Question recommendation method, device and equipment

Abstract

The invention discloses a question recommendation method, device and equipment. The method comprises the following steps: firstly, acquiring a keyword set and word frequency of each keyword in the keyword set in a target court trial record based on the target court trial record; a plurality of feature vectors are retrieved from a pre-constructed co-occurrence matrix based on the keyword set; in the co-occurrence matrix, two dimensions of the matrix respectively represent problems and keywords. And performing feature enhancement processing on the feature vectors corresponding to the corresponding keywords by using the word frequency to obtain a plurality of enhanced feature vectors, and performing summation operation on the plurality of enhanced feature vectors to obtain a result vector. And finally, generating a question recommendation list of the target court trial record according to numerical values in the result vector. Problem recommendation is realized through an automatic technical means, the efficiency and accuracy of extracting problems based on text materials in application scenes such as court trial investigation are improved in an auxiliary manner, and the dependence on individual experience and ability of manual question asking is reduced.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a question recommendation method, apparatus, and device. Background Art

[0002] An arbitration tribunal is a temporary organization tasked with resolving disputes between two parties. It handles a wide range of arbitration cases, including civil and commercial arbitration, labor dispute arbitration (also known as labor arbitration), agricultural contract disputes, and maritime arbitration. During the arbitration hearing process, the tribunal must conduct questioning based on written materials, known as the in-court investigation. This phase is a crucial stage in which arbitrators proactively investigate the facts and legal issues of the case. Its significance goes far beyond simply asking questions; it is a core procedure that directly influences the course of the case and the outcome of the arbitration. It embodies the combination of arbitration's "flexibility" and "inquisitorial" nature, and is a crucial guarantee for ensuring the fairness and efficiency of arbitration.

[0003] The following uses labor arbitration as an example. Labor arbitration specifically addresses disputes between employers and employees arising from labor rights and obligations, such as confirmation of the labor relationship, payment of labor remuneration, and work-related injury benefits. This type of arbitration is a prerequisite for labor dispute litigation; directly filing a lawsuit in court is generally not possible without arbitration. Labor arbitration is a crucial component of the human resources and social security system and the preferred means of resolving labor disputes. A well-functioning labor arbitration system can quickly and effectively protect vulnerable groups at critical moments, effectively safeguard the legitimate rights and interests of workers and employers, reduce unnecessary civil disputes arising from labor disputes, and alleviate some of the pressure on the People's Courts. The hearing process of a labor arbitration trial includes presentation and rebuttal, cross-examination of evidence, questioning by the arbitration tribunal (in-court investigation), and debate between the two parties. However, in practice, the in-court investigation phase faces numerous difficulties.

[0004] First, the in-court investigation phase places high demands on the arbitrator's ability to formulate questions. Arbitrators must thoroughly study and accurately grasp the arbitration application, defense, and evidence submitted by both parties, carefully analyzing them and identifying core questions with substantive impact on the award. This process not only requires the arbitrator to possess a high level of professionalism but also requires a significant investment of time and effort, making it a profound test of their cognitive abilities.

[0005] Furthermore, manual questioning has certain limitations. Current arbitration procedures rely entirely on manual questioning by arbitrators, and their effectiveness is closely tied to the arbitrators' individual experience. Due to varying levels of experience among arbitrators, key questions are inevitably overlooked, potentially leading to the omission of important facts and, in turn, compromising the fairness and accuracy of the award.

[0006] Furthermore, there is a severe shortage of experienced arbitrators. In recent years, with the growing awareness of the rule of law among working people, the number of labor arbitration cases has been on the rise, which has put increasing pressure on arbitrators. Therefore, there is an urgent need for auxiliary means to alleviate pressure and improve efficiency. Summary of the Invention

[0007] Based on the above problems, this application provides a question recommendation method, device and equipment, the purpose of which is to achieve question recommendation through automated technical means, to assist in improving the efficiency and accuracy of extracting questions based on text materials in application scenarios such as court trial investigations, and to reduce dependence on individual experience and ability of manual questioning.

[0008] The embodiments of this application disclose the following technical solutions:

[0009] In a first aspect, the present application provides a question recommendation method, the method comprising:

[0010] Obtaining a keyword set and a word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript;

[0011] A plurality of feature vectors are retrieved from a pre-constructed co-occurrence matrix based on the keyword set; the plurality of feature vectors correspond one-to-one to each keyword in the keyword set; the co-occurrence matrix is a two-dimensional matrix, wherein a first dimension represents questions and a second dimension represents keywords; wherein the questions represented by the first dimension of the co-occurrence matrix are question sentences extracted from a collection of historical court trial transcripts, and the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the collection of historical court trial transcripts;

[0012] Performing feature enhancement processing on feature vectors corresponding to corresponding keywords using the word frequencies to obtain multiple enhanced feature vectors;

[0013] Performing a sum operation on the multiple enhanced feature vectors to obtain a result vector;

[0014] A list of recommended questions for the target court trial transcript is generated according to the values in the result vector.

[0015] A second aspect of the present application provides a question recommendation device, the device comprising:

[0016] A first acquisition module is configured to acquire a keyword set and a word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript;

[0017] A retrieval module is configured to retrieve a plurality of feature vectors from a pre-constructed co-occurrence matrix based on the keyword set; the plurality of feature vectors correspond one-to-one to each keyword in the keyword set; the co-occurrence matrix is a two-dimensional matrix, wherein a first dimension represents questions and a second dimension represents keywords; wherein the questions represented by the first dimension of the co-occurrence matrix are question sentences extracted from a collection of historical court trial transcripts, and the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the collection of historical court trial transcripts;

[0018] An enhancement module, configured to perform feature enhancement processing on feature vectors corresponding to corresponding keywords using the word frequency to obtain a plurality of enhanced feature vectors;

[0019] A summing module, configured to perform a sum operation on the plurality of enhanced feature vectors to obtain a result vector;

[0020] A recommendation module is used to generate a list of recommended questions for the target trial transcript based on the values in the result vector.

[0021] A third aspect of the present application provides a question recommendation device, the device comprising: a memory and a processor;

[0022] The memory is used to store computer programs;

[0023] The processor is used to run the computer program, and when the computer program is running, the steps of the question recommendation method introduced in the first aspect are executed.

[0024] Compared with the existing technology, this application has the following beneficial effects:

[0025] In the question recommendation method proposed in the present application, with the help of a pre-constructed co-occurrence matrix reflecting the association between keywords and questions, the retrieval of feature vectors in the co-occurrence matrix can be realized by only extracting the keywords of the target trial transcript. In the retrieved feature vector, the historical co-occurrence relationship between each question and the keyword is reflected in a numerical value. Based on the retrieved feature vector, the feature enhancement processing of word frequency is performed to strengthen the importance of the word frequency of the keywords in the target trial transcript, and enhance the specific role of the word frequency in the target trial transcript in question recommendation. On this basis, the result vector is obtained by combining and summing multiple enhanced feature vectors. In the result vector, the closeness of the connection between each question and the target trial transcript can be systematically reflected by numerical values, and then the recommendation of questions specifically for the target trial transcript can be completed logically among multiple questions. The present application realizes question recommendation through automated technical means, which helps to improve the efficiency and accuracy of refining questions based on text materials in application scenarios such as trial investigations, and reduces the dependence on the experience and ability of individuals who ask manual questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0027] Figure 1 A flowchart of a question recommendation method provided in an embodiment of the present application;

[0028] Figure 2 An architectural diagram of a question recommendation method provided in an embodiment of the present application;

[0029] Figure 3 This is a flow chart of the data preparation phase in the early stage of co-occurrence matrix construction according to an embodiment of the present application;

[0030] Figure 4 A schematic diagram of constructing a co-occurrence matrix provided in an embodiment of the present application;

[0031] Figure 5 An architectural diagram of another question recommendation method provided in an embodiment of the present application;

[0032] Figure 6 A flowchart of an embodiment of the present application providing a method for optimizing a co-occurrence matrix using a PageRank enhancement mechanism;

[0033] Figure 7 A schematic diagram of the structure of a question recommendation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] In scenarios such as project management, market research and analysis, policy formulation and evaluation, community building and management, and arbitration, it is often necessary to carefully study materials and refine the questions that need to be asked. In arbitration, for example, arbitrators are required to combine multiple written materials to extract core issues that will have a substantive impact on the award. This places high demands on the arbitrators' personal abilities and work experience. However, the current number of experienced arbitrators is insufficient to meet the growing demand for arbitration, and manual questioning has certain limitations. Therefore, manually completing the arbitration hearing investigation stage generally faces challenges in terms of execution efficiency, accuracy, and cost.

[0035] In the face of the above problems, the inventors have proposed a fully automatic question recommendation method after research. With the help of the co-occurrence of questions and keywords displayed by the co-occurrence matrix constructed with historical trial transcript data, when faced with the need to ask questions in a new trial transcript, it is only necessary to extract keywords and the word frequency of keywords in the new trial transcript from the new trial transcript, and then multiple feature vectors can be retrieved from the co-occurrence matrix. Subsequently, the corresponding feature vectors are enhanced using the word frequency, and the multiple enhanced feature vectors are summed up, and finally a new question recommendation list for the trial transcript is generated based on the numerical value reflected in the result vector. This method does not rely on manual experience and ability, and can be executed automatically. It uses historical trial transcript data as a knowledge reserve and uses the co-occurrence matrix to efficiently and accurately recommend questions for the trial materials. In turn, it effectively assists the arbitration tribunal in asking questions.

[0036] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0037] See also Figure 1 and Figure 2 , Figure 1 This is a flowchart of a question recommendation method provided in an embodiment of the present application. Figure 2 This is an architectural diagram of a question recommendation method provided in an embodiment of the present application. Figure 1 Show the recommendation process with a sequential relationship from top to bottom. Figure 2 The implementation architecture for generating a list of recommended questions for a new court transcript is shown from left to right.

[0038] S101. Obtain a keyword set and the word frequency of each keyword in the keyword set in the target court trial transcript based on the target court trial transcript.

