Legal consultation service management system

Abstract

The invention relates to the technical field of consultation service management, and discloses a legal consultation service management system, which comprises a user side, a server side and an interaction side, through the arrangement of the intelligent service module, the user portrait generation unit and the dynamic knowledge graph construction unit are favorable for obtaining a user portrait and constructing a legal knowledge graph by combining external data on the basis of data features fused by the data acquisition and fusion unit, and dynamically adjusting the legal knowledge graph by adopting an incremental updating mechanism; therefore, matching is not merely dependent on keyword extraction and a static database, a cross-modal attention mechanism is introduced to fuse data and analyze deep semantic logic of user questions, and semantic understanding is performed on the user questions based on user portraits, so that the understanding is more comprehensive, and the user experience is improved. And meanwhile, the constructed legal knowledge graph is dynamically adjusted by adopting an incremental updating mechanism, a data island is broken, and data association is performed by utilizing the knowledge graph.

Description

Technical Field

[0001] The present invention relates to the technical field of consulting service management, and more particularly to a legal consulting service management system. Background Art

[0002] The public document with publication number CN119850374A discloses a legal consulting service information management system and method based on big data, including: a consultation partitioning module, which is used to obtain consultation information during user consultation, and judge the consultation information relative to the consultation partition of the legal service according to the user's intention during consultation; a partition mapping module, which is used to obtain the legal partition of the consultation information and construct a mapping relationship between the consultation partition and the legal partition; a law article combination module, which is used to extract key partitions from the consultation partition and the legal partition, and judge the combination matching degree between multiple laws in the legal partition according to the key partition to obtain a law article bias combination; a case update module, which is used to verify the information of the legal partition based on the law article bias combination, identify the update status of the case in the legal partition, and update the mapping relationship between the consultation partition and the legal partition according to the update status of the case; thereby improving the accuracy and comprehensiveness of the legal consulting service.

[0003] However, the existing legal consulting service management system still has some shortcomings:

[0004] 1. They often rely on keyword extraction and static database matching. They only extract keywords through simple word segmentation and are unable to analyze the deep semantic logic of user questions, resulting in limited semantic understanding.

[0005] 2. The case database, legal provisions database, and lawyer information database are stored independently, lacking cross-modal data association and resulting in data silos;

[0006] 3. It is impossible to dynamically adjust the recommendation strategy for legal services based on the user’s historical consultation records, resulting in a high rate of repeated consultations and insufficient response efficiency. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a legal consulting service management system to solve the problems existing in the above-mentioned background technology.

[0008] The present invention provides the following technical solutions: a legal consulting service management system, comprising a user end, a service end, and an interaction end;

[0009] The user terminal is used to support user login and consultation information input;

[0010] The server provides services to users based on the consultation information input by the user, including intelligent service module, manual service module and feedback optimization module;

[0011] The intelligent service module includes a data acquisition and fusion unit, a user portrait generation unit, a dynamic knowledge graph construction unit, and a content generation unit;

[0012] The manual service module includes a user demand acquisition unit and a demand matching unit;

[0013] The feedback optimization module optimizes the feedback content of the intelligent service module and the manual service module based on the interactive terminal;

[0014] The interactive terminal is used for human-computer interaction display and provides feedback to the intelligent service module and the manual service module.

[0015] Preferably, the consulting information, namely the consulting content of the user's legal consulting service, includes text information, image information and voice information. The consulting information is input in the form of text input, voice input and image input. The consulting information input in the form of text input is text information, the consulting information input in the form of voice input is voice information, and the consulting information input in the form of image is image information.

[0016] Preferably, the data acquisition and fusion unit is used to acquire the consulting information input by the user end, and introduce a cross-modal attention mechanism to fuse the consulting information of different modalities to obtain the fused data features;

[0017] The user portrait generation unit obtains a user portrait based on the data features fused by the data acquisition and fusion unit and in combination with external data;

[0018] The dynamic knowledge graph construction unit is used to construct a legal knowledge graph and dynamically adjust the legal knowledge graph using an incremental update mechanism;

[0019] The content generation unit generates service content based on the user portrait of the user portrait generation unit and the legal knowledge graph of the dynamic knowledge graph construction unit.

