Innovation and entrepreneurship coaching question and answer matching method and system based on semantic understanding

Abstract

The invention provides an innovation and entrepreneurship tutoring question and answer matching method and system based on semantic understanding, and relates to the technical field of natural language processing.The method comprises the steps that firstly, semantic understanding features of question sentences input by a user are extracted, and a corresponding knowledge node set is matched based on a preset innovation and entrepreneurship tutoring knowledge base; generating a semantic association path according to the hierarchical relationship and semantic association degree of the knowledge node set, determining a target question and answer strategy and generating optimized tutoring content according to the semantic association path, and finally updating node connection weights and association relationships in the innovation and entrepreneurship tutoring knowledge base according to feedback data of the user on the optimized tutoring content. Through deep semantic understanding and knowledge base dynamic optimization, accurate matching and personalized content generation of innovation and entrepreneurship tutoring questions and answers are realized, and efficient and high-quality tutoring support is provided for innovation and entrepreneurs.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing. Specifically, it relates to a method and system for matching innovation and entrepreneurship counseling questions and answers based on semantic understanding. Background Art

[0002] In the field of innovation and entrepreneurship education and counseling, with the rise of the innovation and entrepreneurship boom, the demand for personalized and precise counseling from a large number of entrepreneurs and potential entrepreneurs is increasing day by day. Traditional innovation and entrepreneurship counseling mostly relies on manual consultation, standardized courses, or simple keyword search question-and-answer systems. Although manual consultation has a certain degree of pertinence, it is limited by the differences in the energy and experience of consultants and is difficult to ensure service standardization and timeliness; standardized courses are difficult to adapt to the personalized questions and needs of different entrepreneurs; simple keyword search question-and-answer systems can only match literal content, cannot understand the deep semantics of users' questions, and are prone to problems such as answering irrelevant questions, information redundancy or omission, and cannot provide truly effective counseling support for innovation and entrepreneurs. Therefore, there is an urgent need for a method that can deeply understand the semantics of users, accurately match knowledge, generate adaptable counseling content, and continuously optimize the knowledge system based on user feedback to improve the efficiency and quality of innovation and entrepreneurship counseling. Summary of the Invention

[0003] In view of the problems mentioned above, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for matching innovation and entrepreneurship counseling questions and answers based on semantic understanding. The method includes: Extracting semantic understanding features of the question statement input by the user; Matching a set of knowledge nodes corresponding to the semantic understanding features based on a preset innovation and entrepreneurship counseling knowledge base; Generating a semantic association path according to the hierarchical relationship and semantic association degree of the set of knowledge nodes; Determining a target question-and-answer strategy according to the semantic association path and generating optimized counseling content; Updating the node connection weights and association relationships in the innovation and entrepreneurship counseling knowledge base according to the feedback data of the user on the optimized counseling content.

[0004] In a possible implementation manner of the first aspect, the extracting semantic understanding features of the question statement input by the user includes: Performing word segmentation on the question statement, dividing it into multiple semantic word units and removing the stop words in the multiple semantic word units; Invoking a pre-trained semantic encoding model to perform context encoding on the semantic word units to generate context vectors for each semantic word unit; Inputting the context vectors into a multi-head attention layer to calculate the semantic association weights between the semantic word units; Weightedly fuse the context vectors based on the semantic association weights to generate the global semantic features of the question statement; Input the global semantic features into a feature dimensionality reduction layer to extract the core intention features and auxiliary logic features as the semantic understanding features; wherein, the core intention features are used to represent the core goal of the user's question, and the auxiliary logic features are used to describe the dependency relationships and semantic hierarchical structures among multiple semantic word units in the question statement; Among them, the semantic encoding model is trained through the following steps: Collect sample question statements and their corresponding labeled intention tags in historical innovation and entrepreneurship counseling scenarios to construct a training data set; Aim to minimize the cross-entropy loss between the predicted intention tags and the labeled intention tags of the sample question statements, and optimize the parameters of the semantic encoding model.

[0005] In a possible implementation manner of the first aspect, the calling of the pre-trained semantic encoding model to perform context encoding on the semantic word units to generate context vectors for each semantic word unit includes: Input the semantic word units into a bidirectional long short-term memory network to respectively obtain the forward hidden state sequence and the backward hidden state sequence of each semantic word unit; Concatenate the forward hidden state sequence and the backward hidden state sequence of the same semantic word unit to generate an initial context vector; Perform layer normalization processing on the initial context vector, and fuse the normalized context vector with the initial context vector through residual connection to obtain an enhanced context vector; Input the enhanced context vector into a non-linear transformation layer to generate a set of context vectors with unified dimensions.

[0006] In a possible implementation manner of the first aspect, the matching of the knowledge node set corresponding to the semantic understanding features based on the preset innovation and entrepreneurship counseling knowledge base includes: Traverse the candidate knowledge nodes in the innovation and entrepreneurship counseling knowledge base, and call a node matching model to calculate the first similarity score between the node features of each candidate knowledge node and the core intention features; Filter out the candidate knowledge nodes with the first similarity score higher than the first preset threshold to form an initial matching set; Call a logic verification model to analyze whether the node description texts of the candidate knowledge nodes in the initial matching set satisfy the dependency relationships and semantic hierarchical structures corresponding to the auxiliary logic features. If not, they are determined as conflict nodes and the conflict nodes are removed to obtain the candidate knowledge nodes after logic verification; Arrange the candidate knowledge nodes after logical verification in descending order according to the first similarity score, and select the top N nodes as the optimized knowledge node set.

[0007] In a possible implementation manner of the first aspect, the training method of the node matching model includes: Extract positive sample knowledge nodes from the historically successfully matched Q&A data, and record the user question semantic understanding features corresponding to the positive sample knowledge nodes; Process the positive sample knowledge nodes to generate negative sample knowledge nodes, specifically including: retaining the hierarchical attributes and associated labels of the positive sample knowledge nodes, replacing the core keywords of the positive sample knowledge nodes to generate a node description text with semantic contradictions, or selecting nodes irrelevant to the user question semantics from the same tutoring field; Call the initialized node matching model to calculate the positive sample similarity between the node features of the positive sample knowledge nodes and the corresponding semantic understanding features, and the negative sample similarity between the node features of the negative sample knowledge nodes and the semantic understanding features; Construct a contrast loss function to maximize the difference between the positive sample similarity and the negative sample similarity, and update the parameters of the initialized node matching model through the backpropagation algorithm until the initialized node matching model converges to obtain a trained node matching model.

[0008] In a possible implementation manner of the first aspect, the generating of the semantic association path according to the hierarchical relationship and semantic association degree of the knowledge node set includes: Obtain the hierarchical attributes, associated labels and historical user interaction data of each node in the knowledge node set; Construct a vertical hierarchical connection relationship between nodes according to the hierarchical attributes; Expand the horizontal semantic association relationship between nodes according to the associated labels; Generate an initial semantic network based on the vertical hierarchical connection relationship and the horizontal semantic association relationship; Call the path generation algorithm to traverse the initial semantic network, generate multiple candidate semantic paths, and score each candidate semantic path based on the path length, node jump probability, node semantic weight and user feedback index, combined with the historical user interaction data, and select the candidate semantic path with the highest score as the target semantic association path.

[0009] In a possible implementation of the first aspect, the call path generation algorithm traverses the initial semantic network, generates multiple candidate semantic paths, and scores each candidate semantic path based on path length, node transition probability, node semantic weight, and user feedback metrics, in combination with the historical user interaction data, and selects the candidate semantic path with the highest score as the target semantic association path, including: Starting from the core intention node in the knowledge node set, traverse the child nodes downward according to the vertical hierarchical connection relationship, and expand the sibling nodes according to the horizontal semantic association relationship at the same time; During the traversal process, record the accessed node sequence and transition relationship to generate an initial path set; Filter redundant paths from the initial path set to obtain multiple candidate semantic paths, specifically including: merging paths with the same node sequence but different transition orders, and removing paths containing loop nodes or repeated transitions; Obtain the path length of each candidate semantic path, where the path length is obtained by counting the total number of all nodes in the candidate semantic path; Calculate the node transition probability of the candidate semantic path, where the node transition probability is determined based on the ratio of the number of transitions between adjacent nodes to the total number of transitions of the starting node in the historical user interaction data Extract the node semantic weight of the candidate semantic path, where the node semantic weight reads the preset weight value of each node from the innovation and entrepreneurship tutoring knowledge base, and the weight value is calculated based on the node citation frequency and the semantic relevance marked by the user; Obtain the user feedback metric, where the user feedback metric is obtained by counting the average satisfaction score generated after the user clicks on all nodes in the candidate semantic path in the historical session and the matching degree of the associated subsequent questions; After standardizing and transforming the path length, node transition probability, node semantic weight, and user feedback metric of each candidate semantic path, perform weighted summation to generate the comprehensive score of each candidate semantic path; Select the candidate semantic path with the highest comprehensive score as the target semantic association path.

