feature of the third problem-solving step network. And so on, until all problem-solving step networks are processed (until a=Z).

[0071] In a possible implementation, step S131 includes:

[0072] Step S1311, based on the question feature derivation branch in the a-th problem-solving step network, derive the input problem-solving idea features under the restriction condition of the question representation logical vector, and generate a question restriction sub-feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning question.

[0073] Step S1312, output the problem-restricted sub-feature vector of the target Mathematical Olympiad learning problem for the Yth problem-solving step network of the target-oriented problem-solving model as the problem-restricted feature vector. The process of deriving the input problem-solving feature under the restriction of the problem-representing logic vector is the process of interactively fusing the problem-representing logic vector and the input problem-resolving feature.

[0074] In this embodiment, the target Mathematical Olympiad learning problem is processed with the target-oriented problem-solving thinking model. For the first problem-solving step network (a=1), its input problem-solving thinking feature is the first initial problem-solving thinking feature of the target-oriented problem-solving thinking model.

[0075] Assuming that the target Olympiad learning question is about a sequence, the question representation logic vector contains information such as the first term, the number of terms, the common difference (if it is an arithmetic sequence) or the common ratio (if it is a geometric sequence) of the sequence, as well as the logical relationship between the terms of the sequence and the requirements of the question (such as finding the value of a certain term or the sum of the sequence, etc.).

[0076] In the question feature derivation branch in the first problem-solving step network, the input problem-solving idea features are deduced according to the constraints of the question representation logic vector. This process is to interactively fuse the question representation logic vector and the input problem-solving idea features. For example, the initial problem-solving idea feature may be to consider the problem from the basic definition of the sequence, and the information such as the first term and number of terms in the question representation logic vector is interactively fused with the initial problem-solving idea feature.

[0077] If the initial problem-solving idea is based on the general formula of the sequence, then the information such as the first term, common difference (common ratio) in the problem representation logic vector will affect the construction and adjustment of this general formula. Through this interactive fusion, the problem restriction sub-feature vector of the first problem-solving step network for the target Olympiad learning problem is generated. This problem restriction sub-feature vector may contain the general formula form preliminarily determined under the problem condition constraints, or the preliminary application of the number of terms, first term and other information in the problem-solving idea.

[0078] When processing the second problem-solving step network (a=2), its input problem-solving idea feature is the derived feature vector of the first problem-solving step network.

[0079] The problem feature derivation branch in the second problem-solving step network derives the input problem-solving idea feature under the constraints of the problem representation logic vector. More information in the problem representation logic vector, such as the relationship between the middle term and the first term and the number of terms in the sequence, will further affect the problem-solving idea.

[0080] For example, if the derived feature vector of the first problem-solving step network contains a preliminary constructed general formula, in the second problem-solving step network, the relationship between the number of series terms and the intermediate terms in the problem representation logic vector will interact and merge with this general formula. If the question requires the value of a certain intermediate term, the general formula is adjusted according to the relationship between the number of terms, the first term, and the common difference (common ratio), thereby generating a problem-restricted sub-feature vector for the second problem-solving step network for the target Olympiad learning question. This problem-restricted sub-feature vector may contain a more accurate general formula, or more specific problem-solving step information for the problem of finding the intermediate term.

[0081] For the subsequent problem-solving step networks (a>2), the process is similar. Taking a=3 as an example, its input problem-solving idea feature is the derived feature vector of the second problem-solving step network.

[0082] The problem feature derivation branch in the third problem-solving step network performs derivation processing under the constraints of the problem representation logic vector. If the problem has more constraints, such as the impact of the parity of the number of sequence terms on the summation method of the sequence, these conditions will interact and merge with the input problem-solving idea features.

[0083] Assuming that the derived feature vector of the second problem-solving step network contains relevant information for finding the middle term, in the third problem-solving step network, the problem-solving ideas are further adjusted according to the parity of the number of terms in the problem representation logic vector and the summation requirements. This may change the application of the summation formula or regroup the number series for summation, thereby generating the problem-restricted sub-feature vector of the third problem-solving step network for the target Mathematical Olympiad learning problem.

[0084] When the Y-th problem-solving step network is reached, the problem-restricted sub-feature vector of the Y-th problem-solving step network for the target Olympiad mathematics learning problem is output as a problem-restricted feature vector.

[0085] For example, assuming Y=3, the problem-restricted sub-feature vector of the third problem-solving step network for the target Mathematical Olympiad learning problem contains information such as the final application of the summation formula, the grouping of the series (if any), and the form of the final answer (such as the sum expressed by a certain expression). This problem-restricted sub-feature vector fully reflects the solution content feature vector of the target Mathematical Olympiad learning problem, so it is output as a problem-restricted feature vector. This problem-restricted feature vector will be subsequently interactively merged with the problem-solving idea-restricted feature vector to generate a problem-solving path feature vector.

[0086] In a possible implementation, step S132 includes:

[0087] Step S1321, based on the derivation branch of the idea features in the a-th problem-solving step network, derive the input problem-solving idea features under the restriction of the target problem-solving idea guidance information, and generate an initial problem-solving idea limiting feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem.

[0088] Step S1322: Generate a sub-feature vector for limiting the solution idea of ​​the a-th problem-solving step network for the target Olympiad mathematics learning problem based on the feature vector limiting the initial solution idea.

