Dispute self-feedback mediation suggestion and solution generation system and method based on big model

Abstract

The present invention discloses a dispute self-feedback mediation suggestion and solution generation system and method based on a large model, and relates to the technical field of data analysis. The present invention obtains access rights to an information base through a management port; performs text data retrieval and retrieval on target case information in a case information base; based on the retrieved text data of the target case, performs paragraph structure division on the text and extracts text features to construct a feature matrix of the target case text; retrieves historical reference cases and performs matrix decomposition analysis based on the feature matrices of the target case and the reference case, respectively, to obtain corresponding singular feature sub-matrices; performs matrix feature correlation value analysis on the singular feature sub-matrices, respectively, and determines sample cases; performs matching correlation analysis on the singular feature sub-matrices of the target case and the sample case, and combines and generates a processing solution for the target case based on the analysis results combined with the sample case processing solution; and outputs the processing solution for the target case and the sample case.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a system and method for generating dispute self-feedback mediation suggestions and solutions based on a large model. Background Art

[0002] Dispute feedback mediation suggestions and solution generation based on big models is a method that combines artificial intelligence technology and legal expertise to improve mediation efficiency and quality through automation and intelligent means;

[0003] In the current environment, due to the gradual increase in population, the complexity and number of social problems have led to a rapid increase in the number of cases, involving many types and fields; this has brought great difficulties to traditional manual case handling, which requires detailed data collection and analysis of cases. This model cannot meet the needs of the current environment, resulting in low case handling efficiency, slow case handling progress and redundant cases, which is not conducive to efficient and high-quality case handling. Summary of the invention

[0004] The purpose of the present invention is to provide a dispute self-feedback mediation suggestion and solution generation system and method based on a large model to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for generating dispute self-feedback mediation suggestions and solutions based on a large model, the method comprising the following steps:

[0007] S100, obtaining access rights to the case information database through the management port; searching and retrieving text data of target case information in the case information database using case keywords;

[0008] S200, based on the retrieved text data of the target case, the text is divided into paragraph structures, and text features are extracted for each paragraph structure, and a feature matrix of the target case text is constructed based on the feature extraction results of each paragraph;

[0009] S300, retrieve historical reference cases and perform matrix decomposition analysis based on the feature matrices of the target case and the reference case, respectively, to obtain singular feature submatrices of the corresponding feature matrices; perform matrix feature correlation value analysis on the singular feature submatrices of the target case and the reference case, respectively, and screen the reference cases based on the analysis results to determine sample cases;

[0010] S400, performing a matching correlation analysis on the singular feature submatrices of the target case and the sample case, combining and generating a processing solution for the target case based on the analysis result and the processing solution for the sample case; and outputting the processing solution for the target case and the sample case.

[0011] The S100 obtains the access right to the case information database through the management port; the specific steps of searching and retrieving the text data of the target case information in the case information database through the case keywords are as follows:

[0012] S101, requesting access rights through the case information database management port, and accessing data in the case information database based on the obtained rights;

[0013] S102. Using the key attribute information of the target case as a search term, search and retrieve the target case data in the case information database; wherein the key attribute information of the target case includes the case field, case type, and information of personnel involved in the case.

[0014] The specific steps of S200, based on the retrieved text data of the target case, dividing the text into paragraph structures, extracting text features from each paragraph structure, and constructing a feature matrix of the target case text based on the feature extraction results of each paragraph are as follows:

[0015] S201, composing text based on retrieved data of the target case; dividing the text data of the target case into paragraphs and numbering and labeling each paragraph based on the paragraph sequence; converting and extracting text feature vectors of each text paragraph of the target case through a text feature vector conversion model; wherein the text feature vector conversion model is a bag-of-words model, TF-IDF, Word2Vec, GloVe, FastText, etc.;

[0016] S202, based on the numbering sequence of each text paragraph, the feature vector of each text paragraph is numbered in correspondence with the text feature vector of the text paragraph, and the feature matrix of the target case text is constructed by corresponding element alignment of the upper and lower adjacent sequence text feature vectors; wherein the feature vectors of each paragraph are combined and spliced ​​in the upper and lower order according to the numbering sequence, which means that the text feature vectors of the paragraphs with the earlier numbering sequence are used as a reference, and the feature vectors of the adjacent lower sequence paragraphs are arranged in correspondence with the element arrangement of the upper reference vector to complete the splicing combination of the adjacent sequence text feature vectors; wherein if there is an inconsistency in the number of elements in the adjacent sequence text feature vectors, the text feature vector with the smaller number of elements is filled with 0 at the end.