[0010] In a possible implementation of the first aspect, the determining the target Q&A strategy and generating optimized tutoring content according to the semantic association path includes: Parse the node sequence in the semantic association path to identify the tutoring topic evolution logic and key knowledge point distribution corresponding to the node sequence; Match the basic Q&A strategy template from the preset Q&A strategy library according to the tutoring topic evolution logic, where the basic Q&A strategy template includes content generation rules, interaction process design, and resource reference methods; Dynamically adjust the basic Q&A strategy template based on the distribution of the key knowledge points to obtain an adjusted basic Q&A strategy template; Generate the optimized tutoring content according to the adjusted basic Q&A strategy template; the optimized tutoring content includes step-by-step guidance text, reference cases adapted to user needs, and extended resource links related to knowledge points; For example, in a possible implementation manner of the first aspect, parsing the node sequence in the semantic association path and identifying the tutoring topic evolution logic and key knowledge point distribution corresponding to the node sequence includes: Obtain the node attributes of each knowledge node in the semantic association path, where the node attributes include node type labels, historical interaction frequencies, and preset semantic weights; Divide the starting node, intermediate nodes, and terminating node in the node sequence according to the node type label, and extract the hierarchical jump direction and semantic association type between adjacent nodes; Construct a vertical evolution branch based on the hierarchical jump direction, generate a horizontal expansion branch according to the semantic association type, and topologically merge the vertical evolution branch and the horizontal expansion branch to generate an initial topic network; Traverse the node connection edges in the initial topic network, and calculate the comprehensive strength value of each connection edge, where the comprehensive strength value is obtained by weighted summing the normalized value of the historical interaction frequency, the logarithmic conversion value of the preset semantic weight, and the priority coefficient of the association type; Select the connection edges above the strength threshold according to the comprehensive strength value to form a core connection edge set, and extract the node subsequence corresponding to the core connection edge set as the key knowledge point distribution; Based on the node subsequence of the core connection edge set, extract the temporal change pattern of the node type label and the combination rule of the semantic association type to generate the tutoring topic evolution logic, where the temporal change pattern is determined by analyzing the occurrence order and frequency distribution of the node type labels in the node subsequence, and the combination rule is obtained by statistically analyzing the co-occurrence frequency and conditional probability of the semantic association type in the node subsequence.

[0011] In a possible implementation manner of the first aspect, the dynamically adjusting the basic Q&A strategy template based on the key knowledge point distribution to obtain an adjusted basic Q&A strategy template includes: Identify the priority parameters of each knowledge node in the key knowledge point distribution, where the priority parameters are calculated based on the semantic weight, historical user interaction frequency, and hierarchical attribute of the knowledge node; Divide the key knowledge point distribution into a core knowledge point set and an auxiliary knowledge point set according to the priority parameters, where the priority parameters of the core knowledge point set are higher than the preset priority threshold; Extract the content generation rules in the basic Q&A strategy template, set the content generation rules corresponding to the core knowledge point set as the default activation module, and generate extended trigger conditions according to the semantic relevance degree of the auxiliary knowledge point set; Analyze the unvisited knowledge point sequence in the user's historical interaction data, calculate the matching degree between the unvisited knowledge point sequence and the auxiliary knowledge point set, and screen out the unvisited knowledge points with a matching degree higher than the preset matching threshold as the insertion nodes of supplementary cases; According to the auxiliary logic features in the semantic understanding features, determine the logical embedding position of the insertion node of the supplementary case in the basic Q&A strategy template, and the logical embedding position is obtained by matching the dependency relationship in the auxiliary logic features with the interaction process node sequence in the basic Q&A strategy template; Integrate the parameters of the default activation module, extended trigger conditions and logical embedding position to generate an adjusted basic Q&A strategy template; For example, in a possible implementation manner of the first aspect, the matching of the basic Q&A strategy template from the preset Q&A strategy library according to the tutoring theme evolution logic includes: Extract the path attributes in the tutoring theme evolution logic, and the path attributes include the number of path stages, node type distribution and semantic association combination mode; Convert the number of path stages into a stage division vector, convert the node type distribution into a type density matrix, and convert the semantic association combination mode into an association coding sequence; Call the strategy matching model to calculate the first matching degree between the stage division vector and the stage labels of each strategy template in the preset Q&A strategy library, the second matching degree between the type density matrix and the node compatibility matrix of each strategy template, and the third matching degree between the association coding sequence and the association rule library of each strategy template; Perform standardization processing on the first matching degree, second matching degree and third matching degree, and perform weighted fusion based on the preset weight coefficient to generate the comprehensive adaptation score of each strategy template; Screen out the strategy templates with a comprehensive adaptation score higher than the adaptation threshold to form a candidate strategy set, and sort and optimize the candidate strategy set according to the time sequence change mode in the tutoring theme evolution logic, and the sorting and optimization are realized by matching the consistency of the time sequence change mode and the stage evolution in the historical application scenarios of the candidate strategy templates; Select the strategy template with the highest ranking in the sorted and optimized candidate strategy set as the basic Q&A strategy template.

[0012] For example, in a possible implementation of the first aspect, updating the connection weights and association relationships of nodes in the innovation and entrepreneurship coaching knowledge base according to the feedback data of the user on the optimized coaching content includes: Collecting the interaction behavior data of the user and the optimized coaching content, where the interaction behavior data includes the content click position, the number of case accesses, the resource link opening rate, and subsequent question statements; Cleaning and structuring the interaction behavior data to generate a feedback data set; Calculating the contribution score of each knowledge node in the semantic association path according to the feedback data set; For nodes with a contribution score higher than the second preset threshold, increasing the connection weight between the node and the node corresponding to the core intention feature; For nodes with a contribution score lower than the third preset threshold, reducing the connection weight of the node or removing invalid connections; Creating new nodes in the innovation and entrepreneurship coaching knowledge base according to the new semantic understanding features extracted from the subsequent question statements of the user and establishing association relationships with existing nodes; For example, in a possible implementation of the first aspect, calculating the contribution score of each knowledge node in the semantic association path according to the feedback data set includes: Extracting the interaction index set of each knowledge node from the feedback data set, where the interaction index set includes the click-through rate increment, the case access duration, the change value of the resource link opening rate, and the subsequent question matching degree; Performing time decay weighting on the click-through rate increment to generate a time-corrected click-through rate, where the time decay weighting is calculated based on the interval between the click behavior occurrence time and the current time and a preset decay factor; Converting the case access duration into a duration distribution percentile, converting the change value of the resource link opening rate into a change intensity coefficient, and generating a matching score for the subsequent question matching degree through a semantic similarity model; Normalizing the time-corrected click-through rate, the duration distribution percentile, the change intensity coefficient, and the matching score to obtain a standardized index set; Allocating the weights of each index in the standardized index set according to the preset contribution weights and generating an initial contribution score through linear weighting; Performing path correction on the initial contribution score based on the position attribute of the knowledge node in the semantic association path, where the path correction is achieved by calculating the weight coefficient of the path level where the node is located and the jump distance between the node and the core intention node; Taking the score after path correction as the final contribution score of the knowledge node.

[0013] On the other hand, an embodiment of the present invention further provides an innovation and entrepreneurship counseling Q&A matching system based on semantic understanding, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0014] Based on the above aspects, the embodiment of the present invention deeply extracts the semantic understanding features of the user's question statement, breaks through the limitations of traditional simple keyword matching, and accurately grasps the user's core needs. Based on the preset innovation and entrepreneurship counseling knowledge base, the corresponding knowledge node set is matched, and the semantic association path is generated by combining the node hierarchy relationship and semantic association degree, realizing the efficient screening and accurate connection of relevant content from a large amount of knowledge, constructing a knowledge context that conforms to the user's question logic, determining the target Q&A strategy based on this semantic association path and generating optimized counseling content to ensure that the output content highly adapts to the user's needs, improving the pertinence and effectiveness of the counseling. In addition, according to the feedback data of the user on the optimized counseling content, the node connection weights and association relationships in the innovation and entrepreneurship counseling knowledge base are dynamically updated, so that the knowledge system continuously evolves and continuously optimizes the Q&A matching effect, providing an intelligent and adaptive solution for innovation and entrepreneurship counseling. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a schematic execution flow diagram of an innovation and entrepreneurship counseling Q&A matching method based on semantic understanding provided by an embodiment of the present invention.

[0016] Figure 2 is a schematic diagram of exemplary hardware and software components of an innovation and entrepreneurship counseling Q&A matching system based on semantic understanding provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a schematic flow diagram of an innovation and entrepreneurship counseling Q&A matching method based on semantic understanding provided by an embodiment of the present invention. The innovation and entrepreneurship counseling Q&A matching method will be introduced in detail below.

[0018] Step S110: Extract the semantic understanding features of the question statement input by the user.

[0019] In this embodiment, in the scenario of innovation and entrepreneurship counseling, users may input various question statements. For example, "In the current market background where competition is intense and the policy environment is constantly changing, for a startup in the seed-round financing stage that mainly focuses on the innovative application of natural language processing technology in the field of artificial intelligence, how to formulate an efficient and sustainable financing and market expansion strategy while controlling capital costs and risks, combining the enterprise's own technological advantages and market positioning, and also considering the communication skills with potential investors and the possibility of subsequent enterprise strategic transformation?" To effectively process such question statements, it is necessary to extract their semantic understanding features.

[0020] Step S111: Perform word segmentation on the question statement, divide it into multiple semantic word units, and remove the stop words in the multiple semantic word units.