[0089] Step S1323, the Y-th problem-solving step network of the target-oriented problem-solving idea model is output as the problem-solving idea limiting feature vector for the problem-solving idea of ​​the target Olympiad learning problem. The process of deriving the input problem-solving idea feature under the restriction of the target problem-solving idea guidance information is the process of interactively fusing the target problem-solving idea guidance information and the input problem-solving idea feature.

[0090] In this embodiment, when processing a target Olympiad mathematics learning problem, for the first problem-solving step network (a=1) in the target-oriented problem-solving thinking model, its input problem-solving thinking feature is the first initial problem-solving thinking feature of the target-oriented problem-solving thinking model.

[0091] Assume that the target Olympiad mathematics learning topic is about the calculation of the area of ​​geometric figures, and the target problem-solving guidance information is to calculate the area of ​​complex figures according to the segmentation method.

[0092] In the first problem-solving step network, the feature derivation branch of the idea derives the input problem-solving idea features under the constraint of the target problem-solving idea guidance information. This process is an interactive fusion of the target problem-solving idea guidance information and the input problem-solving idea features.

[0093] For example, the first initial problem-solving idea feature may be to consider the area calculation from the perspective of the overall figure, and the target problem-solving idea guidance information (segmentation method) interacts and merges with it. The idea feature derivation branch will think about how to transition from the initial analysis of the overall figure to the segmentation idea according to the requirements of the segmentation method. It may be based on factors such as the shape of the figure and the characteristics of the edges to preliminarily determine which basic figures (such as triangles, rectangles, etc.) the figure can be segmented into, thereby generating the first problem-solving step network for the target Olympiad learning problem. The initial problem-solving idea limiting feature vector. This vector contains information related to the problem-solving idea, such as the segmentation direction and basic figure type preliminarily determined according to the segmentation method.

[0094] When processing the second problem-solving step network (a=2), its input problem-solving idea feature is the derived feature vector of the first problem-solving step network.

[0095] The thought feature derivation branch in the second problem-solving step network derives the input problem-solving thought feature under the restriction of the target problem-solving thought guidance information. Continuing with the example of geometric figure area calculation, the target problem-solving thought guidance information (segmentation method) still plays a restrictive role.

[0096] If the derived feature vector of the first problem-solving step network contains the type of segmented basic figures that are preliminarily determined, in the second problem-solving step network, the idea feature derivation branch will further refine the solution idea based on this information. For example, based on the target problem-solving idea guidance information, think about how to determine the position of the dividing line to make it easier to calculate the area of ​​the segmented figure. Factors such as the symmetry of the figure and the ratio of the side lengths may be considered to generate the initial solution idea limit feature vector of the second problem-solving step network for the target Olympiad learning problem. This vector contains more specific information about the solution idea for determining the position of the dividing line.

[0097] For the subsequent problem-solving step networks (a>2), the process is similar. Taking a=3 as an example, its input problem-solving idea feature is the derived feature vector of the second problem-solving step network.

[0098] The feature derivation branch in the third problem-solving step network is deduced under the constraints of the target problem-solving guidance information. In the example of calculating the area of ​​geometric figures, the target problem-solving guidance information (segmentation method) continuously affects the problem-solving ideas.

[0099] If the derived feature vector of the second problem-solving step network contains information about the location of the dividing line, in the third problem-solving step network, the idea feature derivation branch will further determine the order of calculating the area based on this information. For example, according to the requirement of simple calculation in the target problem-solving idea guidance information, the area of ​​the small figures after segmentation is first calculated to be more conducive to the subsequent summation operation, thereby generating the initial solution idea limiting feature vector of the third problem-solving step network for the target Olympiad learning problem. This vector contains the problem-solving idea information about the order of area calculation.

[0100] Next, in the first problem-solving step network, a problem-solving idea limiting sub-feature vector is generated based on the initial problem-solving idea limiting feature vector.

[0101] For the example of calculating the area of ​​geometric figures mentioned above, the initial problem-solving idea of ​​the network in the first problem-solving step limits the feature vector to include information such as the segmentation direction and basic figure type preliminarily determined according to the segmentation method.

[0102] The derivation adjustment unit in the derivation branch of the idea feature of the first problem-solving step network will adjust the initial problem-solving idea limiting feature vector. For example, if the derivation adjustment unit detects that the current segmentation direction may cause subsequent calculations to be complicated, it may fine-tune the segmentation direction. If there are multiple choices for segmenting basic graphics in the segmentation method, it may re-determine a more suitable basic graphic type based on the specific characteristics of the graphics, thereby generating the first problem-solving step network for the target Olympiad learning problem. The sub-feature vector of the problem-solving idea limiting. This vector contains more optimized problem-solving idea information after adjustment, such as more reasonable segmentation directions and basic graphic types.

[0103] When processing the second problem-solving step network (a=2), the problem-solving idea limiting sub-feature vector is also generated based on the initial problem-solving idea limiting feature vector.

[0104] If the initial problem-solving idea limiting feature vector of the second problem-solving step network contains problem-solving idea information determined by the position of the dividing line, the derivation adjustment unit in the idea feature derivation branch will adjust it.

[0105] For example, the derivation and adjustment unit may further refine the position of the dividing line based on the actual side length value of the figure. If there are multiple possible dividing line positions, it will select the optimal dividing line position based on the requirements of computational simplicity in the target problem-solving guidance information (segmentation method), thereby generating a second problem-solving step network for the solution idea of ​​the target Olympiad learning problem. This vector contains more accurate information about the dividing line position.