[0017] The specific steps of S300 for retrieving historical reference cases and performing matrix decomposition analysis based on the characteristic matrices of the target case and the reference case to obtain the singular characteristic submatrices of the corresponding characteristic matrices; performing matrix characteristic correlation value analysis on the singular characteristic submatrices of the target case and the reference case, and screening the reference cases based on the analysis results to determine the sample cases are as follows:

[0018] S301, retrieve the case information database, retrieve the reference case data of the same field and type as the target case, and process it through the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; perform matrix decomposition based on the text feature matrix of the target case and the reference case to analyze the feature distribution status in each text matrix, and extract the singular feature submatrix in the corresponding text matrix by decomposing the text feature matrix, and its calculation formula is:

[0019] ; ;

[0020] Among them, A(i,j) is the i*j feature matrix of the target case; U(i,i) is the i*i orthogonal matrix; V T (j,j) is the transpose of the j*j orthogonal matrix; E n (i,j) is the i*j feature matrix of the reference case numbered n; H(i,j) is the i*j singular feature submatrix of the target case; G n (i,j) is the i*j singular feature submatrix of the reference case numbered n; i,j are the number of rows and columns of the matrix; the matrix decomposition calculation is obtained by the matrix singular value decomposition calculation, so the corresponding singular feature submatrix is ​​a diagonal matrix, whose eigenvalues ​​are distributed on the diagonal, and the elements in other positions are 0;

[0021] S302, based on the decomposition calculation and analysis of each text feature matrix, extract the singular feature submatrices in the target case text feature matrix and each reference case text feature matrix respectively, and determine the feature distribution status of each text feature matrix; perform matrix feature correlation value analysis based on the singular feature submatrices of each text feature matrix, and the calculation formula is:

[0022] ;

[0023] ;

[0024] Among them, MF[H(i,j)] is the matrix feature correlation value of the singular feature submatrix of the target case feature matrix; MF[G n (i, j)] is the matrix feature correlation value of the singular feature submatrix of the reference case feature matrix numbered n; ||H(i, j)|| and ||G n (i, j)|| is the Euclidean norm calculation of the matrix; C is the correlation parameter; Y w [H(i,j)] and Y w [G n(i, j)] are the eigenvalues ​​on the diagonal of the singular feature submatrix of the target case feature matrix and the singular feature submatrix of the corresponding reference case feature matrix numbered n; w is the number of eigenvalues ​​on the diagonal of each singular feature submatrix;

[0025] Based on the analysis results of the matrix feature correlation values ​​of the singular feature submatrices of each text feature matrix, the matrix feature correlation values ​​of the target case are used as the reference value, and the matrix feature correlation values ​​of the reference cases are screened. The absolute value of the difference between the matrix feature correlation values ​​of each reference case and the matrix feature correlation values ​​of the target case is calculated, and the calculation results are numerically sorted. The reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case.

[0026] The S400 performs a matching correlation analysis on the singular feature submatrices of the target case and the sample case, and generates a processing solution for the target case based on the analysis result and the processing solution for the sample case; the specific steps of outputting the processing solution for the target case and the sample case are as follows:

[0027] S401. Perform a matching correlation analysis based on the singular feature submatrix of the target case and the sample. The calculation formula is:

[0028] ;

[0029] Among them, SO[H(i,j),G n (i,j) min1 ] w and SO[H(i,j),G n (i,j) min2 ] w are the matching correlation values ​​of the target case singular feature submatrix and the first sample case singular feature submatrix, and the target case singular feature submatrix and the second sample case singular feature submatrix corresponding to the eigenvalue numbered w; Y w [G n (i,j)] min1 and Y w [G n (i,j)] min2 are the eigenvalues ​​numbered w in the first sample case singular feature submatrix and the second sample case singular feature submatrix respectively; θ1 and θ2 are the angles between the target case singular feature submatrix and the first sample singular feature submatrix, and the corresponding derived eigenvectors between the target case singular feature submatrix and the second sample singular feature submatrix respectively;

[0030] S402, based on the target case singular feature submatrix, the first sample case singular feature submatrix and the second sample case singular feature submatrix, respectively, the calculation results of the matching correlation values ​​of the eigenvalues ​​are compared and analyzed, and according to the comparison results, the processing scheme of the sample case of the corresponding numbered section to which the corresponding numbered eigenvalue belongs is taken as the reference processing scheme of the corresponding numbered section to which the target case belongs; if SO[H(i,j),G n (i,j) min1 ] w >SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue w in the singular feature submatrix of the first sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w <SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the eigenvalue numbered w in the singular feature submatrix of the second sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w =SO[H(i,j),G n (i,j) min2 ] w , then the processing schemes of the corresponding numbered sections described by the numbered eigenvalues ​​in the first sample case and the second sample case different-feature submatrices are all the reference processing schemes of the corresponding numbered sections of the target case;

[0031] Based on the reference processing solution selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing solution for the target case; the comprehensive reference processing solution for the target case and the sample case are output.