[0021] First, use a word segmentation tool to process the above question statement. The word segmentation tool can be based on various algorithms, such as rule-based word segmentation algorithms, statistics-based word segmentation algorithms, or a hybrid algorithm that combines the two. Taking the hybrid algorithm as an example, it first performs a preliminary segmentation of the statement using rules and then optimizes the segmentation results through a statistical model. After word segmentation, the statement is divided into numerous semantic word units, such as "current", "intense competition", "policy environment", "constantly changing", "market background", "seed-round financing stage", "field of artificial intelligence", "natural language processing technology", "innovative application", "startup", "control", "capital cost", "risk", "enterprise itself", "technological advantages", "market positioning", "formulate", "efficient", "sustainable", "financing", "market expansion strategy", "potential investors", "communication skills", "subsequent", "possibility of enterprise strategic transformation", etc. Then, remove the stop words among them. Stop words are usually some words that frequently appear in language expressions but contribute little to semantic understanding, such as "of", "in", "for", "in terms of", "how", "and", "at the same time", "also", etc. After the operation of removing stop words, a set of semantic word units with actual semantic value is obtained, denoted as M = {m1, m2, m3,..., mn}, where mi represents the i-th semantic word unit.

[0022] Step S112: Call a pre-trained semantic encoding model to perform context encoding on the semantic word units and generate context vectors for each semantic word unit.

[0023] Step S1121: Input the semantic word units into a bidirectional long short-term memory network to obtain the forward hidden state sequence and the backward hidden state sequence of each semantic word unit respectively.

[0024] The bidirectional long short-term memory network (Bi-LSTM) consists of a forward LSTM and a backward LSTM. The set of semantic word units M is sequentially input into the Bi-LSTM. For the forward LSTM, it processes the semantic word units from left to right. When processing the i-th semantic word unit mi, it can update its own state according to the information of the previously processed semantic word units m1 to mi-1, generating a forward hidden state ht_f. The backward LSTM processes the semantic word units from right to left. When processing the i-th semantic word unit mi, it will update its own state according to the information of the subsequent semantic word units mi+1 to mn, generating a backward hidden state ht_b. Specifically, the update process of the forward LSTM can be described as follows: First, according to the current input semantic word unit mi and the forward hidden state ht-1_f at the previous moment, calculate the input gate it_f, the forget gate ft_f, the cell state ct_f, and the output gate ot_f. The input gate it_f determines how much information of the current input semantic word unit will be added to the cell state; the forget gate ft_f determines how much information of the cell state at the previous moment will be retained; the cell state ct_f will be updated according to the results of the input gate and the forget gate; the output gate ot_f determines the hidden state ht_f at the current moment. The update process of the backward LSTM is similar to that of the forward LSTM, except that the processing order is reversed. Finally, for each semantic word unit mi, a corresponding forward hidden state ht_f and a backward hidden state ht_b can be obtained, thus forming a forward hidden state sequence H_f={h1_f, h2_f, …, hn_f} and a backward hidden state sequence H_b={h1_b, h2_b, …, hn_b}.

[0025] Step S1122: Concatenate the forward hidden state sequence and the backward hidden state sequence of the same semantic word unit to generate an initial context vector.

[0026] Concatenate the hidden states corresponding to the same semantic word unit in the forward hidden state sequence H_f and the backward hidden state sequence H_b. For example, for the i-th semantic word unit mi, concatenate its forward hidden state ht_f and backward hidden state ht_b together to obtain an initial context vector vt_0=[ht_f; ht_b], where ";" represents the concatenation operation. In this way, for all semantic word units, an initial context vector set V_0={v1_0, v2_0, …, vn_0} can be obtained.

[0027] Step S1123: Perform layer normalization on the initial context vector and fuse the normalized context vector with the initial context vector through a residual connection to obtain an enhanced context vector.

[0028] Layer Normalization normalizes each sample's feature dimensions. For each vector vt_0 in the initial context vector set V_0, first calculate its mean μt and variance σt, and then normalize it through the formula vt_n = (vt_0 - μt) / σt to obtain the normalized context vector vt_n. Then, through the residual connection, fuse the normalized context vector vt_n and the initial context vector vt_0 to enhance the context vector vt_e = vt_0 + vt_n. The role of the residual connection is to alleviate the vanishing gradient problem, enabling the model to better learn deep features. After processing, the enhanced context vector set V_e = {v1_e, v2_e,..., vn_e} is obtained.

[0029] Step S1124: Input the enhanced context vectors into the non-linear transformation layer to generate a context vector set with unified dimensions.

[0030] The non-linear transformation layer can use an activation function, such as the ReLU function. Input each vector vt_e in the enhanced context vector set V_e into the ReLU function, and the expression of the ReLU function is f(x) = max(0, x). Through the action of the ReLU function, set the negative part of the input vector to 0 and retain the positive part, thereby introducing non-linear features. At the same time, to ensure the unified dimensions of the context vectors, a linear transformation can also be performed after the non-linear transformation layer. Through a weight matrix W and a bias vector b, transform the vector to obtain the final context vector vt_c = W * vt_e + b. Finally, the context vector set V_c = {v1_c, v2_c,..., vn_c} with unified dimensions is obtained.

[0031] Step S113: Input the context vectors into the multi-head attention layer to calculate the semantic association weights between each semantic word unit.

[0032] The multi-head attention layer consists of multiple attention heads. For each attention head, first, the context vector set V_c is respectively passed through three linear transformation matrices W_q, W_k, and W_v to obtain the query vector set Q, the key vector set K, and the value vector set V. For each query vector qi, calculate its similarity score with all key vectors kj. The similarity score can be obtained through the dot product operation, that is, sij = qi * kj. Then, the similarity score is scaled and smoothed by dividing it by a scaling factor √d_k (where d_k is the dimension of the key vector), and then passing through the softmax function to obtain the attention weight αij, αij = softmax(sij / √d_k). Finally, the value vectors are weighted and summed according to the attention weights to obtain the output vector oi of each attention head. The output vectors of all attention heads are concatenated and then passed through a linear transformation matrix W_o to obtain the final output vector of the multi-head attention layer. In this process, the attention weight αij represents the semantic association weight between each semantic word unit.

[0033] Step S114: Weightedly fuse the context vectors based on the semantic association weights to generate the global semantic features of the question sentence.

[0034] According to the calculated semantic association weights αij, the context vector set V_c is weightedly fused. For each context vector vi_c, it is weighted according to the corresponding attention weight αij. Here, the weighted fusion is to splice the weighted context vectors according to the set rules. Specifically, the weighted context vectors vi_c * αij are connected end to end in sequence to form a new vector. This new vector contains the information of all semantic word units and takes into account the semantic association between them, and finally obtains the global semantic feature vector v_g. Through this splicing method, the feature information of each semantic word unit can be better retained, avoiding the problems of information loss or feature blur that may occur during the addition process, so as to generate global semantic features that can accurately reflect the semantics of the question sentence.

[0035] Step S115: Input the global semantic features into the feature dimensionality reduction layer to extract the core intention features and auxiliary logical features as semantic understanding features; among them, the core intention features are used to represent the core goal of the user's question, and the auxiliary logical features are used to describe the dependency relationship and semantic hierarchical structure between multiple semantic word units in the question sentence.

[0036] The feature dimension reduction layer can adopt methods such as principal component analysis (PCA) or autoencoders. Taking PCA as an example, first calculate the covariance matrix C of the global semantic feature vector v_g, then perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors. Select the eigenvectors corresponding to the top k largest eigenvalues, project the global semantic feature vector into the subspace formed by these eigenvectors, and obtain the dimension-reduced feature vector. Among the dimension-reduced feature vectors, some features can be identified as core intention features, and some features can be identified as auxiliary logic features.

[0037] The core intention features can reflect the core goal of the user's question. For example, in the above question statement, the core goal may be to formulate financing and market expansion strategies. The auxiliary logic features describe the dependency relationships and semantic hierarchical structures between semantic units. For example, there is a conditional dependency relationship between "under the premise of controlling capital costs and risks" and "formulating financing and market expansion strategies". Finally, denote the core intention features as f_c, the auxiliary logic features as f_l, and the semantic understanding features F = {f_c, f_l}.

[0038] Step S120: Match the knowledge node set corresponding to the semantic understanding features based on a preset innovation and entrepreneurship counseling knowledge base.

[0039] Step S121: Traverse the candidate knowledge nodes in the innovation and entrepreneurship counseling knowledge base, and call the node matching model to calculate the first similarity score between the node features of each candidate knowledge node and the core intention features.

[0040] The innovation and entrepreneurship counseling knowledge base contains a large number of knowledge nodes, and each knowledge node has its corresponding node features. For each candidate knowledge node in the knowledge base, input its node features and the core intention feature f_c extracted in step S115 into the node matching model. The node matching model can be a neural network model based on deep learning, which learns the mapping relationship between the node features and the core intention features and outputs a similarity score. Assume that the node feature of the candidate knowledge node is nk, and the output function of the node matching model is S(nk, f_c), then the first similarity score s1 = S(nk, f_c).

[0041] Step S122: Screen out the candidate knowledge nodes with the first similarity score higher than the first preset threshold to form an initial matching set.

[0042] Set a first preset threshold t1, traverse the first similarity scores s1 of all candidate knowledge nodes, and screen out the candidate knowledge nodes with scores higher than t1 to form an initial matching set I = {nk1, nk2,..., nkm}, where nki represents the i-th candidate knowledge node that meets the conditions.

[0043] Step S123: Invoke the logical verification model to analyze whether the node description text of the candidate knowledge nodes in the initial matching set satisfies the dependency relationships and semantic hierarchical structures corresponding to the auxiliary logical features. If not, it is determined as a conflict node, and the conflict node is removed to obtain the candidate knowledge nodes after logical verification.