[0106] For the subsequent problem-solving step network (a>2), take a=3 as an example.

[0107] The initial solution idea limiting feature vector of the network in the third problem-solving step contains the solution idea information about the area calculation order. The derivation adjustment unit in the idea feature derivation branch will adjust it.

[0108] For example, the derivation and adjustment unit may readjust the calculation order according to the specific numerical size of the area of ​​the small figure after segmentation. If there is a special numerical relationship (such as the area of ​​some small figures is equal or multiple), it will optimize the calculation order according to the requirements of using numerical relationships to simplify calculations in the target problem-solving guidance information (segmentation method), thereby generating the third problem-solving step network for the solution idea of ​​the target Olympiad learning problem. The vector contains the optimized area calculation order information.

[0109] When the Y-th problem-solving step network is reached, the problem-solving idea limiting sub-feature vector of the Y-th problem-solving step network for the target Olympiad mathematics learning problem is output as a problem-solving idea limiting feature vector.

[0110] For example, assuming Y=3, the solution idea sub-feature vector of the third problem-solving step network for the target Mathematical Olympiad learning question contains information related to the solution idea, such as the optimized area calculation order, the precise location of the dividing line, and the final basic graphic type. This solution idea sub-feature vector fully reflects the solution idea feature vector of the solution idea label guided by the target solution idea guidance information, so it is output as the solution idea feature vector. This solution idea feature vector will be interactively merged with the question-limited feature vector in the future to generate a solution path feature vector.

[0111] In a possible implementation, the idea feature derivation branch of the a-th problem-solving step network includes a derivation adjustment unit, and the derivation adjustment unit is used to adjust the feature vector output by the idea feature derivation branch of the a-th problem-solving step network and the granularity value when interacting and fusing with the feature vector output by the question feature derivation branch of the a-th problem-solving step network.

[0112] Step S1322 includes:

[0113] The initial problem-solving idea limiting feature vector is adjusted according to the derivation adjustment unit to generate a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem.

[0114] In a possible implementation, the derivation adjustment unit is in a state switching state or a derivation granularity adjustment state, the state switching state has an enabled state and a disabled state, and the derivation granularity adjustment state has a preset derivation granularity parameter.

[0115] The step of adjusting the initial problem-solving idea limiting feature vector according to the derivation adjustment unit to generate a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem includes:

[0116] Step S1322-1, if the derivation adjustment unit is in the state switching state, and the state switching state is in the enabled state, then the initial problem-solving idea limiting feature vector is output as the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem.

[0117] Step S1322-2, if the derivation adjustment unit is in the state switching state, and the state switching state is in the disabled state, it is determined that the sub-feature vector of the a-th problem-solving step network for the solution idea of ​​the target Olympiad learning problem is empty.

[0118] Step S1322-3, if the derivation adjustment unit is in the derivation granularity adjustment state, the initial problem-solving idea limiting feature vector is processed according to the derivation granularity parameter to generate a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem.

[0119] In this embodiment, when processing a target Mathematical Olympiad learning problem, for the a-th problem-solving step network in the target-oriented problem-solving thinking model, its thinking feature derivation branch includes a derivation adjustment unit. The function of this derivation adjustment unit is to adjust the granularity value of the feature vector output by the thinking feature derivation branch of the a-th problem-solving step network when interacting and fusing with the feature vector output by the problem feature derivation branch of the a-th problem-solving step network.

[0120] The derivation and adjustment unit plays a key role in the process of generating the a-th problem-solving step network based on the problem-solving idea limiting feature vector for the target Olympiad learning problem.

[0121] Next, let's take the target Mathematical Olympiad learning problem of calculating the area of ​​geometric figures (using the segmentation method) as an example. Assume that in the a-th problem-solving step network (such as a=2), the initial problem-solving idea limiting feature vector contains information about the preliminary determination of the location of the segmentation line, which is obtained based on the previous problem-solving steps and the target problem-solving idea guidance information (segmentation method).

[0122] If the derivation adjustment unit is in the state switching state, and this state switching state is in the enabled state, it means that in this particular problem-solving step, no additional adjustment or refinement is required to the initial problem-solving idea limiting feature vector.

[0123] At this point, the initial solution idea limiting feature vector is directly output as the solution idea limiting sub-feature vector of the a-th solution step network (here, the second solution step network) for the target Mathematical Olympiad learning problem. In other words, in this step, the information about the initial determination of the position of the dividing line remains unchanged and is directly used as part of the solution idea limiting sub-feature vector for subsequent interactive fusion or other operations with the feature vector output by the problem feature derivation branch.

[0124] Taking the area calculation of geometric figures as an example, suppose that in the a-th problem-solving step network (such as a=3), the initial problem-solving idea limits the feature vector to include preliminary ideas about the order of area calculation.

[0125] If the derivation adjustment unit is in a state switching state, and this state switching state is in a disabled state, it means that in this problem-solving step, the current initial problem-solving idea limiting feature vector does not meet the requirements or should not be used.

[0126] Therefore, it is determined that the solution idea of ​​the a-th problem-solving step network (here, the third problem-solving step network) for the target Olympiad learning problem is limited to an empty sub-eigenvector. This means that in this problem-solving step, it is necessary to rethink or adjust the solution idea, because the preliminary idea about the area calculation order obtained before cannot be used for subsequent operations.

[0127] Still taking the area calculation of geometric figures as an example, assume that in the a-th problem-solving step network (such as a=2), the initial problem-solving idea limits the feature vector to include a preliminary method for calculating the area of ​​the small figure after segmentation.

