Methods and systems for managing chat logs based on storage networks

Abstract

This invention discloses a method and system for managing chat logs based on a storage network, relating to the field of log management technology. The method includes the following steps: S1, constructing an interaction matrix; S2, constructing an association matrix; S3, calculating participation; S4, outputting summary text; and S5, assessing risk coefficients. This invention effectively divides different chat logs by constructing an interaction matrix and marks chat logs within the same conversation segment. After the initial construction of the interaction matrix, to further conserve storage resources, the position values ​​in the matrix are optimized: if the logical correlation between two chat logs is weak, the value at the corresponding matrix position is reset to zero, thus avoiding redundant storage; while for strongly correlated log pairs, their position value is retained as 1. This design quantifies the connection strength between chat logs, thereby reflecting the closeness of the relationship between users.

Description

Technical Field

[0001] This invention relates to the field of record management technology, specifically to a method and system for managing chat logs based on a storage network. Background Technology

[0002] A chat log management system based on a storage network enables centralized storage and refined management of chat logs. It automatically associates the user ID and content of each log entry, provides data access and content auditing, and ensures the security, compliance, and traceability of chat data during storage, transmission, and use. This meets the management needs of enterprises for internal communication. Patent application number 202110547942.1 discloses "A method and system for identifying WeChat chat logs by integrating conversation scenario information, including: acquiring WeChat chat log text and preprocessing it; determining the word vector of each text and, based on social support theory, integrating the conversation scenario of each text..." The process involves classifying text into categories and outputting a classification vector for each text. Two text classification vectors are merged to obtain the corresponding text scene information. A domain-adaptive BERT model is used to extract semantic features from any two texts. Combining the scene information and semantic features of the two texts, the contextual relationship between them is predicted, and the probability that the two texts belong to a contextual relationship is output. By comparing the probability of each text belonging to a contextual relationship with other texts, the two texts with the highest probability of a contextual relationship are selected to establish a dialogue relationship, thus achieving dialogue recognition in WeChat chat history. This invention integrates the contextual information of the text, resulting in higher accuracy for the dialogue recognition model.

[0003] The aforementioned existing technologies have solved problems such as the inability to integrate conversation context information for chat record recognition. However, when the system is running, it needs to browse through a massive amount of chat records one by one to assess risks and understand the context. This not only causes a huge waste of human resources, but the comprehensive review process itself will seriously infringe on users' personal privacy. At the same time, the lack of intelligent risk assessment based on participation makes it impossible for the system to identify key users with high influence in group chats, resulting in serious inaccuracies in risk assessment of key figures. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for managing chat logs based on a storage network, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for managing chat logs based on a storage network, comprising the following steps:

[0006] S1. Constructing the interaction matrix: Obtain all chat records in the group, divide them into multiple conversation segments, construct an interaction matrix based on the conversation segments corresponding to each chat record, and calculate the number of connections between each chat record and other chat records in the same conversation segment. Optimize the interaction matrix according to the number of connections.

[0007] S2. Constructing the association matrix: After obtaining the serial numbers of all users in the group, construct the association matrix based on the interaction matrix and the user serial number corresponding to each chat record;

[0008] S3. Statistical participation: Analyze the other user numbers associated with each user number according to the position value in the association matrix, and calculate the participation of all user numbers based on the position value of each user number and the other user numbers.

[0009] S4. Output summary text: After receiving all chat records in the conversation segment, use the Transformer model to analyze them and output the summary text corresponding to the current conversation segment;

[0010] S5. Assess Risk Factors: Obtain classified project records, set corresponding thresholds according to the project classification level in each project record, extract the entity list corresponding to the summary text of the conversation fragment, compare the entity list corresponding to the summary with each project record, calculate the correlation coefficient, and trigger an emergency warning when the correlation coefficient exceeds the threshold. Obtain the project classification level and the value coefficient corresponding to the current user's access permissions, and determine the risk factor of the current user's number based on the value coefficient.

[0011] Preferably, step S1 specifically includes the following steps:

[0012] S101. Retrieve chat history from the group. and the corresponding generation time ,in This indicates the total number of chat records. Indicates the first A chat log, Indicates the first The creation time of each chat record, based on the total number of records. Construct a size of An interaction matrix is ​​generated, where the rows and columns represent the serial numbers of chat records, and all positions in the matrix are initialized to 2.