[0044] The logical verification model analyzes the node description text of each candidate knowledge node in the initial matching set I. The auxiliary logical feature f_l contains the dependency relationships and semantic hierarchical structure information between the semantic word units in the question statement. The logical verification model checks whether the description text of the candidate knowledge nodes conforms to these dependency relationships and semantic hierarchical structures. For example, if the auxiliary logical feature indicates an association between "financing strategy" and "market expansion strategy", and a certain candidate knowledge node only mentions "financing strategy" without reflecting the association with "market expansion strategy", then this node may be determined as a conflict node. The conflict nodes that do not meet the conditions are removed from the initial matching set I to obtain the set of candidate knowledge nodes after logical verification I_l = {nk1_l, nk2_l,..., nkp_l}.

[0045] Step S124: Arrange the candidate knowledge nodes after logical verification in descending order according to the first similarity score, and select the top N nodes as the optimized knowledge node set.

[0046] For each node in the set of candidate knowledge nodes I_l after logical verification, it is arranged in descending order according to its corresponding first similarity score s1. Then, the top N nodes are selected to form the optimized knowledge node set K = {nk1_k, nk2_k,..., nkN_k}.

[0047] Step S130: Generate semantic association paths based on the hierarchical relationships and semantic association degrees of the knowledge node set.

[0048] Step S131: Obtain the hierarchical attributes, association labels, and historical user interaction data of each node in the knowledge node set.

[0049] For each node nki_k in the optimized knowledge node set K, obtain its hierarchical attribute from the innovation and entrepreneurship tutoring knowledge base. The hierarchical attribute can represent the hierarchical position of the node in the knowledge base, such as a first-level node, a second-level node, etc. At the same time, obtain the association label of the node. The association label is used to describe the semantic association between the node and other nodes, such as labels like "financing", "market expansion", "risk control", etc. In addition, it is also necessary to obtain the historical user interaction data of this node. The historical user interaction data includes information such as the number of clicks, stay time, and subsequent questions of the user for this node.

[0050] Step S132: Construct the vertical hierarchical connection relationships between nodes according to the hierarchical attributes.

[0051] Construct the vertical hierarchical connection relationship between nodes according to the hierarchical attributes of knowledge nodes. If one node is the superior node of another node, there is a vertical connection between them. For example, in the knowledge base of innovation and entrepreneurship guidance, the "financing strategy" node may be the superior node of the "seed round financing strategy" node, and there is a vertical hierarchical connection between them. In this way, a vertical hierarchical connection graph Gl is constructed.

[0052] Step S133: Expand the horizontal semantic association relationship between nodes according to the associated tags.

[0053] Expand the horizontal semantic association relationship between nodes according to the associated tags of knowledge nodes. If two nodes have the same or related associated tags, there is a horizontal semantic association between them. For example, the "financing strategy" node and the "investment analysis" node both have associated tags related to "finance", and a horizontal semantic association can be established between them. In this way, a horizontal semantic association graph Gs is constructed.

[0054] Step S134: Generate an initial semantic network based on the vertical hierarchical connection relationship and the horizontal semantic association relationship.

[0055] Merge the vertical hierarchical connection graph Gl and the horizontal semantic association graph Gs to generate an initial semantic network G. In the initial semantic network G, there are both vertical hierarchical connection relationships and horizontal semantic association relationships between nodes.

[0056] Step S135: Invoke the path generation algorithm to traverse the initial semantic network, generate multiple candidate semantic paths, and score each candidate semantic path based on the path length, the node jump probability, the node semantic weight, and the user feedback index, combined with the historical user interaction data, and select the candidate semantic path with the highest score as the target semantic association path.

[0057] Step S1351: Starting from the core intention node in the knowledge node set, traverse the child nodes downward according to the vertical hierarchical connection relationship, and expand the sibling nodes according to the horizontal semantic association relationship at the same time.

[0058] Determine the core intention node in the knowledge node set K. For example, in the above question statement, the core intention node may be the "financing and market expansion strategy" node. Starting from this node, traverse all its child nodes downward according to the vertical hierarchical connection relationship, and expand its sibling nodes according to the horizontal semantic association relationship at the same time. During the traversal process, record the accessed node sequence and jump relationship to generate an initial path set P0.

[0059] Step S1352: Filter the redundant paths in the initial path set to obtain multiple candidate semantic paths.

[0060] Redundancy path filtering is performed on the initial path set P_0. First, paths that contain the same node sequence but have different jump orders are merged because these paths essentially express the same semantic information. Second, paths that contain loop nodes or repeated jumps are removed because these paths do not conform to normal semantic logic. After filtering, a set of multiple candidate semantic paths P = {p1, p2,..., pq} is obtained.

[0061] Step S1353: Obtain the path length of each candidate semantic path. The path length is obtained by counting the total number of all nodes in the candidate semantic path.

[0062] For each path pi in the candidate semantic path set P, count the total number of nodes it contains and use it as the path length li of this path.

[0063] Step S1354: Calculate the jump probability between nodes of the candidate semantic path. The jump probability between nodes is determined based on the ratio of the number of jumps between adjacent nodes to the total number of jumps of the starting node in the historical user interaction data.

[0064] According to the historical user interaction data, count the number of jumps between adjacent nodes and the total number of jumps of the starting node. For the adjacent node pair (nj, nk) in the candidate semantic path pi, calculate the jump probability pjk between them, where pjk = the number of jumps between adjacent nodes (nj, nk) / the total number of jumps of the starting node nj. For the entire candidate semantic path pi, the jump probabilities of all its adjacent node pairs can be comprehensively considered, such as taking the average value or the weighted average value, to obtain the jump probability pi_p between nodes of this path.

[0065] Step S1355: Extract the node semantic weights of the candidate semantic path. The node semantic weights read the preset weight values of each node from the innovation and entrepreneurship guidance knowledge base, and the weight values are calculated based on the node citation frequency and the semantic relevance marked by the user.

[0066] Read the preset weight values of each node in the candidate semantic path pi from the innovation and entrepreneurship guidance knowledge base. The weight value of a node is calculated based on the node citation frequency and the semantic relevance marked by the user. For example, if a node is frequently cited and has a high semantic relevance to the user's question, then its weight value will be relatively large. Comprehensively consider the weight values of all nodes in the path, such as taking the average value or the weighted average value, to obtain the node semantic weight pi_w of this path.

[0067] Step S1356: Obtain the user feedback index. The user feedback index is obtained by counting the average satisfaction score generated after the user clicks on all nodes in the candidate semantic path in the historical session and the matching degree of the associated subsequent questions.

[0068] Statistically calculate the average satisfaction score generated after all nodes in the candidate semantic path pi are clicked by the user in the historical conversation and the matching degree with the associated subsequent questions. The average satisfaction score can be calculated from the scoring data of the user on the node content, and the matching degree with the associated subsequent questions can be calculated by a semantic similarity model. Considering the average satisfaction score and the matching degree with the associated subsequent questions comprehensively, the user feedback index pi_f of this path is obtained.

[0069] Step S1357: After standardizing and transforming the path length, node transition probability, node semantic weight, and user feedback index of each candidate semantic path, perform weighted summation to generate the comprehensive score of each candidate semantic path.

[0070] Perform standardizing transformation on the path length li, node transition probability pi_p, node semantic weight pi_w, and user feedback index pi_f so that they are within the same dimensional range. The standardizing transformation can adopt common methods, such as Z-score standardization, that is, subtract the mean value of each eigenvalue and then divide by its standard deviation. Let the standardized path length be li_s, the node transition probability be pi_p_s, the node semantic weight be pi_w_s, and the user feedback index be pi_f_s. Then, assign a weight to each standardized feature, denoted as wl, wp, ww, and wf respectively. These weights can be adjusted according to the actual situation to reflect the importance of each feature in the comprehensive score. Through weighted summation, generate the comprehensive score si of each candidate semantic path, that is, si = wl * li_s + wp * pi_p_s + ww * pi_w_s + wf * pi_f_s.

[0071] Step S1358: Select the candidate semantic path with the highest comprehensive score as the target semantic association path.

[0072] Compare the comprehensive scores si of all candidate semantic paths, and select the candidate semantic path with the highest score as the target semantic association path p_t. This target semantic association path can best reflect the hierarchical relationship and semantic association degree between knowledge nodes.

[0073] Step S140: Determine the target Q&A strategy according to the semantic association path and generate optimized tutoring content.

[0074] Step S141: Analyze the node sequence in the semantic association path and identify the tutoring topic evolution logic and key knowledge point distribution corresponding to the node sequence.

[0075] Step S1411: Obtain the node attributes of each knowledge node in the semantic association path. The node attributes include node type labels, historical interaction frequencies, and preset semantic weights.

[0076] For each knowledge node in the target semantic association path p_t, obtain its node attributes from the innovation and entrepreneurship tutoring knowledge base. The node type labels can be divided into types such as "concept explanation", "strategy suggestion", "case analysis", etc., which are used to describe the main content nature of the node. The historical interaction frequency reflects the number of times the node has been accessed by the user in the historical conversation, and the preset semantic weight reflects the importance of the node in the knowledge base.

[0077] Step S1412: Divide the starting node, intermediate nodes, and ending node in the node sequence according to the node type label, and extract the hierarchical jump direction and semantic association type between adjacent nodes.