[0128] If the derivation adjustment unit is in the derivation granularity adjustment state and has a preset derivation granularity parameter, the derivation granularity parameter may be related to the calculation accuracy, the subdivision degree of the calculation steps, etc.

[0129] For example, the preset derivation granularity parameter specifies the extent to which the special shape features of the figure need to be considered when calculating the area of ​​the small figure. If this parameter requires a more detailed consideration of the relationship between the hypotenuse and the right angle side of the figure (for a triangular small figure), then the initial solution idea is limited to the feature vector according to this derivation granularity parameter.

[0130] Specifically, the formula or method for calculating the area of ​​small figures may be adjusted to consider more geometric relationship factors. In this way, the a-th problem-solving step network (here is the second problem-solving step network) is generated for the solution idea sub-feature vector of the target Olympiad learning problem. This solution idea sub-feature vector contains the small figure area calculation method adjusted according to the derivation granularity parameter, which is used for subsequent problem-solving operations, such as interactive fusion with the feature vector output by the problem feature derivation branch or further derivation of the solution idea.

[0131] In a possible implementation, the question-defining feature vector includes Q question logic components on Q knowledge logic dimensions, and the problem-solving idea-defining feature vector includes Q idea logic components on the Q knowledge logic dimensions, where Q is a positive integer.

[0132] Step S133 includes:

[0133] Step S1331, merge the question logic components and the idea logic components on the same knowledge logic dimension respectively, and generate the merged operation components of the goal-oriented problem-solving idea model on each knowledge logic dimension.

[0134] Step S1332, preliminarily connect the question logic components and the idea logic components on the same knowledge logic dimension, and generate the connection components of the goal-oriented problem-solving idea model on each knowledge logic dimension.

[0135] Step S1333, the merging operation components and the connection components of the target-oriented problem-solving idea model on the Q knowledge logic dimensions are fused to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information. The interactive fusion includes the merging operation, the preliminary connection processing and the fusion processing.

[0136] In a possible implementation, step S1333 includes:

[0137] Step S1333-1, performing advanced connection operations on the merging operation components and connection components of the goal-oriented problem-solving thinking model on the same knowledge logic dimension, and generating advanced connection components of the goal-oriented problem-solving thinking model on each knowledge logic dimension.

[0138] Step S1333-2, obtain an Olympiad logic operation component. The Olympiad logic operation component includes Q logic operators.

[0139] Step S1333-3, use the Q logical operators to perform logical operations on the advanced connection components of the goal-oriented problem-solving idea model on the Q knowledge logical dimensions, and generate the operation result characteristics of the goal-oriented problem-solving idea model on the Q knowledge logical dimensions. One logical operator is used to perform logical operations on the advanced connection components of the goal-oriented problem-solving idea model on one knowledge logical dimension.

[0140] Step S1333-4, based on the Mathematical Olympiad logic operation component, the operation result features of the target-oriented problem-solving thinking model on the Q knowledge logic dimensions are spliced ​​to generate a problem-solving path feature vector corresponding to the target problem-solving thinking guidance information. The mapping conversion process includes the logic operation and the splicing process.

[0141] In this embodiment, it is assumed that the target Olympiad mathematics learning topic is about geometric figures, Q is 3, and the three knowledge logic dimensions are respectively the graphic shape logic dimension, the graphic relationship logic dimension, and the numerical calculation logic dimension.

[0142] For the question-limited feature vector, the question logic component in the graphic shape logic dimension may contain information related to the shape of the graphic, such as whether it is a triangle, quadrilateral, etc., such as the type of triangle (equilateral, isosceles, etc.); the question logic component in the graphic relationship logic dimension may contain the positional relationship between the graphics (such as parallel, vertical, etc.); the question logic component in the numerical calculation logic dimension may contain numerical information such as side length and angle size.

[0143] For the feature vector limited by the problem-solving ideas, the idea logic component in the graphic shape logic dimension may include the problem-solving entry point determined according to the graphic shape, such as considering the special properties when seeing an equilateral triangle; the idea logic component in the graphic relationship logic dimension may include how to use the graphic relationship to construct the problem-solving steps; the idea logic component in the numerical calculation logic dimension may include ideas such as the order of calculating the length of sides or angles.

[0144] In the dimension of shape logic, the logical component of the question (triangle type, etc.) and the logical component of the train of thought (the entry point for solving the problem based on the triangle type) are combined. For example, if the logical component of the question indicates that it is an equilateral triangle, and the logical component of the train of thought is to solve the problem by using the property that the three sides of the equilateral triangle are equal, the combined operation component generated after the combined operation may be a comprehensive information including the shape information of the equilateral triangle and how to use its properties to solve the problem.

[0145] Similarly, in the graphic relationship logic dimension and the numerical calculation logic dimension, the corresponding question logic components and idea logic components are merged and operated respectively to generate the merged operation components of the goal-oriented problem-solving idea model in these two knowledge logic dimensions.

[0146] In the logical dimension of graphic shape, the logical components of the question (triangle type, etc.) and the logical components of the thinking (the entry point for solving the problem based on the triangle type) are preliminarily connected. This connection is not a simple merger, but a logical connection. For example, the triangle type and the thinking of how to use its properties to solve the problem are connected in a logical structure to generate the connection component of the goal-oriented problem-solving thinking model in the logical dimension of graphic shape.