[0020] Preferably, the external data includes user basic attribute data and interactive behavior data; the user basic attribute data includes user name, age, occupation and industry; the interactive behavior data includes historical consultation content, consultation frequency and user operation logs, and the user operation logs include the type of legal provisions clicked by the user, the length of time the user stays on a certain type of legal content, and the legal content collected and shared by the user.

[0021] Preferably, the user demand acquisition unit uses the data features fused by the data acquisition and fusion unit as the user demand features;

[0022] The demand matching unit performs demand matching based on user demand characteristics, obtains successfully matched lawyers, and connects the successfully matched lawyers with the user for a manual dialogue.

[0023] Preferably, the cross-modal attention mechanism is expressed as follows:

[0024] Among them, Attention(Q,K,V) represents the cross-modal attention mechanism; Q represents the query vector, which is generated by linear transformation of the feature matrix of the target modality and represents the target modality information that currently needs to be paid attention to; K represents the key vector, which is generated by the feature matrix of the source modality and stores the original features of the source modality for matching with the query vector; V represents the value vector, which comes from the source modality features, carries the actual semantic information of the source modality, and is output after weighting by the attention weight; the target modality is the fused data modality, and the source modality is the initial data modality; softmax represents the softmax function; M represents the mask matrix; Represents the scaling factor.

[0025] Preferably, the user profile is obtained based on a deep learning model, and the model includes an embedding layer, an attention layer, and a fully connected layer;

[0026] The embedding layer maps discrete features in external data into low-dimensional dense vectors;

[0027] The attention layer obtains the relevance weight of the user's historical behavior and the current consultation question based on the cross-modal attention mechanism;

[0028] The fully connected layer concatenates the output features of the embedding layer and the attention layer, and integrates them into a unified user portrait vector through nonlinear transformation for output.

[0029] Preferably, the legal knowledge graph mainly includes entities, relationships between entities, and entity attributes. The entities mainly include legal subjects, legal objects, and legal acts. The legal subjects are people and organizations that participate in legal relationships and enjoy rights and obligations. The legal objects are things and interests that are regulated by the rights and obligations of the subjects. The legal acts include prosecution, defense, mediation, and judgment. The entity attributes are used to describe entity information.

[0030] The inter-entity relationship is used to describe the legal logical association between entities. The inter-entity relationship includes legal reference relationship, causal relationship, conflict relationship and temporal relationship. The legal reference relationship refers to the cited legal provisions and clauses, the causal relationship refers to the cause leading to the result, the conflict relationship refers to the existence of a conflict between the two, and the temporal relationship refers to the relationship between the time sequence.

[0031] The incremental update mechanism is based on the incremental update formula and dynamically adjusts the legal knowledge graph.

[0032] Preferably, the incremental update formula is expressed as: ZL t+1 =ZL t ∪ΔZL; where ZL t+1 Represents the knowledge graph at time t+1, ZL t represents the knowledge graph at time t, ΔZL represents the incremental update content; the incremental update content is divided into two categories, namely entity update and entity relationship update, which are expressed as follows:

[0033] Among them, (E new ,A,B) represents entity update, E new Indicates a newly added entity, A indicates the attributes of the newly added entity, and B indicates the specific content of the newly added entity when the newly added entity is a legal object; (D new ,T,C) represents entity relationship update, D new Indicates a newly added entity relationship, and T indicates the relationship type.

[0034] Preferably, the demand matching formula is expressed as:

[0035] M ab =μ1·Sim(F a ,H b )+μ2·Avail(W b )+μ3·P b Among them, M ab represents the final matching score between the a-th user and the b-th lawyer; Sim(F a ,H b ) represents the professional similarity, that is, the demand feature F of the ath user a The lawyer's expertise characteristic H of the bth lawyer b The matching degree satisfies Sim(F a ,H b )∈[-1,1]; μ1 represents the weight of professional similarity; Avail(W b ) represents the time slack of the b-th lawyer, that is, the serviceability of the b-th lawyer within the time window W, satisfying Avail(W b )∈[0,1]; μ2 represents Avail(W b ) weight; P b Indicates the service quality score of the b-th lawyer, satisfying P b ∈[0,1]; μ3 represents P b weights; μ1 satisfies μ1∈(0,1), μ2 satisfies μ2∈(0,1), and μ3 satisfies μ3∈(0,1).