[0078] According to the node type label, divide the node sequence in the target semantic association path p_t into a starting node, intermediate nodes, and an ending node. The starting node is usually the node most relevant to the core intention of the user's question. The intermediate nodes are used to further elaborate and expand relevant content, and the ending node gives summary or conclusive information. At the same time, extract the hierarchical jump direction between adjacent nodes. For example, a jump from a superior node to a subordinate node indicates in-depth refinement of content, and a jump between peer nodes indicates horizontal association and expansion. The semantic association type can include "causal association", "parallel association", "progressive association", etc., which reflects the semantic logical relationship between nodes.

[0079] Step S1413: Construct a vertical evolution branch based on the hierarchical jump direction, generate a horizontal expansion branch according to the semantic association type, and topologically merge the vertical evolution branch and the horizontal expansion branch to generate an initial topic network.

[0080] Construct a vertical evolution branch according to the hierarchical jump direction between adjacent nodes. For example, if there is a jump from the "financing strategy" node to the "seed round financing strategy" node, a vertical evolution branch can be constructed. At the same time, generate a horizontal expansion branch according to the semantic association type. For example, if there is a parallel association between the "financing strategy" node and the "market expansion strategy" node, a horizontal expansion branch can be constructed. Topologically merge the vertical evolution branch and the horizontal expansion branch to form an initial topic network N_0 containing nodes and connection edges.

[0081] Step S1414: Traverse the node connection edges in the initial topic network, calculate the comprehensive strength value of each connection edge, and the comprehensive strength value is obtained by weighted summing the normalized value of the historical interaction frequency, the logarithmic conversion value of the preset semantic weight, and the priority coefficient of the association type.

[0082] For each node connection edge in the initial topic network N_0, calculate its comprehensive strength value. First, normalize the historical interaction frequency and convert it to a unified range. Then, perform a logarithmic transformation on the preset semantic weight to reduce the differences between weight values. The priority coefficient of the association type is preset according to the importance of different semantic association types. Perform a weighted sum of the normalized historical interaction frequency, the logarithmically transformed preset semantic weight, and the priority coefficient of the association type to obtain the comprehensive strength value of each connection edge. Let the historical interaction frequency be f, the preset semantic weight be w, the priority coefficient of the association type be c, and the weighting coefficients be wf, ww, and wc respectively. Then the comprehensive strength value s_c of the connection edge = wf * normalize(f) + ww * logarithmically transform(w) + wc * c.

[0083] Step S1415: Filter the connection edges above the strength threshold according to the comprehensive strength value to form a core connection edge set, and extract the node subsequence corresponding to the core connection edge set as the key knowledge point distribution.

[0084] Set a strength threshold t2, filter out the connection edges with a comprehensive strength value greater than t2 to form a core connection edge set E_c. The core connection edge set reflects the most important semantic association relationships between nodes. Extract the node subsequence corresponding to the core connection edge set. These node subsequences contain the most critical knowledge points and use them as the key knowledge point distribution K_d.

[0085] Step S1416: Based on the node subsequence of the core connection edge set, extract the temporal change pattern of the node type labels and the combination rules of the semantic association types to generate the tutoring topic evolution logic. The temporal change pattern is determined by analyzing the occurrence order and frequency distribution of the node type labels in the node subsequence. The combination rules are obtained by counting the co-occurrence frequency and conditional probability of the semantic association types in the node subsequence.

[0086] Analyze the node subsequence corresponding to the core connection edge set and extract the temporal change pattern of the node type labels. For example, there may be an order where a "concept explanation" node appears first, then a "strategy suggestion" node, and finally a "case analysis" node. At the same time, count the co-occurrence frequency and conditional probability of the semantic association types in the node subsequence to obtain the combination rules of the semantic association types. Integrate the temporal change pattern of the node type labels and the combination rules of the semantic association types to generate the tutoring topic evolution logic L_e.

[0087] Step S142: Match the basic Q&A strategy template from the preset Q&A strategy library according to the tutoring topic evolution logic. The basic Q&A strategy template includes content generation rules, interaction process design, and resource reference methods.

[0088] Step S1421: Extract the path attributes in the tutoring topic evolution logic. The path attributes include the number of path stages, the node type distribution, and the semantic association combination pattern.

[0089] Extract the path attributes from the tutoring topic evolution logic L_e. The number of path stages reflects the stage division in the tutoring topic evolution process. The node type distribution describes the distribution of different types of nodes in the path. The semantic association combination pattern reflects the combination method of the semantic association types between nodes.

[0090] Step S1422: Convert the number of path stages into a stage division vector, convert the node type distribution into a type density matrix, and convert the semantic association combination pattern into an association coding sequence.

[0091] Convert the number of path stages into a stage division vector. Each element of the vector represents the relevant information of each stage. Convert the node type distribution into a type density matrix. The rows of the matrix represent different node types, the columns represent different stages, and the matrix elements represent the density of the node type in that stage. Convert the semantic association combination pattern into an association coding sequence. Each element in the sequence represents the coding of a semantic association type.

[0092] Step S1423: Call the policy matching model to calculate the first matching degree between the stage division vector and the stage labels of each policy template in the preset Q&A policy library, the second matching degree between the type density matrix and the node compatibility matrix of each policy template, and the third matching degree between the association coding sequence and the association rule library of each policy template.

[0093] Input the stage division vector, the type density matrix, and the association coding sequence into the policy matching model respectively. The policy matching model will calculate the first matching degree between the stage division vector and the stage labels of each policy template in the preset Q&A policy library, the second matching degree between the type density matrix and the node compatibility matrix of each policy template, and the third matching degree between the association coding sequence and the association rule library of each policy template. Let the stage division vector be v_p, the stage label of the policy template be t_p, the type density matrix be M_t, the node compatibility matrix of the policy template be M_c, the association coding sequence be s_a, and the association rule library of the policy template be R_a. Then the first matching degree m1 = policy matching model(v_p, t_p), the second matching degree m2 = policy matching model(M_t, M_c), and the third matching degree m3 = policy matching model(s_a, R_a).

[0094] Step S1424: Standardize the first matching degree, the second matching degree, and the third matching degree, and perform weighted fusion based on the preset weight coefficients to generate the comprehensive adaptation score of each policy template.

[0095] Normalize the first matching degree m1, the second matching degree m2, and the third matching degree m3 so that they are within the same dimension range. Then, assign a preset weight coefficient to each normalized matching degree, denoted as wm1, wm2, and wm3 respectively. Through weighted fusion, generate the comprehensive adaptation score s_m of each policy template, that is, s_m = wm1 * normalize(m1) + wm2 * normalize(m2) + wm3 * normalize(m3).

[0096] Step S1425: Screen out the policy templates with comprehensive adaptation scores higher than the adaptation threshold to form a candidate policy set, and sort and optimize the candidate policy set according to the time series change pattern in the tutoring theme evolution logic. The sorting and optimization are achieved by matching the consistency of the time series change pattern with the stage evolution in the historical application scenarios of the candidate policy templates.

[0097] Set an adaptation threshold t3, screen out the policy templates with comprehensive adaptation scores higher than t3 to form a candidate policy set S_c. Then, sort and optimize the candidate policy set S_c according to the time series change pattern in the tutoring theme evolution logic L_e. Match the consistency of the time series change pattern with the stage evolution in the historical application scenarios of the candidate policy templates, and preferentially select the policy templates with high stage evolution consistency.

[0098] Step S1426: Select the policy template with the highest ranking in the sorted and optimized candidate policy set as the basic Q&A policy template.

[0099] Rank the policy templates in the sorted and optimized candidate policy set S_c, and select the policy template with the highest ranking as the basic Q&A policy template T_b. This basic Q&A policy template contains information such as content generation rules, interaction process design, and resource reference methods.

[0100] Step S143: Dynamically adjust the basic Q&A policy template based on the distribution of key knowledge points to obtain the adjusted basic Q&A policy template.

[0101] Step S1431: Identify the priority parameters of each knowledge node in the distribution of key knowledge points. The priority parameters are calculated based on the semantic weight of the knowledge node, the historical user interaction frequency, and the hierarchical attribute.

[0102] For each knowledge node in the key knowledge point distribution \(K_d\), calculate its priority parameter. The priority parameter is calculated based on the semantic weight of the knowledge node, the historical user interaction frequency, and the hierarchical attribute it belongs to. The semantic weight reflects the importance of the node in the knowledge base, the historical user interaction frequency reflects the popularity of the node, and the hierarchical attribute indicates the hierarchical position of the node in the knowledge base. The priority parameter can be calculated by weighted summation. Let the semantic weight be \(w\), the historical user interaction frequency be \(f\), and the hierarchical attribute be \(l\), and the weighting coefficients be \(ww\), \(wf\), and \(wl\), then the priority parameter \(p = ww*w + wf*f + wl*l\).

[0103] Step S1432: Divide the key knowledge point distribution into a core knowledge point set and an auxiliary knowledge point set according to the priority parameter, where the priority parameter of the core knowledge point set is higher than the preset priority threshold.

[0104] Set a preset priority threshold \(t4\), and divide the key knowledge point distribution \(K_d\) into a core knowledge point set \(K_c\) and an auxiliary knowledge point set \(K_a\) according to the priority parameter \(p\). The priority parameters of the knowledge nodes in the core knowledge point set are higher than \(t4\), and these nodes contain the most critical information. The priority parameters of the knowledge nodes in the auxiliary knowledge point set are lower than \(t4\), and are used to further supplement and expand the core knowledge points.