[0032] A dispute self-feedback mediation suggestion and solution generation system based on a large model, the system comprising a database retrieval module, a matrix construction module, a matrix comprehensive correlation analysis module and a solution output module;

[0033] The database retrieval module obtains the access rights of the case information library through the management port; retrieves and retrieves the text data of the target case information in the case information library through the case keywords; the matrix construction module divides the text into paragraph structures based on the retrieved text data of the target case, extracts text features of each paragraph structure, and constructs the feature matrix of the target case text based on the feature extraction results of each paragraph; the matrix comprehensive association analysis module retrieves historical reference cases and performs matrix decomposition analysis based on the feature matrices of the target case and the reference case respectively, and obtains the singular feature submatrix of the corresponding feature matrix; performs matrix feature association value analysis on the singular feature submatrices of the target case and the reference case respectively, and screens the reference case based on the analysis results to determine the sample case; the solution output module performs a matching association analysis on the singular feature submatrices of the target case and the sample case, and combines and generates the processing solution of the target case based on the analysis results and the sample case processing solution; and outputs the processing solution of the target case and the sample case.

[0034] The database retrieval module includes a permission access request unit and a case information retrieval unit;

[0035] The permission access request unit requests access rights through the case information database management port, and accesses data in the case information database based on the obtained permissions;

[0036] The case information retrieval unit uses the key attribute information of the target case as a search term to search and retrieve the target case data in the case information database.

[0037] The matrix construction module includes a text feature conversion unit and a text feature matrix construction unit;

[0038] The text feature conversion unit forms text based on the retrieved data of the target case; divides the text data of the target case into paragraphs and numbers and labels each paragraph based on the paragraph sequence; and converts and extracts text feature vectors of each text paragraph of the target case through a text feature vector conversion model;

[0039] The text feature matrix construction unit numbers the feature vectors of each text paragraph in a corresponding order based on the numbering sequence of each text paragraph, and constructs a feature matrix of the target case text by combining corresponding elements of the text feature vectors in the upper and lower adjacent order.

[0040] The matrix comprehensive association analysis module includes a matrix feature decomposition unit and a singular feature submatrix association analysis unit;

[0041] The matrix feature decomposition unit retrieves reference case data of the same type and field as the target case from the case information database, and processes the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; based on the text feature matrices of the target case and the reference case, matrix decomposition is performed to analyze the feature distribution status in each text matrix, and the singular feature submatrix in the corresponding text matrix is ​​extracted by decomposing the text feature matrix;

[0042] The singular feature submatrix association analysis unit extracts the singular feature submatrices from the target case text feature matrix and the reference case text feature matrices based on the decomposition calculation and analysis of each text feature matrix, and determines the feature distribution status of each text feature matrix; performs matrix feature association value analysis based on the singular feature submatrices of each text feature matrix; based on the matrix feature association value analysis results of the singular feature submatrices of each text feature matrix, the matrix feature association values ​​of the reference cases are screened with the matrix feature association values ​​of the target case as the reference value, and the calculation results are numerically sorted by calculating the absolute value of the difference between the matrix feature association values ​​of each reference case and the matrix feature association values ​​of the target case, and the reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case.

[0043] The solution output module includes a case feature matching correlation analysis unit and an abnormal equipment warning unit;

[0044] The case feature matching association analysis unit combines the target case with the sample's singular feature submatrix to perform matching association analysis;

[0045] The abnormal equipment warning unit compares and analyzes the calculation results of the matching correlation values ​​of each eigenvalue in the singular feature submatrix of the target case and the first sample case singular feature submatrix and the second sample case singular feature submatrix respectively, and according to the comparison results, takes the processing plan of the sample case of the corresponding numbered section to which the corresponding numbered eigenvalue belongs as the reference processing plan of the corresponding numbered section to which the target case belongs; based on the reference processing plan selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing plan for the target case; and the comprehensive reference processing plan and the sample case of the target case are output.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention calls a case database to perform big data information retrieval in combination with a target case, and performs similar retrieval on cases of the same type in the same field; constructs a corresponding feature matrix based on the retrieved data through text feature vector quantization processing, obtains the text singular feature submatrix of the corresponding case by decomposing the feature matrix, and obtains the feature distribution of the corresponding case text; calculates matrix feature association values ​​based on the singular feature submatrix of each case, and selects a sample matrix by analyzing the feature association values ​​of each matrix; performs a fit analysis based on the eigenvalues ​​in the sample matrix and the eigenvalues ​​of the target case, and selects a reference case processing solution based on the analysis results; the present invention performs a big data retrieval of similar cases for the target case, and outputs a sample case and a reference processing solution for the target case through a series of automated data analyses such as feature conversion analysis, association screening and fit comparison analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a structural schematic diagram of the dispute self-feedback mediation suggestion and solution generation system based on the big model of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] Example: Figure 1 As shown, the present invention provides a technical solution:

[0051] A method for generating dispute self-feedback mediation suggestions and solutions based on a large model, the method comprising the following steps:

[0052] S100, obtaining access rights to the case information database through the management port; searching and retrieving text data of target case information in the case information database using case keywords;