[0036] Technical effects and advantages of the present invention:

[0037] (1) The present invention is provided with an intelligent service module, which is conducive to obtaining user portraits based on the data features after fusion of the data acquisition and fusion unit through the user portrait generation unit and the dynamic knowledge graph construction unit, combining external data to construct a legal knowledge graph, and dynamically adjust the legal knowledge graph using an incremental update mechanism; thereby not only relying on the extraction of keywords and matching with a static database, but also introducing a cross-modal attention mechanism to fuse data, analyze the deep semantic logic of user questions, and at the same time, perform semantic understanding of user questions based on user portraits, so as to have a more comprehensive understanding, and at the same time adopt an incremental update mechanism to dynamically adjust the constructed legal knowledge graph, breaking the data island, and using the knowledge graph to associate data, and obtaining user portraits according to the user's historical consultation records to dynamically adjust the recommendation strategy of legal services, that is, the generated service content.

[0038] (2) The present invention is provided with a manual service module, which is conducive to obtaining user demand characteristics through the user demand acquisition unit and the demand matching unit and matching the needs based on the user demand characteristics, obtaining successfully matched lawyers, and connecting the successfully matched lawyers with the users for manual dialogue. Manual dialogue can be conducted according to the actual needs of the users, preventing repeated consultations and improving consultation efficiency. When the intelligent service cannot meet the needs of the users, manual dialogue can be conducted to improve the response rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a structural diagram of the legal consulting service management system of the present invention.

[0040] Figure 2 This is a structural diagram of the intelligent service module of the present invention.

[0041] Figure 3 It is a structural diagram of the manual service module of the present invention. DETAILED DESCRIPTION

[0042] The technical solutions of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the present invention. In addition, the forms of the various structures described in the following embodiments are merely examples. The legal consulting service management system involved in the present invention is not limited to the various structures described in the following embodiments. All other implementations obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0043] like Figure 1As shown, the present invention provides a legal consulting service management system, including a user end, a service end, and an interactive end; the user end is used to support user login and consultation information input; the consultation information is the consultation content of the user's legal consulting service, including but not limited to text information, image information, and voice information, etc. The consultation information can be input in the form of any one or more of text input, voice input, and image input. Consultation information input in the form of text input is text information, consultation information input in the form of voice input is voice information, and consultation information input in the form of image is image information;

[0044] The server provides services to users based on the consultation information input by the user, including intelligent service module, manual service module and feedback optimization module;

[0045] The intelligent service module includes a data acquisition and fusion unit, a user portrait generation unit, a dynamic knowledge graph construction unit, and a content generation unit;

[0046] The manual service module includes a user demand acquisition unit and a demand matching unit;

[0047] The feedback optimization module optimizes the feedback content of the intelligent service module and the manual service module based on the interactive terminal;

[0048] The interactive terminal is used for human-computer interaction display and provides feedback to the intelligent service module and the manual service module.

[0049] In this embodiment, it should be specifically explained that the data acquisition and fusion unit is used to obtain consultation information input by the user end, and introduce a cross-modal attention mechanism to fuse consultation information of different modalities to obtain the fused data features; the text information and image information are consultation information of different modalities, the image information and voice information are also consultation information of different modalities, and the text information and voice information are also consultation information of different modalities;

[0050] The user portrait generation unit obtains a user portrait based on the data features fused by the data acquisition and fusion unit in combination with external data; the external data includes but is not limited to user basic attribute data and interactive behavior data; the user basic attribute data includes but is not limited to the user's name, age, occupation, and industry; the interactive behavior data includes but is not limited to historical consultation content, consultation frequency, and user operation logs, and the user operation logs include but are not limited to the type of legal provisions clicked by the user, the length of time the user stays on a certain type of legal content, and the legal content collected or shared by the user;

[0051] The dynamic knowledge graph construction unit is used to construct a legal knowledge graph and dynamically adjust the legal knowledge graph using an incremental update mechanism to ensure the timeliness and consistency of the knowledge graph and achieve dynamic association and real-time updating of legal knowledge;

[0052] The content generation unit generates service content based on the user portrait of the user portrait generation unit and the legal knowledge graph of the dynamic knowledge graph construction unit; the service content is the legal content related to the consulting information, including but not limited to legal procedures, legal provisions and similar cases.