[0039] The target court transcript can be understood as a newly received court transcript. In this application scenario, question recommendations are required for the target transcript. In this step, key features are extracted from the target transcript, including keywords and their frequency. Since the number of extracted keywords is often more than one, a keyword set is defined to specifically contain the keywords extracted from the target transcript.

[0040] When this step is specifically implemented, as a first optional implementation method, it is possible to combine a pre-built keyword library, extract words that match the keyword library from the target trial transcript, and add them to the keyword set of the target trial transcript. In the second optional implementation method, considering that there may be semantically similar connections between words of different expressions, therefore, when pre-building the keyword library, it is possible to further construct a mapping relationship between similar words and their representative words, and further derive a representative word set corresponding to the keyword library based on the above mapping relationship on the basis of the keyword library. When executing this step, combined with the mapping relationship between similar words and representative words, as well as the specific vocabulary in the representative word library, the target trial transcript is queried for matching representative words, and the matching representative words are added to the keyword set of the target trial transcript.

[0041] The word frequency of keywords can be obtained through data statistics. If the number of words in the text material is relatively small, the word frequency of the keywords will be very high compared to the text material with a larger number of words, thus causing the contribution of this trial transcript to the construction of the co-occurrence matrix to be greater than that of other trial transcripts. For this reason, the present application proposes that in the stage of constructing and applying the co-occurrence matrix, the number of occurrences of keywords is not used as the word frequency, but the ratio of the number of occurrences of a keyword in a trial transcript to the sum of the number of occurrences of all keywords in the trial transcript is used as the word frequency of the keyword in the trial transcript. By analogy, the word frequency of each keyword in the target trial transcript in the keyword set of the target trial transcript is obtained. Through this standardized processing method, the influence of the number of words in the text material on the application of word frequency is weakened.

[0042] Similarly, considering that there may be semantically similar connections between words of different expressions, for the second implementation method of obtaining the keyword set mentioned above, when counting the keyword frequency, the sum of the similar words that have a mapping relationship with a representative word in the target trial transcript and the number of occurrences of the representative word can be used as the numerator for the standardized calculation of the representative word frequency, and then the keyword frequency is obtained through division. Thus, through this implementation method, and with the help of the mapping relationship between similar words and representative words, the frequency of each representative word in the keyword set of the target trial transcript can be obtained. To facilitate the index, these statistical word frequencies can be displayed in the form of a word frequency list.

[0043] The purpose of obtaining keywords in this step is to retrieve the feature vector corresponding to each word in the keyword set from the pre-constructed co-occurrence matrix, thereby reflecting the close connection between each question and the word through the feature vector. The purpose of obtaining the word frequency of each word in the keyword set in this step is to enhance the specific impact of these words in the target trial materials, so that they can help more accurately recommend questions that are relevant to the scenario. The following first describes the process of retrieving feature vectors in conjunction with step S102; the application of word frequency is further explained in step S103.

[0044] S102 : Retrieve a plurality of feature vectors from a pre-built co-occurrence matrix based on the keyword set.

[0045] In an embodiment of the present application, the co-occurrence matrix is a two-dimensional matrix, in which the first dimension represents questions and the second dimension represents keywords. Here, the first dimension and the second dimension are used to distinguish the two different dimensions of the co-occurrence matrix: rows and columns. As an example, the first dimension is the rows of the matrix, and the second dimension is the columns of the matrix. Then, each row of the co-occurrence matrix corresponds to a question, and each column corresponds to a keyword. In another example, the first dimension is the columns of the matrix, and the second dimension is the rows of the matrix. Then, each column of the co-occurrence matrix corresponds to a question, and each row corresponds to a keyword. In the embodiments below, in order to facilitate understanding and description, an example in which the first dimension is the rows of the matrix and the second dimension is the columns of the matrix is used for introduction. It can be understood that when the first dimension is the columns of the matrix and the second dimension is the rows of the matrix, the technical concept expressed in the technical solution of the present application can still be adaptively applied.

[0046] A co-occurrence matrix is constructed by analyzing the co-occurrence of keywords and questions based on a large number of historical trial transcripts. This matrix reflects the close connection between keywords and questions in historical trial transcripts. The questions represented by the first dimension of the co-occurrence matrix are question statements extracted from the historical trial transcript dataset; the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the historical trial transcript dataset. The larger the value of an element at a certain position in the co-occurrence matrix, the closer the connection between the corresponding keyword and the question. In this application, the co-occurrence matrix, a dataset based on historical data that reflects the close connection between keywords and questions, can be used to recommend appropriate questions for new trial transcripts (i.e., target trial transcripts), assisting the arbitration tribunal's trial investigation process. Overall, the co-occurrence matrix is an efficient and convenient query tool for this application and also a treasure trove of data.

[0047] For ease of understanding, the construction process of the co-occurrence matrix is introduced below.

[0048] First, a dataset of historical court trial transcripts (containing multiple historical trial transcripts) is obtained. Next, questions and keywords are extracted from each historical trial transcript to construct a set of questions and a set of keywords corresponding to each transcript. The word frequency of each keyword in the keyword set in the corresponding historical trial transcript is counted to construct keyword-frequency pairs corresponding to each transcript. Furthermore, a representative word set corresponding to the historical trial transcript dataset is constructed based on the semantic similarity between each constructed keyword set and different keywords. Furthermore, a representative question set corresponding to the historical trial transcript dataset is constructed based on the semantic similarity between each constructed question set and different questions. Based on the semantic similarity between different keywords and the representative word set, the keyword-frequency pairs corresponding to each historical trial transcript are corrected to obtain the corrected keyword-frequency pairs for each transcript. A co-occurrence matrix is constructed based on the representative word set, the representative question set, and the corrected keyword-frequency pairs corresponding to each transcript. This process is described in detail below.

[0049] Figure 3 This is a flow chart of the data preparation phase in the early stage of building the co-occurrence matrix in the embodiment of the present application. Figure 3 As can be seen in the figure, keywords can be extracted from the applicant's application, the respondent's defense and the arbitration tribunal's cross-examination in the court transcript to form a keyword library; questions can be extracted from the arbitrators' questions in the court investigation to form a question library.

[0050] As an example, we retrieved 4,919 historical court transcripts from arbitration cases in a specific region and extracted questions and keywords from each. A large language model can be applied to each transcript to extract keywords relevant to the question. For example, the prompt reads: "[Instructions] The materials contain the applicant's application, the respondent's defense, and the cross-examination of evidence from a court transcript. Please identify keywords that are important to the subsequent court investigation. For example, if the question involves 'Will the applicant's nursing expenses be paid?' and the document contains words related to "nursing expenses," these keywords are identified as keywords. [Materials]... [Court Investigation Questions]..." This constructs a keyword set corresponding to each historical court transcript. The keyword sets from each historical court transcript are deduplicated and merged to form a keyword library.

[0051] Extract all the questions the arbitrators asked the applicant and the respondent from the trial investigation section of each historical trial transcript, segment them, remove stop words, and retain only the core semantics. This step can also be achieved by designing prompt words and calling a large language model for each historical trial transcript. The prompt word content is: "[Instruction] I want to convert the question list into vector form, but I need you to help me process the questions in advance, including removing stop words, as well as irrelevant information such as time, place, name and amount. If the text is empty, please output 'none' [Requirements] 1. If a question contains several sub-questions, it must be broken down and refined, and the sub-questions must be complete (for example: When did the applicant resign? Why?: 'Resignation date', 'Reason for resignation'). All questions must be output as a list Don't make up questions or output anything other than the list; [Example] Question 1 {When did the applicant start working for the respondent and how did he / she do so?} Question 2 {Did the applicant request compensatory leave for 2018 in 2019?}: ['Joining employment date', 'Joining employment method', 'Request for paid leave'] [Text]...". This constructs the question set corresponding to the historical trial transcript. Optionally, the question set can be further divided into the applicant's question set and the respondent's question set. After de-duplicating and merging the question sets of each historical trial transcript, a question library is formed, which can be further divided into the applicant's question library and the respondent's question library.

[0052] Given that some documents are short on words, their keywords often have high frequency, causing these transcripts to contribute more to the co-occurrence matrix than others. Therefore, this solution uses a standardized frequency formula: keyword frequency = number of occurrences of the keyword / total number of occurrences of all keywords. This ensures that each transcript contributes 1 to the matrix. Suppose a document contains the following keywords: "labor" 3 times; "dispute" 2 times; "arbitration" 1 time. The standardized frequency calculation is as follows:

[0053]

[0054]

[0055]

[0056] When constructing the matrix, the semantic similarity between keywords and the semantic similarity between questions also need to be considered. By analyzing semantic similarity, it is possible to establish semantic connections between similar questions and representative questions, and to establish semantic connections between similar words and representative words, thus avoiding the diversified presentation of questions with similar semantics and keywords with similar semantics, which would interfere with the efficient construction of the co-occurrence matrix or increase the complexity of the co-occurrence matrix application. For keywords, the keyword library can be vectorized, such as Figure 3 As shown in the figure, the K-means clustering analysis method is used: all similar words are clustered, and then the optimal cluster number K is determined by the elbow rule. Finally, each cluster and its center are obtained, and then the mapping relationship between similar words and representative words is constructed. Regarding the problem, compared with words, the semantics of sentences are more complex, and the category boundaries are fuzzy, which is not suitable for forced clustering. Therefore, the following can be used for the problem: Figure 3 The cosine similarity threshold classification method shown in the figure ultimately determines the mapping relationship between each representative question and its similar questions. Based on the above mapping relationship, the keyword library mentioned above can be optimized to the representative word set corresponding to the historical court trial transcript dataset, and the question library mentioned above can be optimized to the representative question set corresponding to the historical court trial transcript dataset.