[0013] S102, Select chat history After using it as the base record, iterate through... The chat history that followed Corresponding generation time If the benchmark record Time of creation and Time of creation If the difference between them is less than or equal to the threshold, then... Store to Within the same conversation segment, it iterates through the next chat record; otherwise, it iterates through the next chat record. The chat logs generated after traversing the baseline record serve as the new baseline record;

[0014] S103. Analyze all conversation segments and randomly select two chat records. and If the first A chat log With the A chat log If they are in the same session segment, then the first one in the interaction matrix will be... Line number Column position points and the first Line number All column positions are set to 1; otherwise, the first column position in the interaction matrix is ​​set to 1. Line number Column position points and the first Line number All column position points are set to 0, where The parameter is used to repeat the operation until there are no more points with a value of 2 in the interaction matrix.

[0015] S104. Retrieve chat history from the group. The text is then cleaned and segmented, and the processed record text is fed into the BERT model for analysis to obtain the feature vectors of the corresponding records. The specific process of cleaning and segmentation is as follows:

[0016] a. After converting the original chat log text into a format structure suitable for the BERT model, remove invalid information and use the corresponding BERT tokenizer to split the sentences in the text into sub-words in the model's vocabulary;

[0017] b. Insert a start marker at the beginning of each sentence and an end marker at the end, convert the words after segmentation of the recorded text into corresponding numbers, and use attention masks to fill each record to the same length;

[0018] c. Input the tags, corresponding numbers of words, and attention mask padding contained in the records into the BERT model for analysis, and output the feature vector of the chat records;

[0019] S105. After calculating the feature vectors of all chat records, combine them into a feature matrix. Analyze the fused feature vector of each chat record using the interaction matrix and the feature matrix. Based on the fused feature vectors of all chat records, determine whether all positions with a value of 1 in the interaction matrix are reasonable. When the fused feature vector in the interaction matrix... Line number When the value of the column position point is 1, according to the first... The first chat log and the first The fusion feature vector of chat logs and Calculate the feature vector of this location point. ,in The value is mapped to the number of connections. If the current number of connections is greater than the threshold, the value of the location point is determined to be reasonable. Otherwise, the value of the location point is determined to be unreasonable and the value is changed to 0.

[0020] Preferably, step S2 specifically includes the following steps:

[0021] S201. Obtain the user serial number in the group. ,in This indicates the total number of users in the group. Indicates the first Individual user serial number, based on the total number of users in the group. Construct a size of The correlation matrix, in which rows and columns represent user serial numbers, and all positions in the matrix are initialized to 0;

[0022] S202. Read all positions with a value of 1 in the interaction matrix. Randomly select an unmarked position in the interaction matrix, mark it, and determine the row number of that position. Column number is Then, directly extract the number of connections at that location. And the first chat history With the chat history User serial number and ,in Indicates the serial number, according to the user serial number. and Determine the first in the correlation matrix Line number Current position value of the column ,according to and Calculate the new position value ,in ,Will The value is updated to Repeat the operation until all positions with a value of 1 in the interaction matrix are marked, and then output the correlation matrix.

[0023] Preferably, step S3 specifically includes the following steps:

[0024] S301, Extract the first element from the correlation matrix. After considering all position values ​​in a row, remove the first... Line number Column position value After determining the position value as zero, the remaining position values ​​are used to analyze the relationship with the user serial number. Other related user serial numbers ,in Indicates the relationship with the user serial number The associated first Each user serial number will Store to candidate set In the middle, according to the user serial number Corresponding position value Calculate and Interaction coefficient between ,in , Indicates the relationship between the user number and the correlation matrix. The associated first The position value corresponding to each user's serial number. express With user serial number The associated first User serial number Interaction coefficient between them This represents a parameter and counts the serial numbers of all users. Number of chat messages sent ,in express The number of chat messages sent depends on Calculate Corresponding level of enthusiasm ,in , express The corresponding level of positivity is then repeated until every row in the association matrix has been extracted.

[0025] S302. Determine the maximum number of iterations, and set the current number of iterations and the participation of all user numbers to the default values.

[0026] Preferably, step S3 further includes the following steps:

[0027] S303. After reading the current iteration number, use the participation analysis algorithm based on the user sequence number. and Interaction coefficient between and The user's activity level is used to calculate the user's serial number. Participation level;

[0028] S304. After determining the participation degree of all user numbers, automatically increment the current iteration count by 1, and determine whether the current iteration count is equal to the maximum iteration count. If the current iteration count is equal to the maximum iteration count, directly output the participation degree of all current user numbers; otherwise, use the participation degree of all current users to calculate the participation degree of the next iteration process.

[0029] Preferably, step S4 specifically includes the following steps:

[0030] S401. After receiving all chat records in the conversation segment, sort them by time, use the data in the storage network to determine the user number and text content corresponding to each chat record, use the Stanford tool to count all nouns in the text of different chat records that point to the same entity, and call the reply relationship parser to analyze the relationship structure between chat records, thereby outputting the reply relationship between each chat record and other chat records.