[0053] In this embodiment, it should be specifically explained that the user demand acquisition unit uses the data features fused by the data acquisition and fusion unit as the user demand features;

[0054] The demand matching unit performs demand matching based on user demand characteristics, obtains a successfully matched lawyer, and connects the successfully matched lawyer with the user for a manual dialogue;

[0055] In the process of the server providing legal consulting services to users, after the user enters the consulting information, the intelligent service module first generates the service content. If the user is not satisfied with the current content and needs a manual dialogue, the manual service module will connect with the lawyer for a manual dialogue.

[0056] In this embodiment, it should be specifically explained that the cross-modal attention mechanism of the data acquisition and fusion unit achieves deep fusion of data features of text information, voice information, and image information by dynamically calculating the correlation weights between data features of different modalities. The cross-modal attention mechanism is expressed as follows:

[0057] Among them, Attention(Q,K,V) represents the cross-modal attention mechanism; Q represents the query vector, which is generated by linear transformation of the feature matrix of the target modality and represents the target modality information that needs to be paid attention to, such as the word vector sequence of the legal issues consulted by the user; K represents the key vector, which is generated by the feature matrix of the source modality and stores the original features of the source modality for matching with the query vector, such as the image area features in the image information or the Mel spectrum features in the speech information; V represents the value vector, which comes from the source modality features, carries the actual semantic information of the source modality, and is output after weighting by the attention weight, as shown in the figure The target modality is the fused data modality, and the source modality is the initial data modality. For example, if the voice information, image information, and text information are all fused in the form of text information, then the target modality is text, the original modality of the voice information is voice, and the source modality of the image information is image. Ssoftmax represents the softmax function. M represents the mask matrix, which is a binary matrix used to process the problem of time and space misalignment of the data modality and is used to mask invalid areas, such as silent frames in voice or background areas in images. Represents the scaling factor, which is a normalization factor that prevents the softmax gradient from disappearing due to the dot product result being too large. It is used to control the variance of the dot product result within a reasonable range and improve training stability.

[0058] In this embodiment, it should be specifically noted that the user profile is obtained based on a deep learning model, which includes an embedding layer, an attention layer, and a fully connected layer;

[0059] The embedding layer maps discrete features in external data into low-dimensional dense vectors, capturing semantic associations between features; the discrete features include but are not limited to user ID, user occupation, legal document ID, etc.

[0060] The attention layer is based on a cross-modal attention mechanism to obtain the relevance weight of the user's historical behavior and the current consultation question; it dynamically focuses on key information to solve the "information redundancy" problem;

[0061] The fully connected layer concatenates the output features of the embedding layer and the attention layer, and integrates them into a unified user profile vector through a nonlinear transformation such as a ReLU activation function for output. The user profile vector includes but is not limited to the distribution of user interests and user needs.

[0062] The weight of the correlation between the user's historical behavior and the current consultation question is specifically:

[0063] Obtaining a text feature vector of the current consultation question and a text feature vector of the historical consultation question, using the text feature vector of the current consultation question as a query vector, and using the text feature vector of the historical consultation question as a key vector or a value vector, where the key vector and the value vector are usually different representations of the same set of features;

[0064] Get the relevance score: Among them, Score(Q now ,K i ) represents Q now With K i The correlation score, Q now represents the current query vector, that is, the text feature vector of the current consultation question, K i represents the i-th key vector, i.e., the text feature vector of the i-th historical consultation question; r represents the vector dimension, which is used to prevent the gradient from disappearing;