[0057] Of course, before optimizing the keyword and question libraries based on semantic similarity in the previous phase, it was already possible to traverse and obtain keyword and word frequency combinations (i.e., keyword-frequency pairs) for each historical trial transcript. After optimization, these keyword-frequency pairs can be modified to avoid the diverse presentation of semantically similar questions and keywords, which could interfere with the accuracy of the keyword-frequency pairs and improve the accuracy of the co-occurrence matrix to be constructed. The logic for modifying keyword-frequency pairs still relies on the semantic similarity between different keywords. For example, a representative word A has two similar words: similar word B and similar word C. Previously, the keyword-frequency pair for representative word A in a historical trial transcript was calculated to be: Representative word A: 0.12, the keyword-frequency pair for similar word B: 0.1, and the keyword-frequency pair for similar word C: 0.15. 0.12, 0.1, and 0.15 all represent word frequencies. Through modification, the resulting keyword-frequency pair for this historical trial transcript is: Representative word A: 0.37. In other words, the word frequencies of the representative word and its similar words are combined as the accurate word frequencies of the representative word in the historical trial transcript, and then the keyword-word frequency pairs of the representative word and its similar words are corrected.

[0058] After the above preliminary preparations are completed, a co-occurrence matrix can be constructed based on the representative word set, the representative question set, and the revised keyword-frequency pairs corresponding to each historical trial transcript. Figure 4The schematic diagram of constructing the co-occurrence matrix provided in the embodiment of the present application is shown below in combination with Figure 4 Introduce the form of the constructed co-occurrence matrix. Figure 4 The following diagram simply shows the local effects of the intermediate stage during the construction of the co-occurrence matrix, showing only the values of four rows and four columns. Specifically, Figure 4 The right side shows four different representative questions: question 1, question 2, question 3, and question 4; and four different representative words: word 1, word 2, word 3, and word 4. Figure 4 The left side shows the representative questions of the problem set mapping corresponding to a historical trial transcript, including: Question 2 and Question 4. In addition, Figure 4 The left side also shows the revised keyword-word frequency pairs of the historical trial transcript, including word 1: 0.4 and word 3: 0.6.

[0059] In an optional implementation, a co-occurrence matrix is constructed based on the representative word set, the representative question set, and the corrected keyword-frequency pairs corresponding to each historical trial transcript, including:

[0060] Construct an initial matrix with each representative question in the representative question set as the first dimension index, each representative word in the representative word set as the second dimension index, and the elements at each position set to 0. Since in this example, the first dimension of the co-occurrence matrix is the row of the matrix and the second dimension is the column of the matrix, the first dimension index can refer to the row index and the second dimension index can refer to the column index. Figure 4 Example, Figure 4 The right side of shows only the matrix effect after traversing the first historical trial transcript. Before that, the matrix effects of Questions 1 to 4 and Words 1 to 4 in the initial matrix should be that all elements in each position of the four rows and four columns are set to 0.

[0061] Traverse each historical trial transcript in the historical trial transcript dataset, and based on the semantic similarity between different questions, convert the question set corresponding to the currently traversed historical trial transcript into a first representative question set, and based on the semantic similarity between different keywords, convert the keyword set corresponding to the currently traversed historical trial transcript into a first representative word set. It can be understood that the purpose of the above conversion process is to determine which representative questions and representative words are involved in the historical trial transcript. Through these conversion operations, the keywords and questions of the traversed historical trial transcript are aligned with the row and column reference content of the established initial matrix. Figure 4 The words 1 and 3 shown belong to the first representative word set of the historical court transcript. Figure 4 Questions 2 and 4 shown belong to the first representative question set of this historical trial transcript.

[0062] Use the first representative question set and the first representative word set to query the elements at the corresponding position of the current initial matrix, and use the word frequency of the corrected keyword-word frequency pair corresponding to the currently traversed historical trial transcript to superimpose the values of the elements at the corresponding position of the current initial matrix to complete an update of the initial matrix. Figure 4 Example, Figure 4 The historical trial transcript shown shows that the revised keyword-frequency pairs include word 1: 0.4 and word 3: 0.6. And through the above steps, it is known that the first representative question set only includes questions 2 and 4. Therefore, the second row and first column position (2,1) = 0, the second row and third column position (2,3) = 0, the fourth row and first column position (4,1) = 0, and the fourth row and third column position (4,3) = 0 can be located from the current initial matrix. For the two positions (2,1) and (4,1) in the first column, since the first column corresponds to word 1 and the word frequency of word 1 is 0.4, the word frequency of 0.4 is superimposed on the two positions (2,1) and (4,1) of the current initial matrix; for the two positions (2,3) and (4,3) in the third column, since the third column corresponds to word 3 and the word frequency of word 3 is 0.6, the word frequency of 0.6 is superimposed on the two positions (2,3) and (4,3) of the current initial matrix. Thus, it is shown that Figure 4 The matrix effect is shown on the right.

[0063] In an optional implementation, the first representative question set corresponding to the traversed historical trial records includes m representative questions, and the first representative word set includes n representative words; both m and n are positive integers. The following example illustrates the process of a single matrix update. Traverse the first representative question set and the first representative word set. For the i-th representative question in the first representative question set and the j-th representative word in the first representative word set, query the target position (i, j) in the current initial matrix, i.e., position (i, j) in the i-th row and j-th column, and obtain the current value of the element at the target position; i ranges from 1 to m, and j ranges from 1 to n, both i and j being integers. Determine the word frequency of the j-th representative word in the first representative word set in the revised keyword-word frequency pair corresponding to the currently traversed historical trial transcript. Add the word frequency of the j-th representative word to the current value of the element at the target position to complete the numerical update of the element at the target position. After traversing both the first representative question set and the first representative word set, complete the single update of the initial matrix.

[0064] It should be noted that Figure 4The matrix shown here only shows the results after traversing the first historical trial transcript. This is an intermediate stage during the co-occurrence matrix construction and does not represent the final co-occurrence matrix. After traversing all historical trial transcripts in the historical trial transcript dataset, the final updated initial matrix is used as the co-occurrence matrix. The implementation of adding word frequencies to the current matrix during each traversal can be referred to the example process described above and will not be repeated here.

[0065] It's understandable that by traversing the historical court trial transcript dataset, the matrix was updated multiple times. Therefore, the resulting co-occurrence matrix is based on the data of all the historical trial transcripts in the historical court trial transcript dataset. By presenting these historical trial transcripts, it generally reflects the closeness of the connection between these representative questions and representative words. For the same row of the co-occurrence matrix, the higher the value, the closer the relationship between the representative word corresponding to that position and the representative question in that row; and for the same column of the co-occurrence matrix, the higher the value, the closer the relationship between the representative question corresponding to that position and the representative word in that column.

[0066] Based on the co-occurrence matrix that has been constructed, data indexing can be completed using each word in the keyword set obtained in S101. For each word in the keyword set, a vector is indexed from the co-occurrence matrix. Since each column in the co-occurrence matrix corresponds to a representative word, the data on the column corresponding to the word can be used as an element in a feature vector. As an example, if there are five representative words in the keyword set, five feature vectors can be extracted, and the extracted feature vectors correspond one-to-one to the representative words in the keyword set.

[0067] S103 , performing feature enhancement processing on feature vectors corresponding to corresponding keywords using word frequencies to obtain multiple enhanced feature vectors.

[0068] In this step, feature vectors are enhanced using word frequency, reinforcing the critical role of word frequency in the target trial transcripts. This process strongly correlates the enhanced feature vectors with the inherent characteristics of the target trial transcripts. Therefore, the enhanced feature vectors not only reflect the co-occurrence of questions and keywords in the historical trial transcript dataset, but also demonstrate the important contribution of keywords in the target trial transcripts to question extraction and recommendation.

[0069] In an optional implementation of this step, feature enhancement can be performed by multiplying the feature vector corresponding to the keyword by the keyword's frequency in the target court transcript to obtain an enhanced feature vector for the keyword. For example, if the target court transcript contains the keyword "nursing fee" with a frequency of 0.3 and the keyword "work-related injury insurance" with a frequency of 0.7, then the columns corresponding to "nursing fee" and "work-related injury insurance" are extracted from the co-occurrence matrix to construct two feature vectors, and the two feature vectors are multiplied by 0.3 and 0.7, respectively.

[0070] In practical applications, this step is not limited to the above-mentioned feature enhancement processing method of directly multiplying the feature vector by the keyword frequency. For example, feature enhancement processing can also be achieved by summing or other operations including summation, multiplication, or exponential operation.

[0071] S104: performing a sum operation on the multiple enhanced feature vectors to obtain a result vector.

[0072] Summing multiple enhanced eigenvectors essentially means adding the values of the elements at the same position of these vectors. The purpose is to reflect the matching of each question with the target trial transcript horizontally through the value of the elements at each position on the result vector based on the integrity of the target trial transcript.

[0073] S105: Generate a list of recommended questions for the target court trial transcript based on the values in the result vector.

[0074] In an optional implementation, this step can be specifically as follows: based on the one-to-one correspondence between the elements at each position of the result vector and the questions, determine the questions corresponding to the elements with positive values in the result vector; arrange the multiple questions determined in descending order according to the numerical values of the corresponding elements, and generate a recommended list of questions for the target trial transcript.

[0075] For example, the resulting vector from the summation operation is represented as [11 13 25 0 0 10], where each position corresponds to question 1, question 2, question 3, question 4, question 5, and question 6. This vector indicates that question 3 has the highest match with the target trial transcript and is therefore more suitable for recommendation as a question in the trial investigation phase. The resulting ranking is: ① Question 3; ② Question 2; ③ Question 1; ④ Question 6.

[0076] In the question recommendation method introduced in the above embodiment, with the help of a pre-constructed co-occurrence matrix reflecting the association between keywords and questions, only the keywords of the target trial transcript are extracted to realize the retrieval of feature vectors in the co-occurrence matrix. In the retrieved feature vector, the historical co-occurrence relationship between each question and the keyword is reflected in a numerical value. Based on the retrieved feature vector, the feature enhancement processing of word frequency is performed to strengthen the importance of the word frequency of the keywords in the target trial transcript, and enhance the specific role of the word frequency in the target trial transcript in question recommendation. On this basis, the result vector is obtained by combining and summing multiple enhanced feature vectors. In the result vector, the closeness of the connection between each question and the target trial transcript can be systematically reflected by numerical values, and then the recommendation of questions specifically for the target trial transcript can be completed logically among multiple questions. This application realizes question recommendation through automated technical means, assists in improving the efficiency and accuracy of refining questions based on text materials in application scenarios such as trial investigations, and reduces the dependence on the experience and ability of individuals who ask manual questions.