[0053] S200, based on the retrieved text data of the target case, the text is divided into paragraph structures, and text features are extracted for each paragraph structure, and a feature matrix of the target case text is constructed based on the feature extraction results of each paragraph;

[0054] S300, retrieve historical reference cases and perform matrix decomposition analysis based on the feature matrices of the target case and the reference case, respectively, to obtain singular feature submatrices of the corresponding feature matrices; perform matrix feature correlation value analysis on the singular feature submatrices of the target case and the reference case, respectively, and screen the reference cases based on the analysis results to determine sample cases;

[0055] S400, performing a matching correlation analysis on the singular feature submatrices of the target case and the sample case, combining and generating a processing solution for the target case based on the analysis result and the processing solution for the sample case; and outputting the processing solution for the target case and the sample case.

[0056] The S100 obtains the access right to the case information database through the management port; the specific steps of searching and retrieving the text data of the target case information in the case information database through the case keywords are as follows:

[0057] S101, requesting access rights through the case information database management port, and accessing data in the case information database based on the obtained rights;

[0058] S102. Using the key attribute information of the target case as a search term, search and retrieve the target case data in the case information database.

[0059] The specific steps of S200, based on the retrieved text data of the target case, dividing the text into paragraph structures, extracting text features from each paragraph structure, and constructing a feature matrix of the target case text based on the feature extraction results of each paragraph are as follows:

[0060] S201, composing text based on retrieved data of the target case; dividing the text data of the target case into paragraphs and numbering and labeling each paragraph based on the paragraph sequence; performing text feature vector conversion and extraction on each text paragraph of the target case through a text feature vector conversion model;

[0061] S202, based on the numbering sequence of each text paragraph, the feature vectors of each text paragraph are numbered in a corresponding sequence to the text feature vectors of the text paragraphs, and the feature matrix of the target case text is constructed by pairing corresponding elements of the text feature vectors in the upper and lower adjacent sequence.

[0062] The specific steps of S300 for retrieving historical reference cases and performing matrix decomposition analysis based on the characteristic matrices of the target case and the reference case to obtain the singular characteristic submatrices of the corresponding characteristic matrices; performing matrix characteristic correlation value analysis on the singular characteristic submatrices of the target case and the reference case, and screening the reference cases based on the analysis results to determine the sample cases are as follows:

[0063] S301, retrieve the case information database, retrieve the reference case data of the same field and type as the target case, and process it through the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; perform matrix decomposition based on the text feature matrix of the target case and the reference case to analyze the feature distribution status in each text matrix, and extract the singular feature submatrix in the corresponding text matrix by decomposing the text feature matrix, and its calculation formula is:

[0064] ; ;

[0065] Among them, A(i,j) is the i*j feature matrix of the target case; U(i,i) is the i*i orthogonal matrix; V T (j,j) is the transpose of the j*j orthogonal matrix; E n (i,j) is the i*j feature matrix of the reference case numbered n; H(i,j) is the i*j singular feature submatrix of the target case; G n (i,j) is the i*j singular characteristic submatrix of the reference case numbered n; i,j is the number of rows and columns of the matrix;

[0066] S302, based on the decomposition calculation and analysis of each text feature matrix, extract the singular feature submatrices in the target case text feature matrix and each reference case text feature matrix respectively, and determine the feature distribution status of each text feature matrix; perform matrix feature correlation value analysis based on the singular feature submatrices of each text feature matrix, and the calculation formula is:

[0067] ;

[0068] ;

[0069] Among them, MF[H(i,j)] is the matrix feature correlation value of the singular feature submatrix of the target case feature matrix; MF[G n (i, j)] is the matrix feature correlation value of the singular feature submatrix of the reference case feature matrix numbered n; ||H(i, j)|| and ||G n (i, j)|| is the Euclidean norm calculation of the matrix; C is the correlation parameter; Y w [H(i,j)] and Y w [G n (i, j)] are the eigenvalues ​​on the diagonal of the singular feature submatrix of the target case feature matrix and the singular feature submatrix of the corresponding reference case feature matrix numbered n; w is the number of eigenvalues ​​on the diagonal of each singular feature submatrix;

[0070] Based on the analysis results of the matrix feature correlation values ​​of the singular feature submatrices of each text feature matrix, the matrix feature correlation values ​​of the target case are used as the reference value, and the matrix feature correlation values ​​of the reference cases are screened. The absolute value of the difference between the matrix feature correlation values ​​of each reference case and the matrix feature correlation values ​​of the target case is calculated, and the calculation results are numerically sorted. The reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case.