[0065] After obtaining the normalized weights, weighted fusion is performed to obtain the correlation weights; the normalized weights are obtained as follows: α i =Softmax(Scpre(Q now ,K i )); where α i represents the probability distribution of the relevance score corresponding to the i-th key vector, and Softmax represents the Softmax function. The Softmax function is used to convert the relevance score into a probability distribution to ensure that the sum of all weights is 1, which facilitates subsequent weighted fusion;

[0066] The weighted fusion is expressed as follows: Among them, XG represents the correlation weight between the user's historical behavior and the current consultation question; V i represents the i-th value vector; n represents the total number of historical consultation questions;

[0067] The fully connected layer contains a learnable weight matrix and bias term. The output user portrait vector has R dimensions, each dimension corresponds to a user feature, R∈[64,256];

[0068] Other parts of the model that are not described in detail are the same as those of the existing deep learning network model, and will not be described in detail in this embodiment.

[0069] In this embodiment, it should be specifically explained that the legal knowledge graph mainly includes entities, relationships between entities, and entity attributes. The entities mainly include legal subjects, legal objects, and legal acts. The legal subjects are people or organizations that participate in legal relationships and enjoy rights and obligations, including but not limited to plaintiffs, defendants, judges, and law firms. The legal objects are things or interests that are regulated by the rights and obligations of the subjects, including but not limited to contract texts, legal provisions, and evidence materials. The legal acts include but are not limited to prosecution, defense, mediation, and judgment. The entity attributes are used to describe entity information. For example, when the entity is a contract text in the legal object, the entity attributes can be the text date and text terms.

[0070] The inter-entity relationship is used to describe the legal logical association between entities and embody the chain of legal reasoning. The inter-entity relationship includes but is not limited to legal reference relationship, causal relationship, conflict relationship, and temporal relationship. The legal reference relationship refers to the cited legal provisions and clauses, the causal relationship refers to the cause leading to the result, the conflict relationship refers to the existence of a conflict between the two, and the temporal relationship refers to the relationship between the time sequence.

[0071] The incremental update mechanism uses web crawler technology to capture new legal cases in real time, uses graph neural networks to automatically expand entity nodes and corresponding inter-entity relationships, and uses rule engines such as Drools to verify conflicts between new and old legal provisions and mark nodes that require manual review. This achieves dynamic association and real-time updating of legal knowledge to support multi-dimensional reasoning in complex legal consulting scenarios.

[0072] The incremental update mechanism is based on the incremental update formula to dynamically adjust the legal knowledge graph;

[0073] The incremental update formula is expressed as: ZL t+1 =ZL t ∪ΔZL; where ZL t+1 Represents the knowledge graph at time t+1, ZL t represents the knowledge graph at time t, ΔZL represents the incremental update content; the incremental update content is divided into two categories, namely entity update and entity relationship update, which are expressed as follows:

[0074] Among them, (E new ,A,B) represents entity update, E new Indicates a newly added entity. A indicates the attributes of the newly added entity. B indicates the specific content of the newly added entity when the newly added entity is a legal object. For example, when the newly added entity is a legal provision, the specific content of the newly added entity is the original text of the legal provision. (D new ,T,C) represents entity relationship update, D newRepresents the newly added entity relationship, T represents the relationship type, such as reference relationship, causal relationship, conflict relationship, etc., C represents the confidence, and C satisfies C∈[0,1].

[0075] In this embodiment, it should be specifically explained that the demand matching formula is expressed as:

[0076] M ab =μ1·Sim(F a ,H b )+μ2·Avail(W b )+μ3·P b Among them, M ab represents the final matching score between the a-th user and the b-th lawyer; Sim(F a ,M b ) represents the professional similarity, that is, the demand feature F of the ath user a The lawyer's expertise characteristic H of the bth lawyer b The matching degree satisfies Sim(F a ,H b )∈[-1,1]; μ1 represents the weight of professional similarity; Avail(W b ) represents the time slack of the b-th lawyer, that is, the serviceability of the b-th lawyer within the time window W, satisfying Avail(W b )∈[0,1]; μ2 represents Avail(W b ) weight; P b Indicates the service quality score of the b-th lawyer, satisfying P b ∈[0,1]; μ3 represents P b weight; μ1 satisfies μ1∈(0,1), μ2 satisfies μ2∈(0,1), and μ3 satisfies μ3∈(0,1); in this embodiment, the value of μ1 can be obtained by dividing the number of successful matches by the total number of consultations, and μ2=0.2 and μ3=0.2 are selected; the values ​​of μ1, μ2, and μ3 can be set by those skilled in the art while satisfying the value range. If the user's needs are urgent, the value of μ2 can be appropriately increased. When the user's consultation content involves complex legal fields, the value of μ1 can be appropriately increased.

[0077] In this embodiment, it should be specifically explained that the Avail(W b ) is obtained as follows:

[0078] Where KX_b represents the spatial duration of the b-th lawyer, All_b represents the total service duration of the b-th lawyer, and ε represents the urgency compensation factor, which can be ε = 0 or ε = 1. If words similar to urging appear during the user consultation process, ε is ε = 1, otherwise ε is ε = 0.

[0079] The P b The way to obtain is:

[0080] Among them, p b _m represents the score of the b-th lawyer after the m-th service, and m′ represents the total number of services provided by the lawyer;

[0081] The Sim(F a ,H b ) is obtained as follows:

[0082] Among them, ||F a || indicates F a The vector modulus, ||H b || indicates H b The vector modulus of the b-th lawyer's lawyer expertise feature H b Features can be extracted based on information such as the lawyer's personal profile, published professional articles and papers, seminar topics participated in, types of cases successfully handled, and case results.

[0083] In this embodiment, it should be specifically explained that after the demand matching unit matches the user's demand characteristics, the lawyer corresponding to the highest final matching score is selected as the successfully matched lawyer, the successfully matched lawyer is connected with the user to conduct a conversation, and the lawyer's service is scored after the conversation ends;

[0084] The feedback to the intelligent service module and the manual service module specifically includes: the user evaluating the service content generated by the intelligent service module and the user rating the service of the lawyers in the manual service module.

[0085] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

[0086] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A legal consulting service management system, characterized by: Including user side, server side and interactive side; The user terminal is used to support user login and consultation information input; The server provides services to users based on the consultation information input by the user, including intelligent service module, manual service module and feedback optimization module; The intelligent service module includes a data acquisition and fusion unit, a user portrait generation unit, a dynamic knowledge graph construction unit, and a content generation unit; The manual service module includes a user demand acquisition unit and a demand matching unit; The feedback optimization module optimizes the feedback content of the intelligent service module and the manual service module based on the interactive terminal; The interactive terminal is used for human-computer interaction display and provides feedback to the intelligent service module and the manual service module.

2. A legal consulting service management system according to claim 1, characterized in that: The consulting information is the consulting content of the user's legal consulting service, including text information, image information and voice information. The consulting information is input in the form of text input, voice input and image input. The consulting information input in the form of text input is text information, the consulting information input in the form of voice input is voice information, and the consulting information input in the form of image is image information.

3. A legal consulting service management system according to claim 1, characterized in that: The data acquisition and fusion unit is used to obtain the consulting information input by the user end, and introduce a cross-modal attention mechanism to fuse the consulting information of different modalities to obtain the fused data features; The user portrait generation unit obtains a user portrait based on the data features fused by the data acquisition and fusion unit and in combination with external data; The dynamic knowledge graph construction unit is used to construct a legal knowledge graph and dynamically adjust the legal knowledge graph using an incremental update mechanism; The content generation unit generates service content based on the user portrait of the user portrait generation unit and the legal knowledge graph of the dynamic knowledge graph construction unit.

4. A legal consulting service management system according to claim 3, characterized in that: The external data includes user basic attribute data and interactive behavior data; the user basic attribute data includes user name, age, occupation and industry; the interactive behavior data includes historical consultation content, consultation frequency and user operation logs, and the user operation logs include the type of legal provisions clicked by the user, the length of time the user stays on a certain type of legal content, and the legal content collected and shared by the user.