[0077] In step S101 of the above embodiment, it is mentioned that the frequency of keywords in the target court trial transcript is counted. The following describes a specific implementation of this process, namely, the process of normalizing the frequency of keywords.

[0078] In an optional implementation method, taking the first target keyword in the keyword set corresponding to the target trial transcript as an example: obtain the number of occurrences of the first target keyword in the keyword set in the target trial transcript, and obtain the number of occurrences of other keywords in the keyword set in the target trial transcript; use the sum of the number of occurrences of the first target keyword in the target trial transcript and the number of occurrences of other keywords in the target trial transcript as the denominator, use the number of occurrences of the first target keyword in the target trial transcript as the numerator, and use the obtained ratio as the word frequency of the first target keyword in the target trial transcript.

[0079] The above process describes the standardized calculation process of keyword frequencies in the target trial transcripts.

[0080] Similarly, the following describes the technical implementation of extracting keyword frequencies from historical trial transcripts in the historical trial transcript dataset, specifically the process of normalizing the frequency of keywords. The rationale for using this standardized method for calculating frequency has already been discussed above and will not be repeated here.

[0081] In an optional implementation method, taking the target historical trial transcript in the historical trial transcript dataset and the second target keyword in its keyword set as an example: obtain the number of occurrences of the second target keyword in the target historical trial transcript and the number of occurrences of other keywords in the target trial transcript in the keyword set corresponding to the target historical trial transcript in the historical trial transcript dataset; use the sum of the number of occurrences of the second target keyword in the target historical trial transcript and the number of occurrences of other keywords in the target historical trial transcript as the denominator, and use the number of occurrences of the second target keyword in the target historical trial transcript as the numerator, and use the obtained ratio as the word frequency of the second target keyword in the target historical trial transcript to construct a keyword-word frequency pair for the second target keyword.

[0082] The above process describes the standardized calculation process of keyword frequency in historical trial transcripts.

[0083] The following continues to focus on word frequency and introduces the method of correcting the keyword-word frequency pairs corresponding to historical court trial transcripts. The correction of keyword-word frequency has been introduced in the relevant description of S102 above, which is mainly based on the mapping relationship between representative words and their similar words. The real meaning of the correction lies in integrating the word frequencies of semantically similar keywords. In an optional implementation method, based on the semantic similarity between different keywords and the set of representative words, the keyword-word frequency pairs corresponding to each historical court trial transcript are corrected, including:

[0084] If the second target keyword is one of multiple representative words in the representative word set, similar keywords with similar word mapping relationships to the second target keyword are determined in the keyword set corresponding to the target historical trial transcript; the word frequency of the similar keyword in the target historical trial transcript is added to the word frequency of the second target keyword in the target historical trial transcript to obtain the sum of the first word frequencies, and the sum of the first word frequencies is used to correct the keyword-word frequency pair of the second target keyword. This means that the word frequencies of the similar words are added to correct the word frequency of the second target keyword.

[0085] If the second target keyword is not one of the multiple representative words in the representative word set, then the target representative word corresponding to the second target keyword in the representative word set is determined based on the similar word mapping relationship; the word frequency of the second target keyword in the target historical trial transcript, the word frequency of the target representative word in the target historical trial transcript, and the word frequency of other keywords with similar word mapping relationships to the target representative word in the target historical trial transcript are added together to obtain the sum of the second word frequencies, and the sum of the second word frequencies is used to correct the keyword-word frequency pair of the target representative word corresponding to the target historical trial transcript. This means that the word frequency of the similar word is added to the word frequency of its representative word, and the final superposition result is used as the corrected word frequency of the representative word.

[0086] The above describes the correction operations performed for both cases, combining the cases where the second target keyword is a representative word and a non-representative word. Regardless of whether the second target keyword is a representative word or not, the original frequency count for the representative word can be corrected by superimposing it. This ensures the accuracy of the values in the constructed co-occurrence matrix and improves its effectiveness in question recommendation scenarios.

[0087] In actual application scenarios, some questions often have a fixed order of questioning. For example, "When did the applicant start working at the company being applied for?" and "What kind of work do you do?" are two questions asked in a sequence of one after the other. In the embodiment of the question recommendation method described above, the question order factor was not considered in the stage of generating the recommendation list. In this application, in order to make the order of questions presented in the question recommendation list more in line with people's questioning habits and language logic, it is proposed to introduce a sequence-aware weighted algorithm, and use the method of mining subsequences to weight the sequence elements into the scores of the questions in the question recommendation list. Specifically, the question sets of several historical trial transcripts can be first mapped to representative question sets based on the similarity of the question semantics. Optionally, these representative question sets can be mined using the PrefixSpan library in Python to extract subsequences of length 2 with a support greater than 500, and high-frequency binary question sequences are obtained. The first in each binary question sequence is called position 0, and the second is called position 1. For each binary question sequence, if both of the questions are in the question recommendation list generated by this solution, the element in position 1 is moved after position 0. Furthermore, when there are competing orders between two or more binary question sequences, the sequence with higher support is preferred.

[0088] Based on the above concept, the following describes an optional implementation of the question recommendation method. In an optional implementation, the question recommendation method further includes:

[0089] Obtain a historical court trial transcript dataset; the historical court trial transcript dataset includes multiple historical court trial transcripts. Extract questions from each historical court trial transcript, and construct a question set corresponding to each historical court trial transcript; based on the constructed question sets and the semantic similarity between different questions, construct a representative question set corresponding to the historical court trial transcript dataset. It should be noted that if the historical court trial transcript dataset has been obtained in this method and a representative question set has been constructed, this can be omitted here. Mining subsequences with a support that meets the preset conditions and a length of 2 from the representative question set, adding the mined subsequences as high-frequency binary question sequences to the mining sequence set; the question priority of the first question in the high-frequency binary question sequence is higher than the question priority of the second question in the latter position. The preset conditions can be set according to actual needs. For example, the preset conditions can be set to a support greater than 500. There is no limit on the support threshold of the preset conditions here.

[0090] After generating the list of recommended questions for the target trial transcript, the method further includes:

[0091] If there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question with the higher recommendation ranking is the first question in the target high-frequency binary question sequence, and the question with the lower recommendation ranking is the second question in the target high-frequency binary question sequence, then the ranking of the two questions in the question recommendation list that both belong to the target high-frequency binary question sequence will not be adjusted. The target high-frequency binary question sequence can be any high-frequency binary question sequence in the mining sequence set. In this case, the ranking of the two questions in the current question recommendation list is consistent with their ranking in the target high-frequency binary sequence, which is consistent with people's questioning habits and language logic, and therefore no further adjustment is required.

[0092] However, if there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question with the higher recommendation order is the second question in the target high-frequency binary question sequence, and the question with the lower recommendation order is the first question in the target high-frequency binary question sequence, then in the question recommendation list, the question with the higher recommendation order will be adjusted to after the question with the lower recommendation order. In this case, the order of the two questions in the current question recommendation list is inconsistent with their order in the target high-frequency binary sequence, which is inconsistent with people's questioning habits and language logic. Therefore, the order of the two questions in the question recommendation list needs to be adjusted.

[0093] Combined with the introduction of the above embodiments, it is not difficult to know that the question recommendation method proposed in this application can not only facilitate the work of arbitrators, but also extract high-value questions to be asked for the target trial transcript in an automated manner. It can also add a sequence element to the recommended question list, so that when asking questions in the order in the question recommendation list, it can take into account the questioning habits and language logic, and achieve better application effects in the trial investigation link.

[0094] Based on this, the inventors have proposed three other optimization ideas for the co-occurrence matrix. By optimizing the co-occurrence matrix, the effect of question recommendation can be further improved. Figure 5 This is an architectural diagram of another question recommendation method provided in an embodiment of the present application.

[0095] Figure 5 The paper presents four phases: data preparation, co-occurrence matrix construction, question recommendation, and matrix optimization. Data preparation involves constructing a set of representative questions and a set of representative words based on a dataset of historical trial transcripts, along with semantic similarity between questions and keywords. Furthermore, during the data preparation phase, the frequency of each representative word in each historical trial transcript is counted to assist in the construction of the co-occurrence matrix in the next phase. The co-occurrence matrix construction phase combines the data obtained during the data preparation phase and iterates through each historical trial transcript, continuously updating the values of each element in the co-occurrence matrix. This traversal results in the co-occurrence matrix, which serves as a vector search tool in the next phase. In the question recommendation phase, keywords are extracted from the new trial transcripts and converted into representative words. The frequency of each word is then counted. For each extracted representative word, a feature vector is retrieved from the co-occurrence matrix. The feature vectors are then enhanced based on the corresponding frequency. After this step, the enhanced feature vectors are summed to obtain a result vector. The questions are ranked based on the numerical values in the result vector, ultimately forming a list of recommended questions for the new trial transcript. This completes the recommendation of questions for the new trial transcript. Figure 5 The matrix optimization shown in can improve the matrix in many aspects, for example, the co-occurrence matrix can be optimized by one or more of sparsification, semantic similarity weighting, or PageRank enhancement mechanism.

[0096] The following is an introduction to the optimization implementation of the co-occurrence matrix. The above-mentioned sparsification processing, semantic similarity weighting or PageRank enhancement mechanism are mainly aimed at three optimization directions. The reasons for the optimization are as follows ①, ② and ③. ① In the constructed co-occurrence matrix, there may be low-frequency noise interference. Specifically, in the co-occurrence matrix, there may be a large number of low co-occurrence values (accidental co-occurrence of questions and keywords), which dilute the relevance between keywords and questions. ② The semantic granularity of the constructed co-occurrence matrix is coarse. Specifically, the previous co-occurrence matrix is mainly constructed based on word frequency statistics, and it is likely to be limited in its ability to capture the deep semantic connection between questions and materials. ③ Lack of importance transfer. Specifically, the global influence of the question vertex in the network is not considered in the constructed co-occurrence matrix.