[0105] Step S1433: Extract the content generation rules from the basic Q&A strategy template, set the content generation rules corresponding to the core knowledge point set as the default activation module, and generate extended trigger conditions according to the semantic correlation degree of the auxiliary knowledge point set.

[0106] Extract the content generation rules from the basic Q&A strategy template \(T_b\). Set the content generation rules corresponding to the core knowledge point set \(K_c\) as the default activation modules, and these modules will be called first when generating optimized tutoring content. According to the semantic correlation degree of the auxiliary knowledge point set \(K_a\), generate extended trigger conditions. For example, if there is a causal relationship between an auxiliary knowledge point and a core knowledge point, then when the core knowledge point is mentioned and the set conditions are met, the content generation of the auxiliary knowledge point is triggered.

[0107] Step S1434: Analyze the sequence of unvisited knowledge points in the user's historical interaction data, calculate the matching degree between the sequence of unvisited knowledge points and the auxiliary knowledge point set, and screen out the unvisited knowledge points with a matching degree higher than the preset matching threshold as the insertion nodes of supplementary cases.

[0108] Analyze the user's historical interaction data to extract the sequence of unvisited knowledge points U. Calculate the matching degree between the sequence of unvisited knowledge points U and the set of auxiliary knowledge points K_a, which can be calculated through a semantic similarity model. Set a preset matching threshold t5, and filter out the unvisited knowledge points with a matching degree higher than t5, which are used as the insertion nodes I_n of supplementary cases.

[0109] Step S1435: According to the auxiliary logic feature in the semantic understanding feature, determine the logical embedding position of the insertion node of the supplementary case in the basic Q&A strategy template. The logical embedding position is obtained by matching the dependency relationship in the auxiliary logic feature with the sequence of interaction process nodes in the template.

[0110] According to the auxiliary logic feature f_l in the semantic understanding feature F, determine the logical embedding position of the insertion node I_n of the supplementary case in the basic Q&A strategy template T_b. By matching the dependency relationship in the auxiliary logic feature with the sequence of interaction process nodes in the template, find a suitable embedding position. For example, if the auxiliary logic feature indicates that there is a sequential relationship between a certain knowledge point and another knowledge point, then the insertion node of the supplementary case should be inserted at the corresponding position.

[0111] Step S1436: Integrate the parameters of the default activation module, extended trigger conditions, and logical embedding position to generate an adjusted basic Q&A strategy template.

[0112] Integrate the parameters of the default activation module, extended trigger conditions, and logical embedding position to generate an adjusted basic Q&A strategy template T_a. This template is dynamically adjusted according to the distribution of key knowledge points on the basis of the basic Q&A strategy template, and can better meet the needs of users.

[0113] Step S144: Generate optimized tutoring content according to the adjusted basic Q&A strategy template; the optimized tutoring content includes step-by-step guidance text, reference cases adapted to user needs, and extended resource links for related knowledge points.

[0114] Generate optimized tutoring content C according to the adjusted basic Q&A strategy template T_a. The optimized tutoring content includes step-by-step guidance text, which provides detailed guidance for users according to the evolution logic of the tutoring theme and the distribution of key knowledge points. At the same time, it includes reference cases adapted to user needs, which can come from historical successful cases or actual application scenarios. In addition, it also provides extended resource links for related knowledge points to facilitate users to further study relevant knowledge in depth.

[0115] Step S150: Update the node connection weights and association relationships in the innovation and entrepreneurship tutoring knowledge base according to the feedback data of users on the optimized tutoring content.

[0116] Step S151: Collect the interaction behavior data of the user and the optimization tutoring content. The interaction behavior data includes the content click position, the number of case consultations, the resource link opening rate, and subsequent question statements.

[0117] By monitoring the interaction process between the user and the optimization tutoring content C, collect the interaction behavior data. The content click position reflects the user's focus on different contents, the number of case consultations reflects the user's interest in cases, the resource link opening rate indicates the user's utilization of extended resources, and the subsequent question statements reflect the problems and further needs encountered by the user during the learning process.

[0118] Step S152: Clean and structurally process the interaction behavior data to generate a feedback data set.

[0119] Clean the collected interaction behavior data to remove the noise data and invalid data. Then, structurally process the cleaned data, convert it into a format easy to analyze, and generate a feedback data set D.

[0120] Step S153: Calculate the contribution score of each knowledge node in the semantic association path according to the feedback data set.

[0121] Step S1531: Extract the interaction index set of each knowledge node from the feedback data set. The interaction index set includes the click-through rate increment, the case consultation duration, the change value of the resource link opening rate, and the subsequent question matching degree.

[0122] Extract the interaction index set of each knowledge node from the feedback data set D. The click-through rate increment represents the increase in the click-through rate of this knowledge node in the current interaction relative to the historical click-through rate. The case consultation duration reflects the time the user spends consulting the case corresponding to this knowledge node. The change value of the resource link opening rate indicates the change in the resource link opening rate corresponding to this knowledge node. The subsequent question matching degree is calculated through a semantic similarity model and reflects the semantic relevance between the user's subsequent questions and this knowledge node.

[0123] Step S1532: Perform time decay weighting processing on the click-through rate increment to generate a time-corrected click-through rate. The time decay weighting processing is calculated based on the interval between the click behavior occurrence time and the current time and a preset decay factor.

[0124] Perform time decay weighting processing on the click-through rate increment to consider the timeliness of the click behavior. According to the interval between the click behavior occurrence time and the current time and the preset decay factor, weight the click-through rate increment. Let the click-through rate increment be i_c, the interval between the click behavior occurrence time and the current time be t, and the preset decay factor be α. Then the time-corrected click-through rate i_c_t = i_c * decay function(t, α), where the decay function can be an exponential decay function, etc.

[0125] Step S1533: Convert the case access duration into a duration distribution percentile, convert the change value of the resource link opening rate into a change intensity coefficient, and calculate the matching degree score of subsequent questions through a semantic similarity model.

[0126] Converting the case access duration into a duration distribution percentile means comparing the case access duration of this knowledge node with that of all knowledge nodes to obtain its percentile in the duration distribution. The change value of the resource link opening rate is converted into a change intensity coefficient by normalizing the change value. The matching degree score of subsequent questions is calculated through a semantic similarity model, and the semantic similarity model can learn the semantic relationship between the question statement and the content of the knowledge node based on a deep learning algorithm.

[0127] Step S1534: Normalize the time-corrected click-through rate, duration distribution percentile, change intensity coefficient, and matching degree score to obtain a standardized index set.

[0128] Normalize the time-corrected click-through rate \(i_c_t\), duration distribution percentile \(p_t\), change intensity coefficient \(c_r\), and matching degree score \(s_m\) so that they are in the same dimension range. Methods such as Z-score standardization can be used for normalization to obtain the standardized index set \(I_s = \{i_c_t_s, p_t_s, c_r_s, s_m_s\}\).

[0129] Step S1535: Assign weights to each index in the standardized index set according to the preset contribution weights, and generate an initial contribution degree score through linear weighting.

[0130] Assign a preset contribution weight to each index in the standardized index set \(I_s\), denoted as \(w_i\), \(w_p\), \(w_c\), and \(w_s\) respectively. Through linear weighting, generate the initial contribution degree score \(s_i\), that is, \(s_i = w_i * i_c_t_s + w_p * p_t_s + w_c * c_r_s + w_s * s_m_s\).

[0131] Step S1536: Perform path correction on the initial contribution degree score based on the position attribute of the knowledge node in the semantic association path. The path correction is achieved by calculating the weight coefficient of the path level where the node is located and the jump distance between the node and the core intention node.

[0132] Based on the position attribute of the knowledge node in the semantic association path p_t, the initial contribution score s_i is corrected for the path. Calculate the weight coefficient of the path level where the node is located. The higher the level of the node, the greater the weight coefficient. At the same time, calculate the jump distance between the node and the core intention node. The closer the jump distance, the greater the weight coefficient of the node. Let the weight coefficient of the path level where the node is located be wl, the jump distance between the node and the core intention node be d, and the weight coefficient of the jump distance be wd. Then the score s_f after path correction can be obtained by weighting the initial contribution score s_i, that is, s_f = s_i * (wl + wd / (d + 1)). Here, adding 1 is to avoid the denominator being 0 when the jump distance is 0. The score s_f after path correction is used as the final contribution score of the knowledge node.

[0133] Step S154: For the nodes with contribution scores higher than the second preset threshold, increase the connection weight between this node and the node corresponding to the core intention feature.

[0134] Set the second preset threshold t6. For the nodes in the semantic association path with contribution scores higher than t6, increase the connection weight between this node and the node corresponding to the core intention feature. In the innovation and entrepreneurship counseling knowledge base, there is a corresponding weight for the connection between each node, which reflects the association strength between them. By increasing the connection weight, the association between these nodes and the core intention can be strengthened, making it easier to be selected in the subsequent matching process. For example, if a knowledge node has a high contribution score regarding "risk control in seed round financing" and the node corresponding to the core intention feature is "financing and market expansion strategy", then increase the connection weight between these two nodes.

[0135] Step S155: For the nodes with contribution scores lower than the third preset threshold, reduce the connection weight of this node or remove the invalid connection.