[0147] In the logical dimension of graphic relationships, the positional relationship between graphics is preliminarily connected with the idea of ​​using this relationship to construct the problem-solving steps. In the logical dimension of numerical calculation, the numerical information such as side length and angle size is preliminarily connected with the sequential idea of ​​calculating these numerical values, and the connection components of the goal-oriented problem-solving idea model in these two knowledge logical dimensions are generated respectively.

[0148] Next, in the logical dimension of graphic shape, the previously obtained merge operation component (including the shape information of the equilateral triangle and the comprehensive information on how to use its properties to solve the problem) and the connection component (the logical connection between the triangle type and the solution idea) are subjected to advanced connection operations. This operation may further strengthen the logical relationship between the two and generate an advanced connection component of the goal-oriented problem-solving idea model in the logical dimension of graphic shape. For example, a tighter logical structure may be formed to clarify how the specific properties of the equilateral triangle are exactly applied to each step of the problem-solving.

[0149] Similarly, in the graphic relationship logic dimension and the numerical calculation logic dimension, the corresponding merging operation components and connection components are subjected to advanced connection operations to generate advanced connection components in these two knowledge logic dimensions.

[0150] Then, obtain the Olympiad Mathematical Olympiad logic operation component. Assume that the three logic operators contained in this Olympiad Mathematical Olympiad logic operation component are respectively for the graphic shape logic dimension, the graphic relationship logic dimension, and the numerical calculation logic dimension.

[0151] For the graphic shape logic dimension, the corresponding logical operator is used to perform logical operations on the advanced connection components of the goal-oriented problem-solving thinking model on this knowledge logic dimension. This logical operation may be to process the information in the advanced connection components according to specific Olympiad rules, for example, to perform a conversion or optimization based on the logical relationship between the properties of triangles and problem-solving ideas, and generate the operation result characteristics of the goal-oriented problem-solving thinking model on the graphic shape logic dimension.

[0152] Similarly, the other two logical operators are used to perform logical operations on the advanced connection components on the graphic relationship logical dimension and the numerical calculation logical dimension respectively, to generate the operation result features on these two knowledge logical dimensions.

[0153] Based on the Mathematical Olympiad logical operation component, the operation result features of the target-oriented problem-solving thinking model on the three knowledge logic dimensions are spliced ​​together. For example, the operation result features on the graphic shape logic dimension, the operation result features on the graphic relationship logic dimension, and the operation result features on the numerical calculation logic dimension are spliced ​​together in a certain order to generate the problem-solving path feature vector corresponding to the target problem-solving thinking guidance information. This problem-solving path feature vector completely contains the solution content feature vector of the target Mathematical Olympiad learning question and the problem-solving thinking feature vector of the problem-solving thinking label guided by the corresponding problem-solving thinking guidance information, providing a comprehensive foundation for the subsequent generation of problem-solving results.

[0154] In a possible implementation, step S140 includes:

[0155] Step S141, for each problem-solving path feature vector corresponding to the problem-solving idea guidance information, split the problem-solving path feature vector into an answer content feature vector and a problem-solving idea feature vector. For the answer content feature vector, based on the pre-constructed Olympiad knowledge classification system, a semantic analysis algorithm is used to identify the different types of knowledge elements contained in the problem-solving path feature vector. For the problem-solving idea feature vector, the logical elements representing the sequence of problem-solving steps, key logical judgment points, and special problem-solving method identifiers in the problem-solving idea feature vector are parsed out, thereby obtaining the parsed answer content knowledge element set and problem-solving idea logical element set.

[0156] In this embodiment, the server operates on the problem-solving path feature vector corresponding to each problem-solving idea guidance information. First, the problem-solving path feature vector is split into a solution content feature vector and a problem-solving idea feature vector.

[0157] For the feature vector of the answer content, a semantic analysis algorithm is used to identify it based on the pre-built Olympiad knowledge classification system. The feature vector of the answer content contains features related to various knowledge contents related to the Olympiad questions. The semantic analysis algorithm can identify different types of knowledge elements. For example, in the content involving the integration of geometry and algebra, it may identify knowledge elements such as the basic properties of geometric figures (such as shape characteristics, etc.) and the relationship between variables in algebraic expressions. These elements together constitute the knowledge element set of the answer content.

[0158] #For the problem-solving idea feature vector, the server parses the logical elements representing the sequence of problem-solving steps, key logic judgment points, and special problem-solving method identifiers. The sequence of problem-solving steps reflects the order of various operations in the problem-solving process, such as performing a certain calculation first and then performing a logic judgment; the key logic judgment point is the part that plays a key decision-making role in the problem-solving process, such as judging whether a certain condition is met to determine the subsequent problem-solving direction; the special problem-solving method identifier indicates that a specific Olympiad problem-solving method is used, such as using a special series summation formula, etc. These logical elements constitute the problem-solving idea logical element set.

[0159] Step S142, determine the knowledge field involved in the target Olympiad mathematics learning question based on the answer content knowledge element set, and extract basic concepts, theorems and formulas related to the knowledge field from the Olympiad mathematics knowledge classification system to construct a corresponding preliminary knowledge framework. At the same time, based on the problem-solving idea logic element set, determine the corresponding problem-solving process framework, integrate the preliminary knowledge framework and the problem-solving process framework to form a directional problem-solving framework for each of the problem-solving path feature vectors.