[0097] Regarding point ②, in an optional implementation, after constructing the co-occurrence matrix, the problem recommendation method further includes:

[0098] A word semantic vector is extracted for each representative word in the set of representative words, and a question semantic vector is extracted for each representative question in the set of representative questions. This step represents the semantics of the representative words and questions in a vectorized manner, facilitating similarity calculation. This step can be implemented using a pre-trained BERT model. For example, a 768-dimensional question semantic vector is extracted for each representative question using the BERT model, and a 768-dimensional word semantic vector is extracted for each representative word using the BERT model.

[0099] Calculate the cosine similarity between each extracted word semantic vector and each question semantic vector, and construct a cosine similarity matrix based on the calculated cosine similarities. In the cosine similarity matrix, the first dimension represents the representative question, and the second dimension represents the representative word. In one example, if the first dimension of the matrix is the row of the matrix and the second dimension of the matrix is the column of the matrix, then each row in the cosine similarity matrix corresponds to a different representative question, and each column corresponds to a different representative word. In another example, if the first dimension of the matrix is the column of the matrix and the second dimension is the row of the matrix, then each column in the cosine similarity matrix corresponds to a different representative question, and each row corresponds to a different representative word. Regarding the definitions of the first dimension and the second dimension, they should be aligned in the cosine similarity matrix and the co-occurrence matrix to ensure the feasibility of subsequent matrix fusion operations.

[0100] The following is an introduction based on the previous example. In this step, the word semantic vector is compared with the question semantic vector, and the cosine similarity value reflects the semantic gap between the representative word and the representative vector. Then, the deep semantic relationship between the question and the material can be easily captured from the cosine similarity matrix. The element value s at the position of the i-th row and the j-th column in the cosine similarity matrix S is ijIt can be calculated by the following formula:

[0101]

[0102] Where q i represents the question semantic vector representing the question in row i, k j Represents the word semantic vector of the representative word in the jth column.

[0103] The co-occurrence matrix is fused with the cosine similarity matrix to obtain the first optimized co-occurrence matrix. Assuming that before the co-occurrence matrix is optimized, the co-occurrence matrix is represented as M and the cosine similarity matrix is represented as S, the process of fusing them to obtain the first optimized co-occurrence matrix M' can be expressed as:

[0104]

[0105] Where, Indicates the fusion weight coefficient of S, as an example =0.2. By calculating the cosine similarity between the question semantic vector and the word semantic vector, constructing a cosine similarity matrix, and fusing this cosine similarity matrix with the pre-optimized co-occurrence matrix, we can capture the deep semantic connections between the question and the material, making the semantic granularity of the first optimized co-occurrence matrix more refined. Testing has shown that this optimization strategy significantly improves the recall rate for long-tail questions.

[0106] Regarding point ①, in an optional implementation, after obtaining the first optimized co-occurrence matrix, the question recommendation method further includes:

[0107] The numerical values of the elements at each position in the first optimized co-occurrence matrix are numerically filtered using a preset threshold, elements with values lower than the preset threshold are reset to 0, and elements with values higher than or equal to the preset threshold are retained to obtain a second optimized co-occurrence matrix. This optimization strategy essentially implements threshold filtering on the elements at each position in the first optimized co-occurrence matrix. The preset threshold can be set according to actual needs. For example, if a stronger ability to resist low-frequency noise interference is required, a lower threshold can be set. As an example, the preset threshold is set to 1, which means that elements greater than or equal to 1 are retained, and the remaining elements less than 1 are reset to 0. In this way, the sparse processing of the co-occurrence matrix before optimization is completed, which effectively eliminates low-frequency noise interference and increases the accuracy of the recommended questions.

[0108] After optimizing the second optimized co-occurrence matrix, we can see that both issues 1 and 2 have been overcome. Therefore, when applying the co-occurrence matrix to feature vector retrieval for a new court transcript, we can specifically retrieve multiple feature vectors from the second optimized co-occurrence matrix based on the keyword set of the target court transcript (i.e., the new transcript). This will optimize subsequent vector retrieval and question recommendation performance.

[0109] Regarding point ③, through Figure 6 Demonstrate the matrix optimization process. Figure 6 The present invention provides a flowchart for optimizing the co-occurrence matrix using the PageRank enhancement mechanism. The PageRank algorithm is an algorithm used to evaluate the importance of Internet web pages. It determines the importance of web pages by analyzing the link structure between web pages. It is often used to sort search engine results and is very suitable for question recommendation scenarios. However, the traditional PageRank mechanism has two obvious limitations in question recommendation: first, the assumption of uniformly distributed edge weights cannot capture the strength of semantic associations between questions; second, it ignores the domain statistical characteristics of the nodes themselves. Figure 6 , which reveals the improvement strategies of the inventors in this application for the above two problems.

[0110] like Figure 6 As shown, in an optional implementation, the question recommendation method further includes:

[0111] S601. Based on the co-occurrence of two questions in each historical trial transcript, a question transfer matrix of the historical trial transcript dataset is constructed.

[0112] In the problem transfer matrix, both the first and second dimensions represent representative questions. That is, each row corresponds to a different representative question, and each column also corresponds to a different representative question. The representative question corresponding to the row and column of the diagonal elements of the problem transfer matrix is the same. The value of the element in the i-th row and j-th column of the problem transfer matrix represents the probability of transition from the i-th representative question to the j-th representative question.

[0113] In an optional implementation, this step may include:

[0114] Traverse the historical trial transcript dataset, count the co-occurring questions in each pair, and add the counts to the two symmetrical positions in the first transfer matrix associated with the co-occurring questions. This process continues until the traversal is completed, resulting in the second transfer matrix. The first transfer matrix is the initialized question transfer matrix, with all elements in each position set to 0. In the first transfer matrix, both the first and second dimensions represent a representative question, i.e., each row corresponds to a different representative question, and each column also corresponds to a different representative question. Each row element of the second transfer matrix is numerically normalized to obtain the question transfer matrix.

[0115] For example, for each pair of co-occurrence problems (q i , q j ), in the initialized question transfer matrix, the position (i, j) in the i-th row and j-th column and the position (j, i) in the j-th row and i-th column are counted +1. In this way, the co-occurrence relationship between two questions in the historical trial transcript dataset is searched, and the frequency of co-occurrence is reflected by the numerical size in the form of counts. After the second transfer matrix is numerically normalized, it is necessary to ensure that the sum of the values in each row is 1. In other words, the sum of the transition probabilities from each question to other questions is 1. Numerical normalization of a row of data is a mature technology in this field, so the specific process of normalization will not be described here.

[0116] S602: Based on the representative question set and the question transition matrix, construct a question co-occurrence graph with the representative questions as vertices and the transition probabilities as edge weights.

[0117] In this application, a problem co-occurrence graph G can be constructed based on a representative problem set of a historical court trial transcript data set. The vertex set in the graph can be represented by V, which is essentially a representative problem set. Each vertex in the vertex set V represents a representative problem in the representative problem set. As mentioned above, a problem transition matrix is constructed in S601, in which each position represents the transition probability from one problem to another. For example, the value at the position of the i-th row and the j-th column represents the problem q i To question q j The transition probability of the jth row and the ith column represents the problem q j To question q i The problem co-occurrence graph G constructed in this application is a directed graph. The edge between two vertices has a directional direction. The transition probability is used as the edge weight of the directed edge between the vertices in the problem co-occurrence graph G. For example, the problem q i The vertex points to the representative problem q j The edge of the vertex, its edge weight w ij Equal to the value of the element at the i-th row and j-th column in the problem transfer matrix.

[0118] Traditional PageRank defines edge weights as uniformly distributed (all outgoing edge weights = 1 / out-degree), which ignores domain knowledge that certain question combinations require joint questioning. However, this approach uses the co-occurrence value between questions as edge weights, automatically increasing the transition probability of frequently co-occurring questions and ensuring that the recommended question chains align with actual dispute patterns. This overcomes the traditional assumption of uniform edge weight distribution, which fails to capture the strength of semantic connections between questions.

[0119] In traditional PageRank, all nodes have the same initial weight (1 / N). However, in this application, the newly created court transcripts are applied to the previously optimized co-occurrence matrix to generate a list of recommended questions. The score of each question is used as the initial weight for each question vertex. This helps high-frequency questions (such as those at the center of controversy) receive higher initial weights, which continuously influence the results during iterations. The following describes the implementation strategy of this optimization process, described in steps S603-S606.

[0120] S603: Obtain a semantic weighted score for each question in the question recommendation list according to the result vector, and perform normalization based on the semantic weighted scores of each question in the question recommendation list to obtain an initial weight for each question and an initial weight vector.

[0121] The result vector here can be the vector obtained after performing vector retrieval, feature enhancement, and vector summation using the second optimized co-occurrence matrix. Of course, if the PageRank improvement is independent of the improvements in points 1 and 2 above, the result vector here can also be the original, unoptimized co-occurrence matrix. This example uses the second optimized co-occurrence matrix to obtain the result vector and recommend questions in the question recommendation list to expand on this step and subsequent steps.

[0122] It can be understood that the semantic weighted score of each representative question is represented by the numerical values of the elements at different positions in the result vector. Each score can be normalized, and the normalized values obtained are used as the initial weights of the representative questions corresponding to the corresponding positions. The vector containing the initial weights of each question, that is, the vector after the numerical normalization of the result vector, can be called the initial weight vector. The initial weight vector includes the initial weights of each vertex in the question co-occurrence graph. The normalization formula is as follows:

[0123]

[0124] Where q i represents the i-th question, s i Represents the problem q in the result vector i The semantic weighted score, PR0(q i ) represents the normalized problem q iThe initial weight of . For other problems, the initial weight can be calculated by referring to the above formula. In this application, an iterative optimization method is adopted to asynchronously update the weight of each vertex in the problem co-occurrence graph so that it is continuously updated based on the initial weight. See the introduction of S604.

[0125] S604: Iteratively optimize the initial weight vector using the problem transfer matrix, the initial weight vector, and the total number of vertices in the co-occurrence problem graph to obtain a weight vector after multiple iterations.