[0071] The S400 performs a matching correlation analysis on the singular feature submatrices of the target case and the sample case, and generates a processing solution for the target case based on the analysis result and the processing solution for the sample case; the specific steps of outputting the processing solution for the target case and the sample case are as follows:

[0072] S401. Perform a matching correlation analysis based on the singular feature submatrix of the target case and the sample. The calculation formula is:

[0073] ;

[0074] Among them, SO[H(i,j),G n (i,j) min1 ] w and SO[H(i,j),G n (i,j) min2 ] w are the matching correlation values ​​of the target case singular feature submatrix and the first sample case singular feature submatrix, and the target case singular feature submatrix and the second sample case singular feature submatrix corresponding to the eigenvalue numbered w; Y w [G n (i,j)] min1 and Y w [G n (i,j)] min2 are the eigenvalues ​​numbered w in the first sample case singular feature submatrix and the second sample case singular feature submatrix respectively; θ1 and θ2 are the angles between the target case singular feature submatrix and the first sample singular feature submatrix, and the corresponding derived eigenvectors between the target case singular feature submatrix and the second sample singular feature submatrix respectively;

[0075] S402, based on the target case singular feature submatrix, the first sample case singular feature submatrix and the second sample case singular feature submatrix, respectively, the calculation results of the matching correlation values ​​of the eigenvalues ​​are compared and analyzed, and according to the comparison results, the processing scheme of the sample case of the corresponding numbered section to which the corresponding numbered eigenvalue belongs is taken as the reference processing scheme of the corresponding numbered section to which the target case belongs; if SO[H(i,j),G n (i,j) min1 ] w>SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue in the singular feature submatrix of the first sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w <SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue in the singular feature submatrix of the second sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w =SO[H(i,j),G n (i,j) min2 ] w , then the processing schemes of the corresponding numbered sections described by the numbered eigenvalues ​​in the first sample case and the second sample case different-feature submatrices are all the reference processing schemes of the corresponding numbered sections of the target case;

[0076] Based on the reference processing solution selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing solution for the target case; the comprehensive reference processing solution for the target case and the sample case are output.

[0077] A dispute self-feedback mediation suggestion and solution generation system based on a large model, the system comprising a database retrieval module, a matrix construction module, a matrix comprehensive correlation analysis module and a solution output module;

[0078] The database retrieval module obtains the access rights of the case information library through the management port; retrieves and retrieves the text data of the target case information in the case information library through the case keywords; the matrix construction module divides the text into paragraph structures based on the retrieved text data of the target case, extracts text features of each paragraph structure, and constructs the feature matrix of the target case text based on the feature extraction results of each paragraph; the matrix comprehensive association analysis module retrieves historical reference cases and performs matrix decomposition analysis based on the feature matrices of the target case and the reference case respectively, and obtains the singular feature submatrix of the corresponding feature matrix; performs matrix feature association value analysis on the singular feature submatrices of the target case and the reference case respectively, and screens the reference case based on the analysis results to determine the sample case; the solution output module performs a matching association analysis on the singular feature submatrices of the target case and the sample case, and combines and generates the processing solution of the target case based on the analysis results and the sample case processing solution; and outputs the processing solution of the target case and the sample case.

[0079] The database retrieval module includes a permission access request unit and a case information retrieval unit;

[0080] The permission access request unit requests access rights through the case information database management port, and accesses data in the case information database based on the obtained permissions;

[0081] The case information retrieval unit uses the key attribute information of the target case as a search term to search and retrieve the target case data in the case information database.

[0082] The matrix construction module includes a text feature conversion unit and a text feature matrix construction unit;

[0083] The text feature conversion unit forms text based on the retrieved data of the target case; divides the text data of the target case into paragraphs and numbers and labels each paragraph based on the paragraph sequence; and converts and extracts text feature vectors of each text paragraph of the target case through a text feature vector conversion model;

[0084] The text feature matrix construction unit numbers the feature vectors of each text paragraph in a corresponding order based on the numbering sequence of each text paragraph, and constructs a feature matrix of the target case text by combining corresponding elements of the text feature vectors in the upper and lower adjacent order.

[0085] The matrix comprehensive association analysis module includes a matrix feature decomposition unit and a singular feature submatrix association analysis unit;

[0086] The matrix feature decomposition unit retrieves reference case data of the same type and field as the target case from the case information database, and processes the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; based on the text feature matrices of the target case and the reference case, matrix decomposition is performed to analyze the feature distribution status in each text matrix, and the singular feature submatrix in the corresponding text matrix is ​​extracted by decomposing the text feature matrix;

[0087] The singular feature submatrix association analysis unit extracts the singular feature submatrices from the target case text feature matrix and the reference case text feature matrices based on the decomposition calculation and analysis of each text feature matrix, and determines the feature distribution status of each text feature matrix; performs matrix feature association value analysis based on the singular feature submatrices of each text feature matrix; based on the matrix feature association value analysis results of the singular feature submatrices of each text feature matrix, the matrix feature association values ​​of the reference cases are screened with the matrix feature association values ​​of the target case as the reference value, and the calculation results are numerically sorted by calculating the absolute value of the difference between the matrix feature association values ​​of each reference case and the matrix feature association values ​​of the target case, and the reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case.