[0031] S402. Transmit the user serial number, text content, and all nouns in the text that point to the same entity for each chat record to the encoder of the Transformer for analysis to obtain the initial feature vector corresponding to the chat record.

[0032] S403. The initial feature vector of the chat records and the reply relationship between each chat record and other chat records are transmitted to the graph attention network for analysis, and the fused feature vector is output. The decoder of the Transformer is used to determine the summary text corresponding to the current session fragment based on the fused feature vector.

[0033] Preferably, step S5 specifically includes the following steps:

[0034] S501. Obtain classified project records, which include project number, project name, core keyword list, core personnel list and project classification level. Build a classified project knowledge base using all project records and set corresponding thresholds according to the project classification level in each project record.

[0035] S502. After extracting the summary text of the conversation fragment, it is transmitted to the NER model for analysis to obtain the entity list corresponding to the summary text. The entity list includes personnel name, organization name, geographical location, time, project name, core keywords and event.

[0036] S503. Compare the entity list corresponding to the abstract with each project record in the classified project knowledge base, and calculate the relevance coefficient according to the matching dimensions. The matching dimensions include project name matching, core keyword matching, and core personnel matching. The specific process for calculating the relevance coefficient according to the matching dimensions is as follows:

[0037] a. Extract the entity list from the session summary and obtain the project records from the classified project knowledge base, setting the relevance coefficient of all project records matched to zero;

[0038] b. When matching project names and core personnel, the project names and personnel names in the entity list are matched with the project names and core personnel lists in each project record in the classified project knowledge base. For each successful match, a preset high score is added to the original relevance coefficient.

[0039] c. When matching core keywords, the core keywords in the entity list are matched with the core keywords in each project record in the classified project knowledge base. For each matched keyword, a preset low score is added to the original relevance coefficient.

[0040] d. After matching is complete, output the relevance coefficients for all project records;

[0041] S504. Determine the correlation coefficient of all project records. If the correlation coefficient exceeds the threshold, trigger an emergency warning; otherwise, do not perform any operation.

[0042] S505. After receiving an emergency warning, retrieve all chat records in the conversation segment and their corresponding user serial numbers, count the records with correlation coefficients exceeding the threshold, and query the access permissions of all user serial numbers in the conversation segment according to the item number in the record.

[0043] S506. Obtain the value coefficient corresponding to the project's security level and the current user's access permissions. and ,like Then directly apply the penalty coefficient. Set to zero, if According to and Calculate ,in ,like According to and Calculate ,in ,like According to and Calculate ,in , Both represent constraint values, and , Indicates the correction factor. Represents the natural constant;

[0044] S507. After obtaining the participation degree and penalty coefficient of the current user number, perform a weighted summation to obtain the risk coefficient corresponding to the current user number, and output the risk coefficient through the visualization interface.

[0045] The chat log management system based on the storage network includes a matrix generation unit, an engagement calculation unit, a summary determination unit, and a risk assessment unit.

[0046] The matrix generation unit obtains all chat records in the group, divides them into multiple conversation segments, constructs an interaction matrix based on the conversation segments corresponding to each chat record, calculates the number of connections between each chat record and other chat records in the same conversation segment, optimizes the interaction matrix according to the number of connections, and after obtaining the serial numbers of all users in the group, constructs an association matrix based on the interaction matrix and the serial numbers corresponding to each chat record.

[0047] The participation calculation unit analyzes the other user numbers associated with each user number according to the position value in the association matrix, and calculates the participation of each user number based on the position value corresponding to each user number and other user numbers.

[0048] After receiving all chat records in the conversation segment, the summary determination unit uses the Transformer model to analyze them and outputs the summary text corresponding to the current conversation segment.

[0049] The risk assessment unit acquires classified project records, sets corresponding thresholds according to the project classification level in each project record, extracts the entity list corresponding to the summary text of the session fragment, compares the entity list corresponding to the summary with each project record, calculates the correlation coefficient, and triggers an emergency warning when the correlation coefficient exceeds the threshold. It then acquires the project classification level and the value coefficient corresponding to the current user's access permissions, and determines the risk coefficient of the current user's number based on the value coefficient.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] 1. This invention effectively divides different chat records by constructing an interaction matrix and marks chat records within the same conversation segment. After the initial construction of the interaction matrix is ​​completed, to further save storage resources, the position values ​​in the matrix are optimized: if the logical relationship between two chat records is weak, the value of the corresponding matrix position is reset to zero to avoid redundant storage; while for record pairs with strong correlation, their position value is retained as 1. This design can quantify the connection strength between chat records, thereby reflecting the closeness of the relationship between users.