[0126] An optional implementation of this step is shown in the following formula:

[0127]

[0128] Where, represents the damping coefficient, which can be set to 0.8; T represents the problem transfer matrix, which is obtained after step S601; N represents the total number of vertices in the co-occurrence problem graph, which can also be understood as the total number of problems in the problem set; t represents the number of iterations, and when t=0, w (0) represents the initial weight vector, obtained after executing step S603. The 1 in the above formula represents a vector whose elements are all 1. In this way, the sum of a vector and a vector can still be represented by a vector. By using the above formula, the new weight vector containing the weights of each vertex after one iteration can be calculated. As an example, 100 iterations can be performed to finally obtain the vector w (100) .

[0129] S605 . For each question in the question recommendation list, a final score of the question is obtained using the semantic weighted score of the question and the weight of the question in the weight vector after multiple iterations.

[0130] The following shows an example implementation of this step:

[0131]

[0132] For each question, the above formula can be used to calculate its final score. x Indicates the weight of the problem after x iterations, that is, the weight of the vertex representing the problem after the problem co-occurrence graph is iterated x times. semantic It represents the semantic weighted score of the question, and its value is equal to the value of the element corresponding to the position of the question in the result vector of the second optimized co-occurrence matrix. β represents the weight of the semantic weighted score, and accordingly, 1-β represents the weight of the vertex weight in the graph. The Score obtained by calculation is final represents the final score of the problem. For example, β = 0.9. This means that the semantically weighted score takes a greater weight, but at the same time, the weight of the vertex after multiple iterations is also taken into account in the scoring.

[0133] S606: Based on the final score of each question in the question recommendation list, readjust the order of the questions in the question recommendation list.

[0134] For example, the questions in the question recommendation list are sorted from highest to lowest based on the final score of each question obtained in step S605. This sorting method takes into account the statistical characteristics of the domain of the vertex itself, helping high-frequency questions (such as controversial issues) receive higher initial weights, which continuously influences the results during iterations.

[0135] The inventors of this application proposed a three-level optimization framework for the co-occurrence matrix: eliminating noise through sparse processing (threshold filtering) and retaining significant co-occurrence relationships; introducing similarity weighting to calculate the semantic similarity between recommended questions and material keywords, and dynamically enhancing the weight of questions that conform to the core semantics of the case; applying PageRank weighting to iteratively update the importance of questions through the topological structure of the co-occurrence network to ensure that the recommendation results reflect both local co-occurrence characteristics and global importance distribution, taking into account the global influence of vertices in the network.

[0136] This application combines traditional statistical methods, such as co-occurrence matrices, with cutting-edge large language model technology to achieve a deep fusion of statistical learning and domain knowledge. This innovation not only significantly improves the accuracy of semantic understanding but also provides key clues for the intelligent generation of investigative questions. Compared to purely manual questioning, this solution achieves a qualitative leap in efficiency while effectively avoiding the loss of critical information due to human oversight. Traditional manual questioning requires arbitrators to spend an average of 4-6 hours studying the materials and formulating questions. However, this application proposes a comprehensive framework for automatically generating in-court investigative questions. Based on co-occurrence matrix technology, this framework utilizes keywords from pre-court documents submitted by both parties to generate a list of compliance questions in seconds. Arbitrators only need to gain a rough understanding of the needs of both parties during the trial. They can then ask questions based on the intelligently recommended list of questions in this solution, listen to both parties' statements, and uncover the truth of the case, thereby providing a basis for a fair judgment. This significantly reduces the workload of traditional manual questioning. Furthermore, automated processing reduces repetitive manual labor and is expected to reduce the overall cost of arbitration services by over 30%, making it particularly suitable for handling high-concurrency cases.

[0137] Furthermore, this application fills a gap in intelligent questioning in labor arbitration. Currently, there is no fully computer-generated intelligent recommendation solution for investigative questions in labor arbitration hearings. The technology proposed in this application fills this gap, aiming to assist arbitrators in reducing workload, improving efficiency, and preventing the omission of important information due to human oversight.

[0138] The co-occurrence matrix is a commonly used technique in natural language processing and text generation tasks, used to capture the relationship between words. It constructs a matrix representing the correlation between words by counting the frequency of words co-occurring in a specific context, quantifying the co-occurrence relationship between words and thus revealing their semantic association. However, given the limitations of the co-occurrence matrix, it is obviously impossible to directly use the co-occurrence matrix to count the connection between the materials submitted by the applicant and the respondent and the questions raised by the arbitrator in the scenario of intelligent recommendation of investigation questions in arbitration trials. Therefore, for the scenario of intelligent recommendation of investigation questions in arbitration trials, the patent of this invention will integrate labor arbitration business knowledge, analyze the process of raising investigation questions in trial transcripts, and learn the relationship between the materials submitted by both parties and questions, and between questions and questions in the existing transcripts through various algorithms such as PageRank, ultimately achieving the recommendation of the optimal list of questions based on the materials.

[0139] This application demonstrates the first integration and improvement of PageRank in the judicial field: it organically combines co-occurrence statistical features with graph neural representations; and also implements dynamic correction of transition probabilities based on court trial scenarios. This application improves on traditional PageRank by abandoning the average edge weight definition and adopting per-present value to automatically strengthen the transition probability of frequently co-occurring questions, ensuring that the recommended question chain conforms to the actual dispute pattern. It also abandons the average point weight definition and adopts the existing score of each node as the initial point weight, which helps high-frequency questions obtain higher initial weights, continuously influencing the results during iterations, making the recommended questions more accurate and in line with the scenario requirements.

[0140] Based on the question recommendation method introduced in the above embodiment, the present application also provides a question recommendation device. Figure 7 Figure 2 is a schematic diagram of the structure of the device. Figure 7 The problem recommendation devices shown include:

[0141] A first acquisition module 71 is configured to acquire a keyword set and a word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript;

[0142] A retrieval module 72 is configured to retrieve a plurality of feature vectors from a pre-constructed co-occurrence matrix based on the keyword set; the plurality of feature vectors correspond one-to-one to each keyword in the keyword set; the co-occurrence matrix is a two-dimensional matrix, wherein a first dimension represents questions and a second dimension represents keywords; wherein the questions represented by the first dimension of the co-occurrence matrix are question sentences extracted from a collection of historical court trial transcripts, and the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the collection of historical court trial transcripts;

[0143] An enhancement module 73 is configured to perform feature enhancement processing on feature vectors corresponding to corresponding keywords using the word frequency to obtain a plurality of enhanced feature vectors;

[0144] A summing module 74 is configured to perform a sum operation on the plurality of enhanced feature vectors to obtain a result vector;

[0145] The recommendation module 75 is configured to generate a list of recommended questions for the target trial transcript based on the values in the result vector.

[0146] In an optional implementation, the question recommendation device further includes:

[0147] A second acquisition module is configured to acquire the historical court trial transcript dataset; the historical court trial transcript dataset includes a plurality of historical court trial transcripts;

[0148] The first construction module is used to extract questions and keywords from each of the historical trial transcripts, and construct a question set and a keyword set corresponding to each of the historical trial transcripts;

[0149] a statistical module for counting the word frequency of each keyword in the keyword set in the corresponding historical court trial transcript, and constructing a keyword-word frequency pair corresponding to each historical court trial transcript;

[0150] A second construction module is configured to construct a representative word set corresponding to the historical court trial transcript dataset based on the semantic similarities between the constructed keyword sets and different keywords; and to construct a representative question set corresponding to the historical court trial transcript dataset based on the semantic similarities between the constructed question sets and different questions;

[0151] a correction module, configured to correct the keyword-word frequency pairs corresponding to each of the historical court trial transcripts based on the semantic similarity between different keywords and the representative word set, thereby obtaining corrected keyword-word frequency pairs corresponding to each of the historical court trial transcripts;

[0152] A matrix construction module is used to construct the co-occurrence matrix based on the representative word set, the representative question set and the corrected keyword-word frequency pairs corresponding to each of the historical trial transcripts.

[0153] In an optional implementation, the matrix building module includes:

[0154] an initial matrix construction unit, configured to construct an initial matrix with each representative question in the representative question set as a first dimension index, each representative word in the representative word set as a second dimension index, and with elements at each position set to 0;

[0155] An updating unit is configured to traverse each of the historical trial transcripts in the historical trial transcript data set, convert a question set corresponding to the currently traversed historical trial transcript into a first representative question set based on semantic similarity between different questions, and convert a keyword set corresponding to the currently traversed historical trial transcript into a first representative word set based on semantic similarity between different keywords; use the first representative question set and the first representative word set to query the elements at corresponding positions of the current initial matrix, and superimpose the word frequencies of the corrected keyword-word frequency pairs corresponding to the currently traversed historical trial transcript with the values of the elements at corresponding positions of the current initial matrix to complete an update of the initial matrix;

[0156] The co-occurrence matrix determining unit is configured to use the initial matrix that has been last updated as the co-occurrence matrix after the historical court trial transcript data set has been traversed.

[0157] In an optional implementation, the first representative question set includes m representative questions, and the first representative word set includes n representative words; m and n are both positive integers;

[0158] The updating unit is specifically configured to:

[0159] Traversing the first representative question set and the first representative word set, for the i-th representative question of the first representative question set and the j-th representative word of the first representative word set, querying a target position in the current initial matrix and obtaining a current value of the element at the target position;

[0160] i takes values from 1 to m, j takes values from 1 to n, and both i and j are integers;

[0161] Determining the frequency of the jth representative word in the first representative word set in the revised keyword-word frequency pairs corresponding to the currently traversed historical trial transcript;

[0162] Adding the word frequency of the j-th representative word to the current value of the element at the target position to complete the value update of the element at the target position;

[0163] After the first representative question set and the first representative word set are traversed, the initial matrix is updated once.