[0088] The solution output module includes a case feature matching correlation analysis unit and an abnormal equipment warning unit;

[0089] The case feature matching association analysis unit combines the target case with the sample's singular feature submatrix to perform matching association analysis;

[0090] The abnormal equipment warning unit compares and analyzes the calculation results of the matching correlation values ​​of each eigenvalue in the singular feature submatrix of the target case and the singular feature submatrix of the first sample case and the singular feature submatrix of the second sample case respectively, and takes the processing scheme of the sample case of the corresponding numbered section to which the corresponding numbered eigenvalue belongs as the reference processing scheme of the corresponding numbered section to which the target case belongs according to the comparison result; based on the reference processing scheme selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing scheme for the target case; and the comprehensive reference processing scheme and the sample case of the target case are output;

[0091] In the examples:

[0092] At present, a case handling department adopts the dispute self-feedback mediation suggestion and solution generation system based on the large model of the present invention, and then requests access rights through the case information database management port, and accesses the case information database based on the obtained rights; uses the key attribute information of the target case as the search term, searches and retrieves the target case data in the case information database; forms the text based on the retrieved data of the target case; divides the text data of the target case into paragraphs and numbers and annotates each paragraph based on the paragraph sequence; converts and extracts the text feature vectors of each text paragraph of the target case through the text feature vector conversion model; based on the numbering sequence of each text paragraph, the feature vector of each text paragraph is numbered corresponding to the text feature vector of the text paragraph, and the feature matrix of the target case text is constructed by corresponding element pairing of the text feature vectors in the upper and lower adjacent sequence order;

[0093] The case information database is retrieved to retrieve the reference case data of the same type and field as the target case, and the corresponding text paragraph division processing, text feature vector conversion processing and text feature matrix construction processing are performed; based on the text feature matrices of the target case and the reference case, matrix decomposition is performed to analyze the feature distribution status in each text matrix, and the singular feature submatrix in the corresponding text matrix is ​​extracted by decomposing the text feature matrix. The calculation formula is:

[0094] ; ;

[0095] Based on the decomposition and calculation analysis of each text feature matrix, the singular feature submatrices in the target case text feature matrix and each reference case text feature matrix are extracted respectively to determine the feature distribution of each text feature matrix; the matrix feature correlation value analysis is performed based on the singular feature submatrices of each text feature matrix, and the calculation formula is:

[0096] ;

[0097] ;

[0098] Based on the analysis results of the matrix feature correlation values ​​of the singular feature submatrices of each text feature matrix, the matrix feature correlation values ​​of the target case are used as the reference value to screen the matrix feature correlation values ​​of the reference cases, and the absolute value of the difference between the matrix feature correlation values ​​of each reference case and the matrix feature correlation value of the target case is calculated, and the calculation results are numerically sorted, and the reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case;

[0099] Combine the singular feature submatrices of the target case and the sample to perform a matching correlation analysis, and the calculation formula is:

[0100] ;

[0101] Based on the target case singular feature submatrix, the calculation results of the matching correlation values ​​of each eigenvalue in the first sample case singular feature submatrix and the second sample case singular feature submatrix are compared and analyzed. According to the comparison results, the processing plan of the sample case in the corresponding numbered section of the corresponding numbered eigenvalue is taken as the reference processing plan of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w >SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue w in the singular feature submatrix of the first sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w <SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the eigenvalue numbered w in the singular feature submatrix of the second sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w =SO[H(i,j),G n (i,j) min2 ] w , then the processing schemes of the corresponding numbered sections described by the numbered eigenvalues ​​in the first sample case and the second sample case different-feature submatrices are all the reference processing schemes of the corresponding numbered sections of the target case;

[0102] Based on the reference processing solution selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing solution for the target case; the comprehensive reference processing solution for the target case and the sample case are output.