[0052] 2. This invention constructs an association matrix and calculates the interaction coefficient between user numbers based on the values ​​at corresponding positions in the matrix. Simultaneously, it assesses the activity level corresponding to a user number by combining the ratio of the number of user messages to the total number of messages. By combining the user's interaction coefficient and activity level, the user's participation level in the group can be determined. This indicator can effectively measure the user's interaction reach within the group. In subsequent confidentiality risk analysis, the potential risk coefficient of each user is calculated based on their participation level, thereby quantifying the risk.

[0053] 3. This invention can automatically generate text summaries based on conversation segments, eliminating the need for management to review the original chat content line by line. This not only protects user privacy but also provides timely warnings of potential leaks, ensuring the information security of project content before its release. In calculating the risk coefficient, we combine the user's participation level in the group with their permission scope as a key consideration, effectively compensating for the limitations of a single behavioral indicator and making the risk assessment results more convincing. Attached Figure Description

[0054] Figure 1 An overall method flowchart is provided for embodiments of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example:

[0057] Please see Figure 1 This invention provides a technical solution: a method for managing chat logs based on a storage network, comprising the following steps:

[0058] S1. Constructing the interaction matrix: Obtain all chat records in the group, divide them into multiple conversation segments, construct an interaction matrix based on the conversation segments corresponding to each chat record, and calculate the number of connections between each chat record and other chat records in the same conversation segment. Optimize the interaction matrix according to the number of connections.

[0059] S2. Constructing the association matrix: After obtaining the serial numbers of all users in the group, construct the association matrix based on the interaction matrix and the user serial number corresponding to each chat record;

[0060] S3. Statistical participation: Analyze the other user numbers associated with each user number according to the position value in the association matrix, and calculate the participation of all user numbers based on the position value of each user number and the other user numbers.

[0061] S4. Output summary text: After receiving all chat records in the conversation segment, use the Transformer model to analyze them and output the summary text corresponding to the current conversation segment;

[0062] S5. Assess Risk Factors: Obtain classified project records, set corresponding thresholds according to the project classification level in each project record, extract the entity list corresponding to the summary text of the conversation fragment, compare the entity list corresponding to the summary with each project record, calculate the correlation coefficient, and trigger an emergency warning when the correlation coefficient exceeds the threshold. Obtain the project classification level and the value coefficient corresponding to the current user's access permissions, and determine the risk factor of the current user's number based on the value coefficient.

[0063] S1 specifically includes the following steps:

[0064] S101. Retrieve chat history from the group. and the corresponding generation time ,in This indicates the total number of chat records. Indicates the first A chat log, Indicates the first The creation time of each chat record, based on the total number of records. Construct a size of An interaction matrix is ​​generated, where the rows and columns represent the serial numbers of chat records, and all positions in the matrix are initialized to 2.

[0065] S102, Select chat history After using it as the base record, iterate through... The chat history that followed Corresponding generation time If the benchmark record Time of creation and Time of creation If the difference between them is less than or equal to the threshold, then... Store to Within the same conversation segment, it iterates through the next chat record; otherwise, it iterates through the next chat record. The chat logs generated after traversing the baseline record serve as the new baseline record;

[0066] S103. Analyze all conversation segments and randomly select two chat records. and If the first A chat log With the A chat log If they are in the same session segment, then the first one in the interaction matrix will be... Line number Column position points and the first Line number All column positions are set to 1; otherwise, the first column position in the interaction matrix is ​​set to 1. Line number Column position points and the first Line number All column position points are set to 0, where The parameter is used to repeat the operation until there are no more points with a value of 2 in the interaction matrix.

[0067] S104. Retrieve chat history from the group. The text is then cleaned and segmented, and the processed record text is fed into the BERT model for analysis to obtain the feature vectors of the corresponding records. The specific process of cleaning and segmentation is as follows:

[0068] a. After converting the original chat log text into a format structure suitable for the BERT model, remove invalid information and use the corresponding BERT tokenizer to split the sentences in the text into sub-words in the model's vocabulary;

[0069] b. Insert a start marker at the beginning of each sentence and an end marker at the end, convert the words after segmentation of the recorded text into corresponding numbers, and use attention masks to fill each record to the same length;

[0070] c. Input the tags, corresponding numbers of words, and attention mask padding contained in the records into the BERT model for analysis, and output the feature vector of the chat records;