[0164] In an optional implementation, the first acquisition module 71 includes:

[0165] A first acquiring unit is configured to acquire the number of occurrences of a first target keyword in the keyword set in the target trial transcript;

[0166] A second acquiring unit is configured to acquire the number of times other keywords in the keyword set appear in the target trial transcript;

[0167] a first word frequency calculation unit, configured to use the sum of the number of occurrences of the first target keyword in the target court trial transcript and the number of occurrences of the other keywords in the target court trial transcript as a denominator, and the number of occurrences of the first target keyword in the target court trial transcript as a numerator, and to obtain a ratio as the word frequency of the first target keyword in the target court trial transcript;

[0168] The statistics module includes:

[0169] a third acquiring unit, configured to acquire, from a keyword set corresponding to a target historical trial transcript in the historical trial transcript dataset, the number of occurrences of a second target keyword in the target historical trial transcript and the number of occurrences of other keywords in the target trial transcript;

[0170] The second word frequency calculation unit is used to take the sum of the number of times the second target keyword appears in the target historical trial transcript and the number of times other keywords appear in the target historical trial transcript as the denominator, and take the number of times the second target keyword appears in the target historical trial transcript as the numerator, and obtain the ratio as the word frequency of the second target keyword in the target historical trial transcript to construct a keyword-word frequency pair of the second target keyword.

[0171] In an optional implementation, the correction module is specifically configured to:

[0172] If the second target keyword is one of the plurality of representative words in the representative word set, determining a similar keyword in the keyword set corresponding to the target historical trial transcript that has a similar word mapping relationship with the second target keyword;

[0173] Adding the word frequency of the similar keyword in the target historical trial transcript and the word frequency of the second target keyword in the target historical trial transcript to obtain a sum of first word frequencies, and using the sum of the first word frequencies to modify the keyword-word frequency pair of the second target keyword;

[0174] If the second target keyword is not one of the multiple representative words in the representative word set, determining a target representative word corresponding to the second target keyword in the representative word set based on a similar word mapping relationship;

[0175] The word frequency of the second target keyword in the target historical trial transcript, the word frequency of the target representative word in the target historical trial transcript, and the word frequency of other keywords with similar word mapping relationships with the target representative word in the target historical trial transcript are added together to obtain the sum of the second word frequencies, and the sum of the second word frequencies is used to correct the keyword-word frequency pair of the target representative word corresponding to the target historical trial transcript.

[0176] In an optional implementation, the recommendation module 75 is specifically configured to:

[0177] Determine the problem corresponding to the positive element in the result vector based on the one-to-one correspondence between the elements at each position of the result vector and the problem;

[0178] Arrange the determined multiple questions in descending order according to the numerical values of the corresponding elements to generate a recommended list of questions for the target trial transcript.

[0179] In an optional implementation, the question recommendation device further includes:

[0180] A second acquisition module is used to acquire a historical court trial transcript dataset; the historical court trial transcript dataset includes multiple historical court trial transcripts;

[0181] The third construction module is used to extract questions from each of the historical court trial transcripts and construct a set of questions corresponding to each of the historical court trial transcripts;

[0182] A fourth construction module is configured to construct a representative question set corresponding to the historical court trial transcript dataset based on the constructed question sets and the semantic similarity between different questions;

[0183] A sequence mining module is used to mine subsequences of length 2 whose support satisfies a preset condition from the representative question set, and add the mined subsequences as high-frequency binary question sequences to the mining sequence set; the question priority of the first question in the high-frequency binary question sequence is higher than the question priority of the second question in the latter position;

[0184] The question recommendation device also includes a question ranking adjustment module, which is used to:

[0185] If there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question with the higher recommendation ranking is the first question in the target high-frequency binary question sequence, and the question with the lower recommendation ranking is the second question in the target high-frequency binary question sequence, then the ranking of the two questions in the question recommendation list that both belong to the target high-frequency binary question sequence will not be adjusted;

[0186] If there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question recommended first is the second question in the target high-frequency binary question sequence, and the question recommended second is the first question in the target high-frequency binary question sequence, then in the question recommendation list, the question recommended first will be adjusted to after the question recommended second.

[0187] In an optional implementation, the enhancement module 73 is specifically configured to:

[0188] The feature vector corresponding to the keyword is multiplied by the word frequency of the keyword in the target trial transcript to obtain an enhanced feature vector of the keyword.

[0189] In an optional implementation, the question recommendation device further includes:

[0190] a semantic vector extraction module, configured to extract a word semantic vector for each representative word in the representative word set, and to extract a question semantic vector for each representative question in the representative question set;

[0191] A similarity calculation module is used to calculate the cosine similarity between each extracted word semantic vector and each question semantic vector, and construct a cosine similarity matrix based on the calculated cosine similarities; in the cosine similarity matrix, the first dimension represents the representative question, and the second dimension represents the representative word;

[0192] The matrix fusion module is used to fuse the co-occurrence matrix with the cosine similarity matrix to obtain a first optimized co-occurrence matrix.

[0193] In an optional implementation, the question recommendation device further includes:

[0194] a numerical filtering module, configured to perform numerical filtering on the numerical values of elements at each position in the first optimized co-occurrence matrix using a preset threshold, resetting elements with numerical values lower than the preset threshold to 0, and retaining elements with numerical values higher than or equal to the preset threshold, to obtain a second optimized co-occurrence matrix;

[0195] The retrieval module 72 is specifically used to:

[0196] A plurality of feature vectors are retrieved from the second optimized co-occurrence matrix based on the keyword set.

[0197] In an optional implementation, the question recommendation device further includes:

[0198] A fifth construction module is configured to construct a question transition matrix for the historical trial transcript dataset based on the co-occurrence of pairs of questions in each of the historical trial transcripts; in the question transition matrix, the first dimension and the second dimension both represent representative questions, and the representative questions corresponding to the rows where the elements on the diagonal of the question transition matrix are located are consistent with the representative questions corresponding to the columns where the elements are located; wherein the value of the element in the i-th row and j-th column of the question transition matrix represents the probability of transition from the i-th representative question to the j-th representative question;

[0199] A graph construction module, configured to construct a question co-occurrence graph with representative questions as vertices and transition probabilities as edge weights based on the representative question set and the question transition matrix;

[0200] a normalization processing module, configured to obtain a semantically weighted score for each question in the question recommendation list based on the result vector, and to perform normalization processing based on the semantically weighted scores of each question in the question recommendation list to obtain an initial weight for each question and an initial weight vector; the initial weight vector includes the initial weight of each vertex in the question co-occurrence graph;

[0201] A weight vector optimization module, configured to optimize the initial weight vector for multiple iterations using the problem transfer matrix, the initial weight vector, and the total number of vertices in the co-occurrence problem graph to obtain a weight vector after multiple iterations;

[0202] A score calculation module is used to obtain a final score for each question in the question recommendation list using the semantic weighted score of the question and the weight of the question in the weight vector after multiple iterations;

[0203] The question ranking readjustment module is used to readjust the ranking of the questions in the question recommendation list based on the final score of each question in the question recommendation list.

[0204] In an optional implementation, the fifth building block is specifically configured to:

[0205] Traversing the historical court trial transcript dataset, counting the co-occurring questions in each pair in each historical court trial transcript, and adding the count values to two symmetrical positions related to the two co-occurring questions in the first transfer matrix until the traversal is completed to obtain a second transfer matrix; the first transfer matrix is an initialized question transfer matrix, and the elements at each position are set to 0 during initialization; in the first transfer matrix, the first dimension and the second dimension both represent representative questions;

[0206] Perform numerical normalization processing on each row element of the second transfer matrix to obtain the problem transfer matrix.

[0207] Based on the question recommendation method and question recommendation apparatus described in the above embodiments, the present application also provides a question recommendation device, which includes: a memory and a processor;

[0208] The memory is used to store computer programs;

[0209] The processor is configured to run the computer program, and when the computer program is run, the steps of the question recommendation method in any implementation manner described in the method embodiment are executed.

[0210] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and equipment embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and equipment embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0211] The above is merely one specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A question recommendation method, characterized in that: include: Obtaining a keyword set and a word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript; A plurality of feature vectors are retrieved from a pre-constructed co-occurrence matrix based on the keyword set; the plurality of feature vectors correspond one-to-one to each keyword in the keyword set; the co-occurrence matrix is a two-dimensional matrix, wherein a first dimension represents questions and a second dimension represents keywords; wherein the questions represented by the first dimension of the co-occurrence matrix are question sentences extracted from a collection of historical court trial transcripts, and the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the collection of historical court trial transcripts; Performing feature enhancement processing on feature vectors corresponding to corresponding keywords using the word frequencies to obtain multiple enhanced feature vectors; Performing a sum operation on the multiple enhanced feature vectors to obtain a result vector; A list of recommended questions for the target court trial transcript is generated according to the values in the result vector.

2. The method according to claim 1, characterized in that Also includes: Acquire the historical court trial transcript dataset; the historical court trial transcript dataset includes multiple historical court trial transcripts; Extracting questions and keywords from each of the historical court trial transcripts, and constructing a set of questions and a set of keywords corresponding to each of the historical court trial transcripts; Counting the word frequency of each keyword in the keyword set in the corresponding historical court trial transcript, and constructing a keyword-word frequency pair corresponding to each historical court trial transcript; Based on the constructed keyword sets and the semantic similarity between different keywords, construct a representative word set corresponding to the historical trial transcript dataset; and, based on the constructed question sets and the semantic similarities between different questions, constructing a representative question set corresponding to the historical trial transcript dataset; Based on the semantic similarity between different keywords and the representative word set, correcting the keyword-word frequency pair corresponding to each of the historical trial transcripts to obtain a corrected keyword-word frequency pair corresponding to each of the historical trial transcripts; The co-occurrence matrix is constructed based on the representative word set, the representative question set, and the corrected keyword-word frequency pairs corresponding to each of the historical trial transcripts.

3. The method according to claim 2, characterized in that The co-occurrence matrix is constructed based on the representative word set, the representative question set, and the corrected keyword-word frequency pairs corresponding to each of the historical trial transcripts, including: Constructing an initial matrix with each representative question in the representative question set as a first dimension index, each representative word in the representative word set as a second dimension index, and with elements at each position set to 0; Traversing each of the historical court trial transcripts in the historical court trial transcript data set, converting the question set corresponding to the currently traversed historical court trial transcript into a first representative question set based on the semantic similarity between different questions, and converting the keyword set corresponding to the currently traversed historical court trial transcript into a first representative word set based on the semantic similarity between different keywords; using the first representative question set and the first representative word set to query the elements at the corresponding positions of the current initial matrix, and superimposing the values of the elements at the corresponding positions of the current initial matrix with the word frequencies of the corrected keyword-word frequency pairs corresponding to the currently traversed historical court trial transcript, thereby completing an update of the initial matrix; After the historical court trial transcript data set is traversed, the initial matrix that has completed the last update is used as the co-occurrence matrix.