[0103] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for generating dispute mediation suggestions and solutions based on self-feedback of a large model, characterized by: The method comprises the following steps: S100, obtaining access rights to the case information database through the management port; searching and retrieving text data of target case information in the case information database using case keywords; S200, based on the retrieved text data of the target case, the text is divided into paragraph structures, and text features are extracted for each paragraph structure, and a feature matrix of the target case text is constructed based on the feature extraction results of each paragraph; S300, retrieve historical reference cases and perform matrix decomposition analysis based on the feature matrices of the target case and the reference case, respectively, to obtain singular feature submatrices of the corresponding feature matrices; perform matrix feature correlation value analysis on the singular feature submatrices of the target case and the reference case, respectively, and screen the reference cases based on the analysis results to determine sample cases; The specific steps of S300 are as follows: S301, retrieve the case information database, retrieve the reference case data of the same field and type as the target case, and process it through the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; perform matrix decomposition based on the text feature matrix of the target case and the reference case to analyze the feature distribution status in each text matrix, and extract the singular feature submatrix in the corresponding text matrix by decomposing the text feature matrix, and its calculation formula is: ； ； Among them, A(i,j) is the i*j feature matrix of the target case; U(i,i) is the i*i orthogonal matrix; V T (j,j) is the transpose of the j*j orthogonal matrix; E n (i,j) is the i*j feature matrix of the reference case numbered n; H(i,j) is the i*j singular feature submatrix of the target case; G n (i,j) is the i*j singular characteristic submatrix of the reference case numbered n; i,j is the number of rows and columns of the matrix; S302, based on the decomposition calculation and analysis of each text feature matrix, extract the singular feature submatrices in the target case text feature matrix and each reference case text feature matrix respectively, and determine the feature distribution status of each text feature matrix; perform matrix feature correlation value analysis based on the singular feature submatrices of each text feature matrix, and the calculation formula is: ； ； Among them, MF[H(i,j)] is the matrix feature correlation value of the singular feature submatrix of the target case feature matrix; MF[G n (i, j)] is the matrix characteristic correlation value of the singular characteristic submatrix of the reference case characteristic matrix numbered n; ||H(i, j)|| and ||G n (i, j)|| is the Euclidean norm calculation of the matrix; C is the correlation parameter; Y w [H(i,j)] and Y w [G n (i, j)] are the eigenvalues ​​on the diagonal of the singular feature submatrix of the target case feature matrix and the singular feature submatrix of the corresponding reference case feature matrix numbered n; w is the number of eigenvalues ​​on the diagonal of each singular feature submatrix; Based on the analysis results of the matrix feature correlation values ​​of the singular feature submatrices of each text feature matrix, the matrix feature correlation values ​​of the target case are used as the reference value to screen the matrix feature correlation values ​​of the reference cases, and the absolute value of the difference between the matrix feature correlation values ​​of each reference case and the matrix feature correlation value of the target case is calculated, and the calculation results are numerically sorted, and the reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case; S400, performing a matching correlation analysis on the singular feature submatrices of the target case and the sample case, combining and generating a treatment plan for the target case based on the analysis result and the treatment plan for the sample case; and outputting the treatment plan for the target case and the sample case; The specific steps of S400 are as follows: Combine the singular feature submatrices of the target case and the sample to perform a matching correlation analysis, and the calculation formula is: ； Among them, SO[H(i,j),G n (i,j) min1 ] w and SO[H(i,j),G n (i,j) min2 ] w are the matching correlation values ​​of the target case singular feature submatrix and the first sample case singular feature submatrix, and the target case singular feature submatrix and the second sample case singular feature submatrix corresponding to the eigenvalue numbered w; Y w [G n (i,j)] min1 and Y w [G n (i,j)] min2 are the eigenvalues ​​numbered w in the first sample case singular feature submatrix and the second sample case singular feature submatrix respectively; θ1 and θ2 are the angles between the target case singular feature submatrix and the first sample singular feature submatrix, and the corresponding derived eigenvectors between the target case singular feature submatrix and the second sample singular feature submatrix respectively; Based on the target case singular feature submatrix, the calculation results of the matching correlation values ​​of each eigenvalue in the first sample case singular feature submatrix and the second sample case singular feature submatrix are compared and analyzed. According to the comparison results, the processing plan of the sample case in the corresponding numbered section of the corresponding numbered eigenvalue is taken as the reference processing plan of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w >SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue in the singular feature submatrix of the first sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w <SO[H(i,j),G n (i,j) min2 ] w , then the processing scheme of the corresponding numbered section described by the numbered eigenvalue in the singular feature submatrix of the second sample case is taken as the reference processing scheme of the corresponding numbered section of the target case; if SO[H(i,j),G n (i,j) min1 ] w =SO[H(i,j),G n (i,j) min2 ] w , then the processing schemes of the corresponding numbered sections described by the numbered eigenvalues ​​in the first sample case and the second sample case different-feature submatrices are all the reference processing schemes of the corresponding numbered sections of the target case; Based on the reference processing solution selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing solution for the target case; the comprehensive reference processing solution for the target case and the sample case are output.

2. The method for generating dispute mediation suggestions and solutions based on self-feedback of a large model according to claim 1 is characterized by: The S100 obtains the access right to the case information database through the management port; the specific steps of searching and retrieving the text data of the target case information in the case information database through the case keywords are as follows: S101, requesting access rights through the case information database management port, and accessing data in the case information database based on the obtained rights; S102. Using the key attribute information of the target case as a search term, search and retrieve the target case data in the case information database.

3. The method for generating dispute self-feedback mediation suggestions and solutions based on a large model according to claim 2 is characterized by: The specific steps of S200, based on the retrieved text data of the target case, dividing the text into paragraph structures, extracting text features from each paragraph structure, and constructing a feature matrix of the target case text based on the feature extraction results of each paragraph are as follows: S201, composing text based on retrieved data of the target case; dividing the text data of the target case into paragraphs and numbering and labeling each paragraph based on the paragraph sequence; performing text feature vector conversion and extraction on each text paragraph of the target case through a text feature vector conversion model; S202, based on the numbering sequence of each text paragraph, the feature vectors of each text paragraph are numbered in a corresponding sequence to the text feature vectors of the text paragraphs, and the feature matrix of the target case text is constructed by pairing corresponding elements of the text feature vectors in the upper and lower adjacent sequence.