[0071] S105. After calculating the feature vectors of all chat records, combine them into a feature matrix. Analyze the fused feature vector of each chat record using the interaction matrix and the feature matrix. Based on the fused feature vectors of all chat records, determine whether all positions with a value of 1 in the interaction matrix are reasonable. When the fused feature vector in the interaction matrix... Line number When the value of the column position point is 1, according to the first... The first chat log and the first The fusion feature vector of chat logs and Calculate the feature vector of this location point. ,in This is mapped to the number of connections. If the current number of connections is greater than a threshold, the value of the location point is considered reasonable; otherwise, the value of the location point is considered unreasonable and is changed to 0. The specific process of analyzing the fusion feature vector of each chat record using the interaction matrix and feature matrix is ​​as follows:

[0072] a. Select any chat history Statistical interaction matrix For all points in a row with a value of 1, determine the column number of each point with a value of 1. Using the current column number as the sequence number, count all chat records with the same sequence number and set that as the first... A chat log The adjacent records, of which Indicates parameters;

[0073] b. Determine the adjacent records of all chat logs and count the number of records in the next chat log. A chat log eigenvectors and adjacent records eigenvectors ,in This refers to chat history. The Adjacent records, Indicates adjacent records eigenvectors, This represents the number of adjacent records, and the learning weight matrix corresponding to the current network layer is determined using the GNN model. ,use right and After performing a linear transformation, we obtain and ,in Representing records The eigenvector after linear transformation Representing records The eigenvector after linear transformation, according to and Calculate the chat history With adjacent records Attention coefficient between ,in This refers to chat history. With adjacent records Attention coefficient between them ;

[0074] c. After normalizing the attention coefficient using the Softmax function, the chat log is obtained. With adjacent records Weighting coefficients between ,in This refers to chat history. With adjacent records The weighting coefficients between them , Indicates parameters;

[0075] d. Based on the weighting coefficients and the eigenvectors after linear transformation Calculate the chat history New feature vector ,in ;

[0076] e. After determining the new feature vectors for all chat records, calculate the feature vector for the next layer based on the feature vector corresponding to the current layer, and repeat the operation until the current layer reaches the optimal layer. Finally, the final feature vector of each chat record is output and used as the fused feature vector of the chat records;

[0077] S2 specifically includes the following steps:

[0078] S201. Obtain the user serial number in the group. ,in This indicates the total number of users in the group. Indicates the first Individual user serial number, based on the total number of users in the group. Construct a size of The correlation matrix, in which rows and columns represent user serial numbers, and all positions in the matrix are initialized to 0;

[0079] S202. Read all positions with a value of 1 in the interaction matrix. Randomly select an unmarked position in the interaction matrix, mark it, and determine the row number of that position. Column number is Then, directly extract the number of connections at that location. And the first chat history With the chat history User serial number and ,in Indicates the serial number, according to the user serial number. and Determine the first in the correlation matrix Line number Current position value of the column ,according to and Calculate the new position value ,in ,Will The value is updated to Repeat the operation until all points with a value of 1 in the interaction matrix are marked, and then output the association matrix.

[0080] S3 specifically includes the following steps:

[0081] S301, Extract the first element from the correlation matrix. After considering all position values ​​in a row, remove the first... Line number Column position value After determining the position value as zero, the remaining position values ​​are used to analyze the relationship with the user serial number. Other related user serial numbers ,in Indicates the relationship with the user serial number The associated first Each user serial number will Store to candidate set In the middle, according to the user serial number Corresponding position value Calculate and Interaction coefficient between ,in , Indicates the relationship between the user number and the correlation matrix. The associated first The position value corresponding to each user's serial number. express With user serial number The associated first User serial number Interaction coefficient between them This represents a parameter and counts the serial numbers of all users. Number of chat messages sent ,in express The number of chat messages sent depends on Calculate Corresponding level of enthusiasm ,in , express The corresponding level of positivity is then repeated until every row in the association matrix has been extracted.

[0082] S302. Determine the maximum number of iterations, and set the current number of iterations and the participation degree of all user numbers to the default values;

[0083] S3 also includes the following steps:

[0084] S303. After reading the current iteration number, use the participation analysis algorithm based on the user sequence number. and Interaction coefficient between and The user's activity level is used to calculate the user's serial number. The engagement level, and the specific algorithm for engagement analysis are as follows:

[0085]

[0086] in, Indicates user serial number participation This indicates the total number of user serial numbers. Indicates user serial number and user serial number Other related user serial numbers Interaction coefficient between them Indicates user serial number The corresponding level of enthusiasm, Indicates user serial number participation This represents the adjustment coefficient. Indicates user serial number The candidate set;

[0087] S304. After determining the participation degree of all user numbers, automatically increment the current iteration number by 1, and determine whether the current iteration number is equal to the maximum iteration number. If the current iteration number is equal to the maximum iteration number, directly output the participation degree of all current user numbers; otherwise, use the participation degree of all current users to calculate the participation degree of the next iteration process.