4. The method according to claim 3, characterized in that The first representative question set includes m representative questions, and the first representative word set includes n representative words; m and n are both positive integers; The querying of the elements at corresponding positions of the current initial matrix using the first representative question set and the first representative word set, and superimposing the values of the elements at corresponding positions of the current initial matrix with the word frequencies of the corrected keyword-word frequency pairs corresponding to the currently traversed historical trial transcripts, thereby completing an update of the initial matrix, includes: Traversing the first representative question set and the first representative word set, for the i-th representative question of the first representative question set and the j-th representative word of the first representative word set, querying a target position in the current initial matrix and obtaining a current value of the element at the target position; i takes values from 1 to m, j takes values from 1 to n, and both i and j are integers; Determining the frequency of the jth representative word in the first representative word set in the revised keyword-word frequency pairs corresponding to the currently traversed historical trial transcript; Adding the word frequency of the j-th representative word to the current value of the element at the target position to complete the value update of the element at the target position; After the first representative question set and the first representative word set are traversed, the initial matrix is updated once.

5. The method according to claim 2, characterized in that Obtaining the word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript includes: Obtaining the number of occurrences of a first target keyword in the keyword set in the target court trial transcript, and obtaining the number of occurrences of other keywords in the keyword set in the target court trial transcript; The sum of the number of times the first target keyword appears in the target trial transcript and the number of times the other keywords appear in the target trial transcript is used as the denominator, and the number of times the first target keyword appears in the target trial transcript is used as the numerator, and the obtained ratio is used as the word frequency of the first target keyword in the target trial transcript; The counting of the word frequencies of the keywords in the keyword set in the corresponding historical trial transcripts to construct keyword-word frequency pairs corresponding to each historical trial transcript includes: Obtaining, from a keyword set corresponding to a target historical trial transcript in the historical trial transcript dataset, the number of occurrences of a second target keyword in the target historical trial transcript and the number of occurrences of other keywords in the target historical trial transcript; The sum of the number of times the second target keyword appears in the target historical trial transcript and the number of times other keywords appear in the target historical trial transcript is used as the denominator, and the number of times the second target keyword appears in the target historical trial transcript is used as the numerator. The obtained ratio is used as the word frequency of the second target keyword in the target historical trial transcript to construct the keyword-word frequency pair of the second target keyword.

6. The method according to claim 5, characterized in that The method of modifying the keyword-frequency pairs corresponding to each historical trial transcript based on the semantic similarity between different keywords and the representative word set includes: If the second target keyword is one of the plurality of representative words in the representative word set, determining a similar keyword in the keyword set corresponding to the target historical trial transcript that has a similar word mapping relationship with the second target keyword; Adding the word frequency of the similar keyword in the target historical trial transcript and the word frequency of the second target keyword in the target historical trial transcript to obtain a sum of first word frequencies, and using the sum of the first word frequencies to modify the keyword-word frequency pair of the second target keyword; If the second target keyword is not one of the multiple representative words in the representative word set, determining a target representative word corresponding to the second target keyword in the representative word set based on a similar word mapping relationship; The word frequency of the second target keyword in the target historical trial transcript, the word frequency of the target representative word in the target historical trial transcript, and the word frequency of other keywords with similar word mapping relationships with the target representative word in the target historical trial transcript are added together to obtain the sum of the second word frequencies, and the sum of the second word frequencies is used to correct the keyword-word frequency pair of the target representative word corresponding to the target historical trial transcript.

7. The method according to claim 1, characterized in that Generating a list of recommended questions for the target trial transcript based on the values in the result vector includes: Determine the problem corresponding to the positive element in the result vector based on the one-to-one correspondence between the elements at each position of the result vector and the problem; Arrange the determined multiple questions in descending order according to the numerical values of the corresponding elements to generate a recommended list of questions for the target trial transcript.

8. The method according to claim 7, characterized in that Also includes: Obtaining a historical court trial transcript dataset; the historical court trial transcript dataset includes multiple historical court trial transcripts; Extracting questions from each of the historical court trial transcripts to construct a set of questions corresponding to each of the historical court trial transcripts; Based on the constructed question sets and the semantic similarity between different questions, a representative question set corresponding to the historical trial transcript dataset is constructed; Mining subsequences of length 2 whose support satisfies a preset condition from the representative question set, and adding the mined subsequences as high-frequency binary question sequences to the mining sequence set; The priority of the first question in the high-frequency binary question sequence is higher than the priority of the second question in the latter position; After generating the list of recommended questions for the target trial transcript, the method further includes: If there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question with the higher recommendation ranking is the first question in the target high-frequency binary question sequence, and the question with the lower recommendation ranking is the second question in the target high-frequency binary question sequence, then the ranking of the two questions in the question recommendation list that both belong to the target high-frequency binary question sequence will not be adjusted; If there are two questions in the question recommendation list that both belong to the target high-frequency binary question sequence, and the question recommended first is the second question in the target high-frequency binary question sequence, and the question recommended second is the first question in the target high-frequency binary question sequence, then in the question recommendation list, the question recommended first will be adjusted to after the question recommended second.

9. The method according to claim 1, characterized in that The method of performing feature enhancement processing on the feature vectors corresponding to the corresponding keywords using the word frequency to obtain a plurality of enhanced feature vectors includes: The feature vector corresponding to the keyword is multiplied by the word frequency of the keyword in the target trial transcript to obtain an enhanced feature vector of the keyword.

10. The method according to claim 2, characterized in that After constructing the co-occurrence matrix, the method further includes: Extracting a word semantic vector for each representative word in the representative word set, and extracting a question semantic vector for each representative question in the representative question set; Calculating the cosine similarity between each extracted word semantic vector and each question semantic vector, and constructing a cosine similarity matrix based on the calculated cosine similarities; in the cosine similarity matrix, the first dimension represents representative questions, and the second dimension represents representative words; The co-occurrence matrix is fused with the cosine similarity matrix to obtain a first optimized co-occurrence matrix.

11. The method according to claim 10, characterized in that After obtaining the first optimized co-occurrence matrix, the method further includes: Performing numerical filtering on the values of elements at each position in the first optimized co-occurrence matrix using a preset threshold, resetting elements with values lower than the preset threshold to 0, and retaining elements with values higher than or equal to the preset threshold, to obtain a second optimized co-occurrence matrix; The retrieving a plurality of feature vectors from a pre-constructed co-occurrence matrix based on the keyword set includes: A plurality of feature vectors are retrieved from the second optimized co-occurrence matrix based on the keyword set.

12. The method according to claim 11, characterized in that Also includes: Based on the co-occurrence of two questions in each of the historical trial transcripts, construct a question transfer matrix for the historical trial transcript dataset; In the problem transfer matrix, the first dimension and the second dimension both represent representative problems, and the representative problems corresponding to the rows where the elements on the diagonal of the problem transfer matrix are located are consistent with the representative problems corresponding to the columns where the elements are located; wherein the value of the element in the i-th row and j-th column of the problem transfer matrix represents the probability of transition from the i-th representative problem to the j-th representative problem; Based on the representative question set and the question transition matrix, constructing a question co-occurrence graph with the representative questions as vertices and the transition probabilities as edge weights; Obtaining a semantically weighted score for each question in the question recommendation list based on the result vector, and performing normalization processing based on the semantically weighted scores of each question in the question recommendation list to obtain an initial weight for each question and an initial weight vector; the initial weight vector includes the initial weight of each vertex in the question co-occurrence graph; Iteratively optimizing the initial weight vector using the problem transfer matrix, the initial weight vector, and the total number of vertices in the co-occurrence problem graph to obtain a weight vector after multiple iterations; For each question in the question recommendation list, a final score of the question is obtained using the semantic weighted score of the question and the weight of the question in the weight vector after the multiple iterations; Based on the final score of each question in the question recommendation list, the order of the questions in the question recommendation list is readjusted.

13. The method according to claim 12, characterized in that The question transfer matrix of the historical trial transcript dataset is constructed based on the co-occurrence of two questions in each historical trial transcript, including: Traversing the historical court trial transcript dataset, counting the co-occurring questions in each pair in each historical court trial transcript, and adding the count values to two symmetrical positions related to the two co-occurring questions in the first transfer matrix until the traversal is completed to obtain a second transfer matrix; the first transfer matrix is an initialized question transfer matrix, and the elements at each position are set to 0 during initialization; in the first transfer matrix, the first dimension and the second dimension both represent representative questions; Perform numerical normalization processing on each row element of the second transfer matrix to obtain the problem transfer matrix.

14. A question recommendation device, characterized in that: include: A first acquisition module is configured to acquire a keyword set and a word frequency of each keyword in the keyword set in the target trial transcript based on the target trial transcript; A retrieval module is configured to retrieve a plurality of feature vectors from a pre-constructed co-occurrence matrix based on the keyword set; the plurality of feature vectors correspond one-to-one to each keyword in the keyword set; the co-occurrence matrix is a two-dimensional matrix, wherein a first dimension represents questions and a second dimension represents keywords; wherein the questions represented by the first dimension of the co-occurrence matrix are question sentences extracted from a collection of historical court trial transcripts, and the keywords represented by the second dimension of the co-occurrence matrix are text keywords extracted from the collection of historical court trial transcripts; an enhancement module is configured to perform feature enhancement processing on the feature vectors corresponding to the corresponding keywords using the word frequency to obtain a plurality of enhanced feature vectors; A summing module, configured to perform a sum operation on the plurality of enhanced feature vectors to obtain a result vector; A recommendation module is used to generate a list of recommended questions for the target trial transcript based on the values in the result vector.

15. A question recommendation device, characterized in that: include: memory and processor; The memory is used to store computer programs; The processor is configured to run the computer program, and when the computer program is run, the steps of the question recommendation method according to any one of claims 1 to 13 are executed.

Images