5. The legal consulting service management system according to claim 1, characterized in that: The user demand acquisition unit uses the data features fused by the data acquisition and fusion unit as user demand features; The demand matching unit performs demand matching based on user demand characteristics, obtains successfully matched lawyers, and connects the successfully matched lawyers with the user for a manual dialogue.

6. A legal consulting service management system according to claim 3, characterized in that: The cross-modal attention mechanism is expressed as follows: Among them, Attention(Q,K,V) represents the cross-modal attention mechanism; Q represents the query vector, which is generated by linear transformation of the feature matrix of the target modality and represents the target modality information that currently needs to be paid attention to; K represents the key vector, which is generated by the feature matrix of the source modality and stores the original features of the source modality for matching with the query vector; V represents the value vector, which comes from the source modality features, carries the actual semantic information of the source modality, and is output after weighting by the attention weight; the target modality is the fused data modality, and the source modality is the initial data modality; softmax represents the softmax function; M represents the mask matrix; Represents the scaling factor.

7. A legal consulting service management system according to claim 6, characterized in that: The user profile is obtained based on a deep learning model, which includes an embedding layer, an attention layer, and a fully connected layer; The embedding layer maps discrete features in external data into low-dimensional dense vectors; The attention layer obtains the relevance weight of the user's historical behavior and the current consultation question based on the cross-modal attention mechanism; The fully connected layer concatenates the output features of the embedding layer and the attention layer, and integrates them into a unified user portrait vector through nonlinear transformation for output.

8. A legal consulting service management system according to claim 7, characterized in that: The legal knowledge graph mainly includes entities, relationships between entities, and entity attributes. The entities mainly include legal subjects, legal objects, and legal acts. The legal subjects are people and organizations that participate in legal relationships and enjoy rights and obligations. The legal objects are things and interests that are regulated by the rights and obligations of the subjects. The legal acts include prosecution, defense, mediation, and judgment. The entity attributes are used to describe entity information. The inter-entity relationship is used to describe the legal logical association between entities. The inter-entity relationship includes legal reference relationship, causal relationship, conflict relationship and temporal relationship. The legal reference relationship refers to the cited legal provisions and clauses, the causal relationship refers to the cause leading to the result, the conflict relationship refers to the existence of a conflict between the two, and the temporal relationship refers to the relationship between the time sequence. The incremental update mechanism is based on the incremental update formula and dynamically adjusts the legal knowledge graph.

9. A legal consulting service management system according to claim 8, characterized in that: The incremental update formula is expressed as: ZL t+1 =ZL t ∪ΔZL; where ZL t+1 Represents the knowledge graph at time t+1, ZL t represents the knowledge graph at time t, ΔZL represents the incremental update content; the incremental update content is divided into two categories, namely entity update and entity relationship update, which are expressed as follows: Among them, (E new ,A,B) represents entity update, E new Indicates a newly added entity, A indicates the attributes of the newly added entity, and B indicates the specific content of the newly added entity when the newly added entity is a legal object; (D new ,T,C) represents entity relationship update, D new Indicates a newly added entity relationship, and T indicates the relationship type.

10. A legal consulting service management system according to claim 5, characterized in that: The demand matching formula is expressed as: M ab =μ1·Sim(F a ,H b )+μ2·Avail(W b )+μ3·P b ; Among them, M ab represents the final matching score between the a-th user and the b-th lawyer; Sim(F a ,H b ) represents the professional similarity, that is, the demand feature F of the ath user a The lawyer's expertise characteristic H of the bth lawyer b The matching degree satisfies Sim(F a ,H b )∈[-1,1]; μ1 represents the weight of professional similarity; Avail(W b ) represents the time slack of the b-th lawyer, that is, the serviceability of the b-th lawyer within the time window W, satisfying Avail(W b )∈[0,1]; μ2 represents Avail(W b ) weight; P b Indicates the service quality score of the b-th lawyer, satisfying P b ∈[0,1]; μ3 represents P b weights; μ1 satisfies μ1∈(0,1), μ2 satisfies μ2∈(0,1), and μ3 satisfies μ3∈(0,1).

Images