[0088] S4 specifically includes the following steps:

[0089] S401. After receiving all chat records in the conversation segment, sort them by time, use the data in the storage network to determine the user number and text content corresponding to each chat record, use the Stanford tool to count all nouns in the text of different chat records that point to the same entity, and call the reply relationship parser to analyze the relationship structure between chat records, thereby outputting the reply relationship between each chat record and other chat records.

[0090] S402. Transmit the user serial number, text content, and all nouns in the text that point to the same entity for each chat record to the encoder of the Transformer for analysis to obtain the initial feature vector corresponding to the chat record.

[0091] S403. The initial feature vector of the chat records and the reply relationship between each chat record and other chat records are transmitted to the graph attention network for analysis, and the fused feature vector is output. The decoder of the Transformer is used to determine the summary text corresponding to the current conversation segment based on the fused feature vector.

[0092] S5 specifically includes the following steps:

[0093] S501. Obtain classified project records, which include project number, project name, core keyword list, core personnel list and project classification level. Build a classified project knowledge base using all project records and set corresponding thresholds according to the project classification level in each project record.

[0094] S502. After extracting the summary text of the conversation fragment, it is transmitted to the NER model for analysis to obtain the entity list corresponding to the summary text. The entity list includes personnel name, organization name, geographical location, time, project name, core keywords and event.

[0095] S503. Compare the entity list corresponding to the abstract with each project record in the classified project knowledge base, and calculate the relevance coefficient according to the matching dimensions. The matching dimensions include project name matching, core keyword matching, and core personnel matching. The specific process for calculating the relevance coefficient according to the matching dimensions is as follows:

[0096] a. Extract the entity list from the session summary and obtain the project records from the classified project knowledge base, setting the relevance coefficient of all project records matched to zero;

[0097] b. When matching project names and core personnel, the project names and personnel names in the entity list are matched with the project names and core personnel lists in each project record in the classified project knowledge base. For each successful match, a preset high score is added to the original relevance coefficient.

[0098] c. When matching core keywords, the core keywords in the entity list are matched with the core keywords in each project record in the classified project knowledge base. For each matched keyword, a preset low score is added to the original relevance coefficient.

[0099] d. After matching is complete, output the relevance coefficients for all project records;

[0100] S504. Determine the correlation coefficient of all project records. If the correlation coefficient exceeds the threshold, trigger an emergency warning; otherwise, do not perform any operation.

[0101] S505. After receiving an emergency warning, retrieve all chat records in the conversation segment and their corresponding user serial numbers, count the records with correlation coefficients exceeding the threshold, and query the access permissions of all user serial numbers in the conversation segment according to the item number in the record.

[0102] S506. Obtain the value coefficient corresponding to the project's security level and the current user's access permissions. and ,like Then directly apply the penalty coefficient. Set to zero, if According to and Calculate ,in ,like According to and Calculate ,in ,like According to and Calculate ,in , Both represent constraint values, and , Indicates the correction factor. Represents the natural constant;

[0103] S507. After obtaining the participation degree and penalty coefficient of the current user number, perform a weighted summation to obtain the risk coefficient corresponding to the current user number, and output the risk coefficient through the visualization interface.

[0104] A chat log management system based on a storage network includes a matrix generation unit, an engagement calculation unit, a summary determination unit, and a risk assessment unit.

[0105] The matrix generation unit obtains all chat records in the group, divides them into multiple conversation segments, constructs an interaction matrix based on the conversation segments corresponding to each chat record, calculates the number of connections between each chat record and other chat records in the same conversation segment, optimizes the interaction matrix according to the number of connections, and after obtaining the serial numbers of all users in the group, constructs an association matrix based on the interaction matrix and the serial numbers corresponding to each chat record.

[0106] The participation calculation unit analyzes the position values ​​in the association matrix to identify the other user numbers associated with each user number, and calculates the participation of all user numbers based on the position values ​​corresponding to each user number and the other user numbers.

[0107] After receiving all chat records in a conversation segment, the summary determination unit uses the Transformer model to analyze them and outputs the summary text corresponding to the current conversation segment.

[0108] The risk assessment unit obtains records of classified projects, sets corresponding thresholds according to the project classification level in each project record, extracts the entity list corresponding to the summary text of the conversation fragment, compares the entity list corresponding to the summary with each project record, calculates the correlation coefficient, and triggers an emergency warning when the correlation coefficient exceeds the threshold. It then obtains the project classification level and the value coefficient corresponding to the current user's access permissions, and determines the risk coefficient of the current user's number based on the value coefficient.