[0160] In this embodiment, the knowledge domain involved in the target Mathematical Olympiad learning problem is determined based on the set of knowledge elements of the answer content. For example, if the set of knowledge elements of the answer content contains more elements related to the properties and relationships of geometric figures, it is determined that the problem involves the field of geometric knowledge; if it contains more elements related to number series, it is determined that the problem involves the field of number series knowledge, etc.

[0161] The basic concepts, theorems and formulas related to the knowledge field are extracted from the Olympiad knowledge classification system to build a preliminary knowledge framework. Taking the geometry knowledge field as an example, the definitions of various geometric figures, related theorems (such as the triangle congruence theorem, etc.) and formulas for calculating area and perimeter will be extracted; if it is the sequence knowledge field, the general formula concept of the sequence and the summation formula will be extracted. These contents constitute the preliminary knowledge framework.

[0162] The corresponding problem-solving process framework is determined based on the problem-solving logic element set. The problem-solving process framework is constructed according to the problem-solving step sequence, key logic judgment points, and special problem-solving method identifiers. For example, if the problem-solving step sequence in the problem-solving logic element set is to first analyze the properties of the sequence and then select a suitable summation formula based on the properties, then the problem-solving process framework is constructed in this order.

[0163] The preliminary knowledge framework and the problem-solving process framework are integrated to form a directional problem-solving framework for each problem-solving path feature vector. This framework closely combines knowledge content with the problem-solving process, and clarifies how to use relevant knowledge to solve problems in each problem-solving step.

[0164] Step S143, under the guidance of the preliminary knowledge framework, for each element in the answer content knowledge element set, operations are performed according to the order and logic in the problem-solving process framework, wherein each step of the operation is based on the operation result of the previous step and the relevant elements in the problem-solving process framework, and the problem-solving process is gradually advanced to obtain the problem-solving operation process record corresponding to each problem-solving path feature vector, wherein the problem-solving operation process record includes the operation content, operation basis, operation input and operation output of each step.

[0165] Step S144, based on the last operation output in the problem-solving operation process record and the preset requirements of the target Mathematical Olympiad learning problem, determine whether the problem-solving operation process record has obtained a complete problem-solving result, wherein, if it is a numerical problem, check whether the value of the problem-solving result meets all the conditions in the problem. If it is a proof problem, check whether the proof process of the problem-solving result is rigorous and complete.

[0166] Step S145: If it is detected that the problem-solving operation process record has obtained a complete problem-solving result, the preliminary problem-solving result of each problem-solving path feature vector is output; if it is detected that the problem-solving operation process record has not obtained a complete problem-solving result, the problem-solving operation process record is traced back to find the operation steps that may cause problems, and a preliminary result is generated to mark the problem points or error points.

[0167] Step S146, for the preliminary results of marking the problem points or error points, adjustments are made based on the problem-solving process framework according to the marked problem points or error points to generate the final problem-solving results corresponding to each problem-solving path feature vector after improvement and optimization. The final problem-solving results are used as the problem-solving path results of the target Olympiad mathematics learning questions under Y problem-solving idea labels.

[0168] In this embodiment, if it is detected that the problem-solving operation process record does not obtain a complete problem-solving result, for example, the numerical value in a numerical solution type question does not meet the requirements or there is a logical loophole in the proof in a proof type question.

[0169] Go back through the problem-solving process to find the steps that may cause problems. You may find a calculation error or a logical error in a certain step, and generate a preliminary result that marks the problem point or error point.

[0170] For the preliminary results of marking problem points or error points, adjustments are made based on the problem-solving process framework according to the marked problem points or error points. For example, if it is a calculation error, the calculation process is corrected; if it is a logical judgment error, the correct logical judgment is re-performed and the subsequent problem-solving steps are adjusted to generate the final problem-solving result corresponding to each problem-solving path feature vector after improvement and optimization. This final problem-solving result is the problem-solving path result of the target Olympiad learning problem under Y problem-solving idea labels.

[0171] In a possible implementation, step S110 includes:

[0172] Step S111, using a pre-constructed special vocabulary in the field of Mathematical Olympiad to segment the target Mathematical Olympiad learning question text, and generate at least one basic word segmentation unit. The special vocabulary in the field of Mathematical Olympiad contains reference mathematical terms, concept names, and operation symbols in Mathematical Olympiad. The question text is accurately segmented according to the vocabulary in the vocabulary, and the question is divided into meaningful basic units.

[0173] Step S112, identifying basic mathematical elements from the at least one basic word segmentation unit, and generating a processed question text containing basic mathematical elements, wherein the basic mathematical elements include numbers, mathematical symbols, variables, and preset mathematical objects.

[0174] Step S113, for each basic mathematical element, analyze the semantics of the basic mathematical element in the Mathematical Olympiad knowledge system, and construct a set of logical units based on the semantic analysis results. Each logical unit in the set of logical units is a logical structure that combines related basic mathematical elements according to a set logical relationship.

[0175] Step S114, sorting out the logical relationships of each logical unit in the set of logical units, and constructing a logic graph based on the results of sorting out the logical relationships, wherein the logic graph uses nodes to represent logical units and uses edges to represent logical relationships between logical units, and the logic graph has corresponding starting nodes and ending nodes, wherein the starting node is the initial condition logic unit in the target Olympiad mathematics learning problem, and the ending node is the logic unit corresponding to the content that the target Olympiad mathematics learning problem requires to be solved or proved.