4. A system for generating dispute self-feedback mediation suggestions and solutions based on a large model, using a method for generating dispute self-feedback mediation suggestions and solutions based on a large model as claimed in any one of claims 1 to 3, characterized in that: The system includes a database retrieval module, a matrix construction module, a matrix comprehensive association analysis module and a solution output module; The database retrieval module obtains the access rights of the case information library through the management port; retrieves and retrieves the text data of the target case information in the case information library through the case keywords; the matrix construction module divides the text into paragraph structures based on the retrieved text data of the target case, extracts text features of each paragraph structure, and constructs the feature matrix of the target case text based on the feature extraction results of each paragraph; the matrix comprehensive association analysis module retrieves historical reference cases and performs matrix decomposition analysis based on the feature matrices of the target case and the reference case respectively, and obtains the singular feature submatrix of the corresponding feature matrix; performs matrix feature association value analysis on the singular feature submatrices of the target case and the reference case respectively, and screens the reference case based on the analysis results to determine the sample case; the solution output module performs a matching association analysis on the singular feature submatrices of the target case and the sample case, and combines and generates the processing solution of the target case based on the analysis results and the sample case processing solution; and outputs the processing solution of the target case and the sample case.

5. The dispute self-feedback mediation suggestion and solution generation system based on a large model according to claim 4 is characterized by: The database retrieval module includes a permission access request unit and a case information retrieval unit; The permission access request unit requests access rights through the case information database management port, and accesses data in the case information database based on the obtained permissions; The case information retrieval unit uses the key attribute information of the target case as a search term to search and retrieve the target case data in the case information database.

6. The dispute self-feedback mediation suggestion and solution generation system based on a large model according to claim 5 is characterized by: The matrix construction module includes a text feature conversion unit and a text feature matrix construction unit; The text feature conversion unit forms text based on the retrieved data of the target case; divides the text data of the target case into paragraphs and numbers and labels each paragraph based on the paragraph sequence; and converts and extracts text feature vectors of each text paragraph of the target case through a text feature vector conversion model; The text feature matrix construction unit assigns corresponding sequence numbers to the feature vectors of each text paragraph based on the numbering order of each text paragraph, and constructs a feature matrix of the target case text by pairwise combining corresponding elements of the text feature vectors in upper and lower adjacent sequence.

7. The dispute self-feedback mediation suggestion and solution generation system based on a large model according to claim 6 is characterized by: The matrix comprehensive association analysis module includes a matrix feature decomposition unit and a singular feature submatrix association analysis unit; The matrix feature decomposition unit retrieves reference case data of the same type and field as the target case from the case information database, and processes the corresponding text paragraph division, text feature vector conversion and text feature matrix construction; based on the text feature matrices of the target case and the reference case, matrix decomposition is performed to analyze the feature distribution status in each text matrix, and the singular feature submatrix in the corresponding text matrix is ​​extracted by decomposing the text feature matrix; The singular feature submatrix association analysis unit extracts the singular feature submatrices from the target case text feature matrix and the reference case text feature matrices based on the decomposition calculation and analysis of each text feature matrix, and determines the feature distribution status of each text feature matrix; performs matrix feature association value analysis based on the singular feature submatrices of each text feature matrix; based on the matrix feature association value analysis results of the singular feature submatrices of each text feature matrix, the matrix feature association values ​​of the reference cases are screened with the matrix feature association values ​​of the target case as the reference value, and the calculation results are numerically sorted by calculating the absolute value of the difference between the matrix feature association values ​​of each reference case and the matrix feature association values ​​of the target case, and the reference cases corresponding to the first minimum value and the second minimum value are taken as the first sample case and the second sample case.

8. The dispute self-feedback mediation suggestion and solution generation system based on a large model according to claim 7 is characterized by: The solution output module includes a case feature matching correlation analysis unit and an abnormal equipment warning unit; The case feature matching association analysis unit combines the target case with the sample's singular feature submatrix to perform matching association analysis; The abnormal equipment warning unit compares and analyzes the calculation results of the matching correlation values ​​of each eigenvalue in the singular feature submatrix of the target case, the first sample case singular feature submatrix and the second sample case singular feature submatrix respectively, and takes the processing scheme of the sample case of the corresponding numbered section to which the corresponding numbered eigenvalue belongs as the reference processing scheme of the corresponding numbered section to which the target case belongs according to the comparison result; Based on the reference processing solution selection results of each section of the target case, the corresponding case sections are combined to construct a comprehensive reference processing solution for the target case; Output the comprehensive reference treatment plan and sample cases for the target cases.

Images