[0136] Set the third preset threshold t7. For the nodes with contribution scores lower than t7, reduce the connection weight of this node or remove the invalid connection according to the specific situation. If a node has a very low contribution in user interaction, it means that its association with the user's needs is not strong. By reducing the connection weight, its priority in the matching process can be reduced. If the connection of this node with other nodes has no effect at all, that is, it belongs to an invalid connection, then remove it to optimize the structure of the knowledge base. For example, if a knowledge node about "niche marketing methods for early startups" has a very low contribution score, the connection weight between it and other nodes can be reduced. If it is found that the connection between it and the core business related node has never been used, then remove this connection.

[0137] Step S156: Create a new node in the innovation and entrepreneurship counseling knowledge base according to the new semantic understanding features extracted from the user's subsequent question statement and establish an association relationship with the existing nodes.

[0138] Process the subsequent question statements of the user, and extract new semantic understanding features according to the methods in steps S110 - S115. If the content represented by the new semantic understanding features does not have corresponding nodes in the existing innovation and entrepreneurship counseling knowledge base, then create new nodes in the knowledge base. Establish an association relationship between the new nodes and the existing nodes according to the semantic association between the new semantic understanding features and the existing nodes. For example, if the subsequent question of the user involves the "integrated application of blockchain technology in artificial intelligence entrepreneurship", and it is found that there are no relevant nodes in the knowledge base for the extracted new semantic understanding features, then create a new node "Integrated Entrepreneurship Application of Artificial Intelligence and Blockchain", and establish connection relationships with existing nodes such as "Artificial Intelligence Entrepreneurship" and "Blockchain Technology" according to the semantic association.

[0139] Furthermore, the training method of the semantic encoding model includes the following steps: Step S210: Collect sample question statements and their corresponding marked intention labels in historical innovation and entrepreneurship counseling scenarios, and construct a training data set.

[0140] In historical innovation and entrepreneurship counseling scenarios, collect a large number of sample question statements. These sample question statements can come from historical interaction records between users and the counseling system, entrepreneurship-related questions in online forums, etc. At the same time, mark corresponding intention labels for each sample question statement. The intention labels can be "Financing Strategy Consultation", "Market Expansion Method Inquiry", etc. Combine the sample question statements and their corresponding marked intention labels to construct a training data set. During the construction process, ensure the diversity and representativeness of the data, covering different types of innovation and entrepreneurship problems.

[0141] Step S220: Aim to minimize the cross-entropy loss between the predicted intention label and the marked intention label of the sample question statement, and optimize the parameters of the semantic encoding model.

[0142] Input the training data set into the semantic encoding model. The semantic encoding model will process the sample question statements and output predicted intention labels. Calculate the cross-entropy loss between the predicted intention label and the marked intention label. The cross-entropy loss reflects the degree of difference between the prediction result and the true result. Aim to minimize the cross-entropy loss, and use an optimization algorithm (such as the stochastic gradient descent algorithm) to update the parameters of the semantic encoding model. In each round of training, the model will adjust the parameters according to the gradient direction of the loss function, making the loss gradually decrease. After multiple rounds of training, until the performance of the model reaches stability, obtain the trained semantic encoding model.

[0143] Furthermore, the training method of the node matching model includes the following steps: Step S310: Extract positive sample knowledge nodes from the historically successfully matched Q&A data, and record the semantic understanding features of the user questions corresponding to the positive sample knowledge nodes.

[0144] Filter out positive sample knowledge nodes from the historically successfully matched Q&A data. These nodes can match well with the user questions. At the same time, record the semantic understanding features of the user questions corresponding to each positive sample knowledge node. The semantic understanding features are extracted according to the methods in steps S110 - S115. The positive sample knowledge nodes and the corresponding semantic understanding features form the positive sample data set.

[0145] Step S320: Process the positive sample knowledge nodes to generate negative sample knowledge nodes.

[0146] Process the positive sample knowledge nodes to generate negative sample knowledge nodes. There are two specific methods. One is to retain the hierarchical attributes and associated labels of the positive sample knowledge nodes, and replace the core keywords of the positive sample knowledge nodes to generate node description texts with semantic contradictions. For example, if the positive sample knowledge node is "Equity financing strategy for innovation and entrepreneurship", "equity financing" can be replaced with "debt financing" to generate a node description with semantic contradiction. Another method is to select nodes that are semantically irrelevant to the user questions from the same tutoring field as negative samples. The generated negative sample knowledge nodes and the corresponding semantic understanding features of the user questions form the negative sample data set.

[0147] Step S330: Call the initialized node matching model to calculate the positive sample similarity between the node features of the positive sample knowledge nodes and the corresponding semantic understanding features, and the negative sample similarity between the node features of the negative sample knowledge nodes and the semantic understanding features.

[0148] Input the node features of the positive sample knowledge nodes and the corresponding semantic understanding features into the initialized node matching model to calculate the positive sample similarity. Similarly, input the node features of the negative sample knowledge nodes and the semantic understanding features into the initialized node matching model to calculate the negative sample similarity. The initialized node matching model can be a simple neural network model, which will learn the mapping relationship between the node features and the semantic understanding features and output a similarity score.

[0149] Step S340: Construct a contrastive loss function to maximize the difference between the positive sample similarity and the negative sample similarity, and update the parameters of the initialized node matching model through the backpropagation algorithm until the initialized node matching model converges to obtain the trained node matching model.

[0150] Construct a contrastive loss function. The goal of the contrastive loss function is to maximize the difference between the similarity of positive samples and the similarity of negative samples. Through the backpropagation algorithm, the parameters of the initialized node matching model are updated according to the gradient of the loss function. In each round of training, the model adjusts the parameters to make the similarity of positive samples as large as possible and the similarity of negative samples as small as possible. This process is repeated continuously until the performance of the model reaches stability, that is, the value of the loss function no longer decreases significantly. At this time, the initialized node matching model converges, and a trained node matching model is obtained.

[0151] During the data collection process of this embodiment, the principles of legality, legitimacy, and necessity are strictly followed to fully protect the user's right to know and right of choice, and ensure that the collection of all data obtains the user's explicit authorization. For example, in the initial stage of the user's interaction with the system, the purpose, method, scope of data collection, and the use and protection of data can be explained to the user in detail through a clear and easy-to-understand privacy policy and usage agreement. The user is informed that the collected data includes question statements, interaction behavior data (such as content click location, case access times, resource link opening rate, and subsequent question statements), and these data will be used to optimize the system's question-answer matching service to provide the user with more accurate and personalized innovation and entrepreneurship guidance. At the same time, the user is clearly informed that they have the right to withdraw the authorization at any time, and withdrawing the authorization will not affect the normal use of the basic functions of the system.

[0152] When the user registers or first uses the system, a pop-up window or checkbox consent is used to request the user to authorize the data collection matters. Only when the user clearly agrees will the system start collecting relevant data. For the collection of privacy-sensitive data (such as a small amount of possible user background information), the user's special authorization will be obtained separately to ensure that the user clearly understands the purpose and potential risks of the data. In addition, a perfect authorization management mechanism is established to record the user's authorization information and authorization time. If the system needs to adjust the scope, method, or purpose of data collection, the user will be notified in a timely manner and the user's authorization will be obtained again. At the same time, the user's authorization status is regularly checked to ensure the effectiveness and legality of the authorization.

[0153] In addition, during the data collection process, for the handling of privacy-sensitive data, multiple technical means are adopted for privacy protection and anti-disclosure. First, the collected user data is encrypted, and an advanced encryption algorithm (such as the AES encryption algorithm) is used to convert the data into ciphertext form for storage and transmission. In terms of data storage, a secure database system is adopted, and strict access permissions are set, and only authorized personnel can access the data. At the same time, the data is regularly backed up to prevent data loss. During the data usage process, data desensitization technology is adopted to desensitize sensitive information (such as user names, contact information, etc.), and only non-sensitive information related to the business is retained. For example, the user name is replaced with a number, and the contact information is partially masked. During the model training process, technologies such as federated learning are adopted, so that the data does not need to be centrally stored and processed, but the model is trained on local devices, and only the model parameters are transmitted, thus avoiding the leakage of privacy data. Through these technical means, the security and confidentiality of user privacy-sensitive data are ensured.

[0154] Figure 2 FIG. shows a schematic diagram of exemplary hardware and software components of an innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding that can implement the idea of the present application provided by some embodiments of the present application. For example, the processor 120 can be used on the innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding and is used to execute the functions in the present application.

[0155] The innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding can be a general-purpose server or a special-purpose server, both of which can be used to implement the innovation and entrepreneurship tutoring Q&A matching method based on semantic understanding of the present application. Although only one server is shown in the present application, for convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0156] For example, the innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding can include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as disks, ROM, or RAM, or any combination thereof. Exemplarily, the innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding can also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. According to these program instructions, the method of the present application can be implemented. The innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding also includes an I / O interface 150 between the computer and other input / output devices.

[0157] For ease of explanation, only one processor is described in the innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding. However, it should be noted that the innovation and entrepreneurship tutoring Q&A matching system 100 in the present application may also include multiple processors. Therefore, the steps performed by one processor described in the present application may also be jointly performed or separately performed by multiple processors. For example, if the processor of the innovation and entrepreneurship tutoring Q&A matching system 100 based on semantic understanding performs step A and step B, it should be understood that step A and step B may also be jointly performed by two different processors or separately performed in one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0158] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned innovation and entrepreneurship tutoring Q&A matching method based on semantic understanding is implemented.