[0109] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for managing chat records based on a storage network, characterized in that, The method includes the following steps: S1. Constructing the interaction matrix: Obtain all chat records in the group, divide them into multiple conversation segments, construct an interaction matrix based on the conversation segments corresponding to each chat record, and calculate the number of connections between each chat record and other chat records in the same conversation segment. Optimize the interaction matrix according to the number of connections. S2. Constructing the association matrix: After obtaining the serial numbers of all users in the group, construct the association matrix based on the interaction matrix and the user serial number corresponding to each chat record; S3. Statistical participation: Analyze the other user numbers associated with each user number according to the position value in the association matrix, and calculate the participation of all user numbers based on the position value of each user number and the other user numbers. S4. Output summary text: After receiving all chat records in the conversation segment, use the Transformer model to analyze them and output the summary text corresponding to the current conversation segment; S5. Assess Risk Factors: Obtain classified project records, set corresponding thresholds according to the project classification level in each project record, extract the entity list corresponding to the summary text of the conversation fragment, compare the entity list corresponding to the summary with each project record, calculate the correlation coefficient, and trigger an emergency warning when the correlation coefficient exceeds the threshold. Obtain the project classification level and the value coefficient corresponding to the current user's access permissions, and determine the risk factor of the current user's number based on the value coefficient.

2. The method of claim 1, wherein: S1 specifically includes the following steps: S101、Obtain chat records in a group and corresponding generation time wherein represents the total number of chat records, represents the th chat record, represents the generation time of the th chat record, and the total number of records is used to construct an interaction matrix with a size of The row and column in the matrix represent the serial number of the chat record, and the initial value of all positions in the matrix is 2. S102, Select chat history After using it as the base record, iterate through... The chat history that followed Corresponding generation time If the benchmark record Time of creation and Time of creation If the difference between them is less than or equal to the threshold, then... Store to Within the same conversation segment, it iterates through the next chat record; otherwise, it iterates through the next chat record. The chat logs generated after traversing the baseline record serve as the new baseline record; S103. Analyze all conversation segments and randomly select two chat records. and If the first A chat log With the A chat log If they are in the same session segment, then the first one in the interaction matrix will be... Line number Column position points and the first Line number All column positions are set to 1; otherwise, the first column position in the interaction matrix is ​​set to 1. Line number Column position points and the first Line number All column position points are set to 0, where The parameter is used to repeat the operation until there are no more points with a value of 2 in the interaction matrix. S104. Retrieve chat history from the group. The text is cleaned and segmented, and the processed record text is transmitted to the BERT model for analysis to obtain the feature vector of the corresponding record. S105. After calculating the feature vectors of all chat records, combine them into a feature matrix. Analyze the fused feature vector of each chat record using the interaction matrix and the feature matrix. Based on the fused feature vectors of all chat records, determine whether all positions with a value of 1 in the interaction matrix are reasonable. When the fused feature vector in the interaction matrix... Line number When the value of the column position point is 1, according to the first... The first chat log and the first The fusion feature vector of chat logs and Calculate the feature vector of this location point. ,in The value is mapped to the number of connections. If the current number of connections is greater than the threshold, the value of the location point is determined to be reasonable. Otherwise, the value of the location point is determined to be unreasonable and the value is changed to 0.

3. The method for managing chat logs based on a storage network according to claim 1, characterized in that: S2 specifically includes the following steps: S201. Obtain the user serial number in the group. ,in This indicates the total number of users in the group. Indicates the first Individual user serial number, based on the total number of users in the group. Construct a size of The correlation matrix, in which rows and columns represent user serial numbers, and all positions in the matrix are initialized to 0; S202. Read all positions with a value of 1 in the interaction matrix. Randomly select an unmarked position in the interaction matrix, mark it, and determine the row number of that position. Column number is Then, directly extract the number of connections at that location. And the first chat history With the chat history User serial number and ,in Indicates the serial number, according to the user serial number. and Determine the first in the correlation matrix Line number Current position value of the column ,according to and Calculate the new position value ,in ,Will The value is updated to Repeat the operation until all positions with a value of 1 in the interaction matrix are marked, and then output the correlation matrix.

4. The method for managing chat logs based on a storage network according to claim 1, characterized in that: S3 specifically includes the following steps: S301, Extract the first element from the correlation matrix. After considering all position values ​​in a row, remove the first... Line number Column position value After determining the position value as zero, the remaining position values ​​are used to analyze the relationship with the user serial number. Other related user serial numbers ,in Indicates the relationship with the user serial number The associated first Each user serial number will Store to candidate set In the middle, according to the user serial number Corresponding position value Calculate and Interaction coefficient between ,in , Indicates the relationship between the user number and the correlation matrix. The associated first The position value corresponding to each user's serial number. express With user serial number The associated first User serial number Interaction coefficient between them This represents a parameter and counts the serial numbers of all users. Number of chat messages sent ,in express The number of chat messages sent depends on Calculate Corresponding level of enthusiasm ,in , express The corresponding level of positivity is then repeated until every row in the association matrix has been extracted. S302. Determine the maximum number of iterations, and set the current number of iterations and the participation of all user numbers to the default values.