[0176] Step S115, define a vector space, the dimension of the vector space is determined according to the reference logic type and question characteristics in the Olympiad knowledge system, and the dimension of the vector space includes numerical calculation logic dimension, geometric relationship logic dimension, logical reasoning logic dimension, and conditional judgment logic dimension.

[0177] Step S116, mapping each logic unit in the logic diagram to a corresponding dimension in the vector space, and generating a mapping result of each logic unit in the vector space.

[0178] Step S117, performing a synthesis operation on the mapping results on each dimension, and generating a problem representation logic vector of the target Olympiad mathematics learning problem through the synthesis results of each dimension.

[0179] Figure 2 The directional resource push system 100 applied to Mathematical Olympiad learning shown includes: a processor 1001 and a memory 1003. Among them, the processor 1001 and the memory 1003 are connected, such as being connected through a bus 1002. Optionally, the directional resource push system 100 applied to Mathematical Olympiad learning may also include a transceiver 1004, which may be used for data interaction between the server and other servers, such as data transmission and / or data reception. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of the directional resource push system 100 applied to Mathematical Olympiad learning does not constitute a limitation on the embodiments of the present application.

[0180] The memory 1003 is used to store program codes for executing the embodiments of the present application, and the execution is controlled by the processor 1001. The processor 1001 is used to execute the program codes stored in the memory 1003 to implement the steps shown in the above method embodiments.

[0181] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are pre-installed. When the processor executes the computer executable instructions, the targeted resource push method applied to Olympiad mathematics learning as above is implemented.

[0182] Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application, thereby helping to understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A method for pushing targeted resources for Mathematical Olympiad learning, characterized in that: The method comprises: Obtaining a question characterization logic vector of a target Mathematical Olympiad learning question, wherein the question characterization logic vector reflects the question comprehension logic of the target Mathematical Olympiad learning question; Obtaining X directional problem-solving thinking models and Y problem-solving thinking guidance information corresponding to the X directional problem-solving thinking models, wherein each problem-solving thinking guidance information represents a problem-solving thinking logic of a problem-solving thinking label, and each directional problem-solving thinking model corresponds to at least one problem-solving thinking guidance information, wherein X and Y are both positive integers, and X is not greater than Y; Generate, based on the X directional problem-solving model, a problem-solving path feature vector corresponding to each problem-solving idea guidance information under the constraints of the problem-representation logic vector and the corresponding problem-resolving idea guidance information, wherein each problem-resolving path feature vector includes a solution content feature vector of the target Mathematical Olympiad learning problem and a problem-resolving idea feature vector corresponding to the problem-resolving idea label guided by the problem-resolving idea guidance information, wherein the solution content feature vector is generated by performing artificial intelligence problem-resolving learning on the problem-resolving idea logic expressed by the problem-representation logic vector, and the problem-resolving idea feature vector is generated by performing artificial intelligence problem-resolving learning on the problem-resolving idea logic corresponding to the problem-resolving idea label guided by the problem-resolving idea guidance information; Generating problem-solving results for the problem-solving path feature vectors corresponding to the Y problem-solving idea guidance information, respectively, to obtain the problem-solving path results for the target Olympiad mathematics learning problem under the Y problem-solving idea labels; Obtain the target problem-solving idea label configured by the target user, and extract the target problem-solving path result under the target problem-solving idea label from the problem-solving path results under Y problem-solving idea labels of the target Olympiad learning problem, and then push the target problem-solving path result to the target user as a targeted resource.

2. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 1, characterized in that: Any one of the X directional problem-solving thinking models is a target-oriented problem-solving thinking model, and any problem-solving thinking guidance information corresponding to the target-oriented problem-solving thinking model is target problem-solving thinking guidance information. Generating a problem-solving path feature vector corresponding to each problem-solving thinking guidance information based on the X directional problem-solving thinking models under the constraints of the problem representation logical vector and the corresponding problem-solving thinking guidance information includes: generating a problem-limiting feature vector of the target Mathematical Olympiad learning problem based on the target-oriented problem-solving thinking model under the constraint condition of the problem-representing logical vector, wherein the problem-limiting feature vector is used to reflect the solution content feature vector of the target Mathematical Olympiad learning problem; generating a solution-thinking restriction feature vector of the target Mathematical Olympiad learning problem based on the target-oriented problem-thinking model under the restriction of the target problem-thinking guidance information, wherein the solution-thinking restriction feature vector is used to reflect the solution-thinking feature vector of the solution-thinking label guided by the target problem-thinking guidance information; The problem-limiting feature vector and the problem-solving idea-limiting feature vector are interactively fused to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information.

3. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 2, characterized in that: The goal-oriented problem-solving thinking model includes Z problem-solving step networks connected sequentially for feature derivation, and any problem-solving step network includes a question feature derivation branch and a thinking feature derivation branch connected in parallel, where Z is a positive integer; The goal-oriented problem-solving idea model has a first initial problem-solving idea feature to be optimized, the Z problem-solving step networks include an a-th problem-solving step network, a is not greater than Z, and the a-th problem-solving step network has an input problem-solving idea feature; If a is equal to 1, then the input problem-solving idea feature is the first initial problem-solving idea feature; if a is greater than 1, then the input problem-solving idea feature is the derived feature vector of the a-1th problem-solving step network of the target-oriented problem-solving idea model, and the derived feature vector is obtained by interactively fusing the feature vector output by the question feature derivation branch in the a-1th problem-solving step network and the feature vector output by the idea feature derivation branch.

4. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 3, characterized in that: Generating the problem-limiting feature vector of the target Mathematical Olympiad learning problem based on the target-oriented problem-solving thinking model under the restriction of the problem representation logical vector includes: Based on the problem feature derivation branch in the a-th problem-solving step network, the input problem-solving idea feature is deduced under the constraint of the problem representation logic vector to generate a problem constraint sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem; The Y-th problem-solving step network of the target-oriented problem-solving thinking model is targeted at the problem-restricted sub-feature vector of the target Olympiad mathematics learning problem, and outputted as the problem-restricted feature vector; wherein, the process of deriving and processing the input problem-solving thinking feature under the restriction condition of the problem representation logical vector is the process of interactively fusing the problem representation logical vector and the input problem-solving thinking feature.

5. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 3, characterized in that: The generating of a solution limiting feature vector of the target Mathematical Olympiad learning problem based on the target-oriented problem-solving idea model under the restriction of the target problem-solving idea guidance information includes: Derivative branches based on the thought feature in the a-th problem-solving step network, deriving the input problem-solving thought feature under the constraint of the target problem-solving thought guidance information, and generating an initial problem-solving thought limiting feature vector of the a-th problem-solving step network for the target Olympiad mathematics learning problem; Generating a sub-feature vector defining the solution idea of ​​the a-th problem-solving step network for the target Olympiad mathematics learning problem based on the initial problem-solving idea defining feature vector; The Y-th problem-solving step network of the target-oriented problem-solving idea model is used to limit the problem-solving idea of ​​the target Olympiad learning problem to a sub-feature vector, and the output is the problem-solving idea limiting feature vector; wherein, the process of deriving and processing the input problem-solving idea features under the restriction of the target problem-solving idea guidance information is the process of interactively fusing the target problem-solving idea guidance information and the input problem-solving idea features.

6. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 5, characterized in that: The thought feature derivation branch of the a-th problem-solving step network includes a derivation adjustment unit, wherein the derivation adjustment unit is used to adjust the granularity value of the feature vector output by the thought feature derivation branch of the a-th problem-solving step network when interacting and fusing with the feature vector output by the question feature derivation branch of the a-th problem-solving step network; The method of generating, based on the initial problem-solving idea limiting feature vector, a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem, includes: The initial problem-solving idea limiting feature vector is adjusted according to the derivation and adjustment unit to generate a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem.

7. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 6, characterized in that: The derivation adjustment unit is in a state switching state or a derivation granularity adjustment state, the state switching state has an enabled state and a disabled state, and the derivation granularity adjustment state has a preset derivation granularity parameter; The step of adjusting the initial problem-solving idea limiting feature vector based on the derivation and adjustment unit to generate a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem includes: If the derivation and adjustment unit is in the state switching state, and the state switching state is in the enabled state, the initial problem-solving idea limiting feature vector is output as a problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem; If the derivation and adjustment unit is in the state switching state, and the state switching state is in the disabled state, determining that the solution idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem is empty; If the derivation adjustment unit is in the derivation granularity adjustment state, the initial problem-solving idea limiting feature vector is processed according to the derivation granularity parameter to generate the problem-solving idea limiting sub-feature vector of the a-th problem-solving step network for the target Olympiad learning problem.

8. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 2, characterized in that: The problem definition feature vector includes Q problem logic components on Q knowledge logic dimensions, and the problem-solving idea definition feature vector includes Q idea logic components on the Q knowledge logic dimensions, where Q is a positive integer; The interactive fusion of the problem-limiting feature vector and the problem-solving idea-limiting feature vector to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information includes: Performing a merging operation on the question logic component and the train of thought logic component on the same knowledge logic dimension to generate a merging operation component of the goal-oriented problem-solving train of thought model on each knowledge logic dimension; Performing preliminary connection processing on the question logic components and the idea logic components on the same knowledge logic dimension respectively, and generating the connection components of the goal-oriented problem-solving idea model on each knowledge logic dimension; The merging operation components and connection components of the target-oriented problem-solving idea model on the Q knowledge logic dimensions are fused to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information; wherein, the interactive fusion includes the merging operation, the preliminary connection processing and the fusion processing.

9. The method for pushing targeted resources for Mathematical Olympiad learning according to claim 8, characterized in that: The step of fusing the combined operation components and the connection components of the target-oriented problem-solving idea model on the Q knowledge logic dimensions to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information includes: Performing advanced connection operations on the merging operation components and the connection components of the goal-oriented problem-solving thinking model on the same knowledge logic dimension to generate advanced connection components of the goal-oriented problem-solving thinking model on each knowledge logic dimension; Obtaining an Olympiad logic operation component; the Olympiad logic operation component includes Q logic operators; The Q logical operators are used to perform logical operations on the advanced connection components of the goal-oriented problem-solving thinking model on the Q knowledge logical dimensions, respectively, to generate operation result features of the goal-oriented problem-solving thinking model on the Q knowledge logical dimensions, and one logical operator is used to perform logical operations on the advanced connection components of the goal-oriented problem-solving thinking model on one knowledge logical dimension; The operation result features of the target-oriented problem-solving idea model on the Q knowledge logic dimensions are spliced ​​according to the Olympiad logic operation component to generate a problem-solving path feature vector corresponding to the target problem-solving idea guidance information.

10. A targeted resource push system for Mathematical Olympiad learning, characterized by: The targeted resource push system applied to Olympiad mathematics learning includes a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the targeted resource push method applied to Olympiad mathematics learning as described in any one of claims 1-9.

Images