[0159] It should be noted that, in order to simplify the expression of the disclosure of the present invention and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. An innovation and entrepreneurship counseling Q&A matching method based on semantic understanding, characterized in that The method includes: Extracting semantic understanding features of the question statement input by the user; Matching a set of knowledge nodes corresponding to the semantic understanding features based on a preset innovation and entrepreneurship counseling knowledge base; Generating a semantic association path according to the hierarchical relationship and semantic association degree of the set of knowledge nodes; Determining a target Q&A strategy according to the semantic association path and generating optimized counseling content; Updating the node connection weights and association relationships in the innovation and entrepreneurship counseling knowledge base according to the feedback data of the user on the optimized counseling content.

2. The method for matching innovation and entrepreneurship counseling questions and answers based on semantic understanding according to claim 1, wherein The extracting of the semantic understanding features of the question statement input by the user includes: Performing word segmentation on the question statement, splitting it into multiple semantic word units and removing stop words in the multiple semantic word units; Invoking a pre-trained semantic encoding model to perform context encoding on the semantic word units to generate context vectors for each semantic word unit; Inputting the context vectors into a multi-head attention layer to calculate the semantic association weights between the semantic word units; Performing weighted fusion on the context vectors based on the semantic association weights to generate the global semantic features of the question statement; Inputting the global semantic features into a feature dimensionality reduction layer to extract core intention features and auxiliary logic features as the semantic understanding features; wherein, the core intention features are used to represent the core goal of the user's question, and the auxiliary logic features are used to describe the dependency relationship and semantic hierarchical structure between multiple semantic word units in the question statement; Among them, the semantic encoding model is trained through the following steps: Collecting sample question statements and their corresponding marked intention labels in historical innovation and entrepreneurship counseling scenarios to construct a training data set; Taking the minimization of the cross-entropy loss between the predicted intention label and the marked intention label of the sample question statement as the goal to optimize the parameters of the semantic encoding model.

3. The method for matching innovation and entrepreneurship counseling questions and answers based on semantic understanding according to claim 2, characterized in that, The invoking of the pre-trained semantic encoding model to perform context encoding on the semantic word units to generate context vectors for each semantic word unit includes: Inputting the semantic word units into a bidirectional long short-term memory network to respectively obtain the forward hidden state sequence and the backward hidden state sequence of each semantic word unit; Concatenating the forward hidden state sequence and the backward hidden state sequence of the same semantic word unit to generate an initial context vector; Performing layer normalization processing on the initial context vector and fusing the normalized context vector with the initial context vector through a residual connection to obtain an enhanced context vector; Inputting the enhanced context vector into a non-linear transformation layer to generate a set of context vectors with unified dimensions.

4. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 2, wherein The matching of a set of knowledge nodes corresponding to the semantic understanding features based on a preset innovation and entrepreneurship counseling knowledge base includes: Traversing the candidate knowledge nodes in the innovation and entrepreneurship counseling knowledge base, and invoking a node matching model to calculate the first similarity score between the node features of each candidate knowledge node and the core intention features; Screening the candidate knowledge nodes with the first similarity score higher than the first preset threshold to form an initial matching set; Call the logical verification model to analyze whether the node description text of the candidate knowledge nodes in the initial matching set satisfies the dependency relationship and semantic hierarchy corresponding to the auxiliary logical features. If not, it is determined as a conflict node, and the conflict node is removed to obtain the candidate knowledge nodes after logical verification; Sort the candidate knowledge nodes after logical verification in descending order according to the first similarity score, and select the top N nodes as the optimized knowledge node set.

5. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 4, wherein The training method of the node matching model includes: Extract positive sample knowledge nodes from the historically successfully matched question-and-answer data, and record the user question semantic understanding features corresponding to the positive sample knowledge nodes; Process the positive sample knowledge nodes to generate negative sample knowledge nodes, specifically including: retaining the hierarchical attributes and associated labels of the positive sample knowledge nodes, replacing the core keywords of the positive sample knowledge nodes to generate node description texts with semantic contradictions, or selecting nodes irrelevant to the user question semantics from the same tutoring field; Call the initialized node matching model to calculate the positive sample similarity between the node features of the positive sample knowledge nodes and the corresponding semantic understanding features, and the negative sample similarity between the node features of the negative sample knowledge nodes and the semantic understanding features; Construct a contrast loss function to maximize the difference between the positive sample similarity and the negative sample similarity, and update the parameters of the initialized node matching model through the backpropagation algorithm until the initialized node matching model converges to obtain the trained node matching model.

6. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 1, wherein The generation of the semantic association path according to the hierarchical relationship and semantic association degree of the knowledge node set includes: Obtain the hierarchical attributes, associated labels and historical user interaction data of each node in the knowledge node set; Construct the vertical hierarchical connection relationship between nodes according to the hierarchical attributes; Expand the horizontal semantic association relationship between nodes according to the associated labels; Generate an initial semantic network based on the vertical hierarchical connection relationship and the horizontal semantic association relationship; Call the path generation algorithm to traverse the initial semantic network, generate multiple candidate semantic paths, and score each candidate semantic path based on the path length, node jump probability, node semantic weight and user feedback index, combined with the historical user interaction data, and select the candidate semantic path with the highest score as the target semantic association path.

7. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 6, characterized in that The call to the path generation algorithm to traverse the initial semantic network, generate multiple candidate semantic paths, and score each candidate semantic path based on the path length, node jump probability, node semantic weight and user feedback index, combined with the historical user interaction data, and select the candidate semantic path with the highest score as the target semantic association path includes: Start from the core intention node in the knowledge node set, traverse the child nodes downward according to the vertical hierarchical connection relationship, and expand the sibling nodes according to the horizontal semantic association relationship at the same time; Record the accessed node sequence and jump relationship during the traversal process to generate an initial path set; Perform redundant path filtering on the initial path set to obtain multiple candidate semantic paths, specifically including: merging paths with the same node sequence but different jump orders, and removing paths containing loop nodes or repeated jumps; Obtain the path length of each candidate semantic path, where the path length is obtained by counting the total number of all nodes in the candidate semantic path; Calculate the jump probability between nodes of the candidate semantic path, where the jump probability between nodes is determined based on the ratio of the number of jumps between adjacent nodes to the total number of jumps of the starting node in historical user interaction data; Extract the node semantic weights of the candidate semantic path, where the node semantic weights read the preset weight values of each node from the innovation and entrepreneurship tutoring knowledge base, and the weight values are calculated based on the citation frequency of the node and the semantic relevance marked by the user; Obtain the user feedback index, where the user feedback index is obtained by counting the average satisfaction score generated after the user clicks on all nodes in the candidate semantic path in the historical session and the matching degree of the associated subsequent questions; Perform standardized conversion on the path length, jump probability between nodes, node semantic weights, and user feedback index of each candidate semantic path, and then perform weighted summation to generate the comprehensive score of each candidate semantic path; Select the candidate semantic path with the highest comprehensive score as the target semantic association path.

8. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 1, characterized in that The determining the target Q&A strategy and generating optimized tutoring content according to the semantic association path includes: Analyze the node sequence in the semantic association path and identify the tutoring theme evolution logic and key knowledge point distribution corresponding to the node sequence; Match the basic Q&A strategy template from the preset Q&A strategy library according to the tutoring theme evolution logic, where the basic Q&A strategy template includes content generation rules, interaction process design, and resource reference methods; Dynamically adjust the basic Q&A strategy template based on the key knowledge point distribution to obtain the adjusted basic Q&A strategy template; Generate the optimized tutoring content according to the adjusted basic Q&A strategy template; the optimized tutoring content includes step-by-step guidance text, reference cases adapted to user needs, and extended resource links for associated knowledge points.

9. The method for matching innovation and entrepreneurship tutoring questions and answers based on semantic understanding according to claim 8, characterized in that, The dynamically adjusting the basic Q&A strategy template based on the key knowledge point distribution to obtain the adjusted basic Q&A strategy template includes: Identify the priority parameters of each knowledge node in the key knowledge point distribution, where the priority parameters are calculated based on the semantic weights of the knowledge nodes, historical user interaction frequencies, and hierarchical attributes; Divide the key knowledge point distribution into a core knowledge point set and an auxiliary knowledge point set according to the priority parameters, where the priority parameters of the core knowledge point set are higher than the preset priority threshold; Extract the content generation rules in the basic Q&A strategy template, set the content generation rules corresponding to the core knowledge point set as the default activation module, and generate extended trigger conditions according to the semantic association degree of the auxiliary knowledge point set; Analyze the sequence of unvisited knowledge points in the user's historical interaction data, calculate the matching degree between the sequence of unvisited knowledge points and the set of auxiliary knowledge points, and screen out the unvisited knowledge points with a matching degree higher than the preset matching threshold as the insertion nodes for supplementary cases; According to the auxiliary logic features in the semantic understanding features, determine the logical embedding position of the insertion nodes of the supplementary cases in the basic Q&A strategy template, and the logical embedding position is obtained by matching the dependency relationship in the auxiliary logic features with the sequence of interaction process nodes in the basic Q&A strategy template; Integrate the parameters of the default activation module, extended trigger conditions and logical embedding positions to generate an adjusted basic Q&A strategy template.

10. An innovation and entrepreneurship tutoring Q&A matching system based on semantic understanding, characterized in that, It includes a processor and a memory. The memory is connected to the processor. The memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the semantic understanding-based innovation and entrepreneurship tutoring Q&A matching method according to any one of claims 1-9 above.

Images