5. The method for managing chat logs based on a storage network according to claim 4, characterized in that: S3 further includes the following steps: S303. After reading the current iteration number, use the participation analysis algorithm based on the user sequence number. and Interaction coefficient between and The user's activity level is used to calculate the user's serial number. Participation level; S304. After determining the participation degree of all user numbers, automatically increment the current iteration count by 1, and determine whether the current iteration count is equal to the maximum iteration count. If the current iteration count is equal to the maximum iteration count, directly output the participation degree of all current user numbers; otherwise, use the participation degree of all current users to calculate the participation degree of the next iteration process.

6. The method for managing chat logs based on a storage network according to claim 1, characterized in that: S4 specifically includes the following steps: S401. After receiving all chat records in the conversation segment, sort them by time, use the data in the storage network to determine the user number and text content corresponding to each chat record, use the Stanford tool to count all nouns in the text of different chat records that point to the same entity, and call the reply relationship parser to analyze the relationship structure between chat records, thereby outputting the reply relationship between each chat record and other chat records. S402. Transmit the user serial number, text content, and all nouns in the text that point to the same entity for each chat record to the encoder of the Transformer for analysis to obtain the initial feature vector corresponding to the chat record. S403. The initial feature vector of the chat records and the reply relationship between each chat record and other chat records are transmitted to the graph attention network for analysis, and the fused feature vector is output. The decoder of the Transformer is used to determine the summary text corresponding to the current session fragment based on the fused feature vector.

7. The method for managing chat logs based on a storage network according to claim 1, characterized in that: S5 specifically includes the following steps: S501. Obtain classified project records, construct a classified project knowledge base using all project records, and set corresponding thresholds according to the project classification level in each project record. S502. After extracting the summary text of the conversation fragment, transmit it to the NER model for analysis to obtain the entity list corresponding to the summary text; S503. Compare the entity list corresponding to the abstract with the project records in the classified project knowledge base one by one, and calculate the relevance coefficient according to the matching dimensions, including project name matching, core keyword matching, and core personnel matching. S504. Determine the correlation coefficient of all project records. If the correlation coefficient exceeds the threshold, trigger an emergency warning; otherwise, do not perform any operation. S505. After receiving an emergency warning, retrieve all chat records in the conversation segment and their corresponding user serial numbers, count the records with correlation coefficients exceeding the threshold, and query the access permissions of all user serial numbers in the conversation segment according to the item number in the record. S506. Obtain the value coefficient corresponding to the project's security level and the current user's access permissions. and ,like Then directly apply the penalty coefficient. Set to zero, if According to and Calculate ,in ,like According to and Calculate ,in ,like According to and Calculate ,in , Both represent constraint values, and , Indicates the correction factor. Represents the natural constant; S507. After obtaining the participation degree and penalty coefficient of the current user number, perform a weighted summation to obtain the risk coefficient corresponding to the current user number, and output the risk coefficient through the visualization interface.

8. A chat log management system based on a storage network, characterized in that, The chat log management system is applicable to the chat log management method based on storage network as described in any one of claims 1-7, including a matrix generation unit, an engagement calculation unit, a summary determination unit, and a risk assessment unit; The matrix generation unit obtains all chat records in the group, divides them into multiple conversation segments, constructs an interaction matrix based on the conversation segments corresponding to each chat record, calculates the number of connections between each chat record and other chat records in the same conversation segment, optimizes the interaction matrix according to the number of connections, and after obtaining the serial numbers of all users in the group, constructs an association matrix based on the interaction matrix and the serial numbers corresponding to each chat record. The participation calculation unit analyzes the other user numbers associated with each user number according to the position value in the association matrix, and calculates the participation of each user number based on the position value corresponding to each user number and other user numbers. After receiving all chat records in the conversation segment, the summary determination unit uses the Transformer model to analyze them and outputs the summary text corresponding to the current conversation segment. The risk assessment unit acquires classified project records, sets corresponding thresholds according to the project classification level in each project record, extracts the entity list corresponding to the summary text of the session fragment, compares the entity list corresponding to the summary with each project record, calculates the correlation coefficient, and triggers an emergency warning when the correlation coefficient exceeds the threshold. It then acquires the project classification level and the value coefficient corresponding to the current user's access permissions, and determines the risk coefficient of the current user's number based on the value coefficient.

Images