Legal consultation self-service system and method based on human-computer interaction

Abstract

The invention relates to a legal consultation self-service system and method based on human-computer interaction, and relates to the field of consultation interaction systems. The system comprises a multi-mode front-end interaction module, an intention recognition module, a dialogue state management module, a dynamic decision tree engine, a risk assessment and classification module, a result presentation and document generation module and an artificial lawyer upgrade interface. The method comprises the steps of receiving multi-modal input, identifying intentions, interactively collecting facts by a dynamic decision tree, evaluating and grading risks, generating reports or suggestions, and triggering manual intervention when necessary. According to the method, technologies such as multi-modal interaction, a dynamic decision tree, quantitative risk assessment and seamless manual docking are introduced, so that the problems of static interaction, single mode, lack of risk assessment and the like of an existing legal consultation system are solved, more efficient, more accurate, more intelligent, more humanized legal consultation services with quantitative risks are provided, and the user experience and satisfaction are improved.

Description

Technical Field

[0001] The present invention relates to the field of consultation interactive systems, and in particular to a legal consultation self-service system and method based on human-computer interaction. Background Art

[0002] With social development and rising legal awareness, the public's demand for legal advice is growing. Traditional legal consultation methods often rely on manual services, which are subject to problems such as low efficiency, high costs, and limited service hours. To meet the demand for large-scale and convenient legal consultation, online legal consultation platforms have emerged. However, existing online legal consultation systems still have some technical shortcomings.

[0003] Existing legal consultation systems mainly rely on preset question-and-answer pairs or static templates for interaction. The user consultation process is fixed and cannot dynamically adjust the consultation path based on the user's actual situation and feedback. This often causes users to repeatedly enter information and go through lengthy processes, reducing user experience and consultation efficiency.

[0004] In addition, many existing systems only support a single text input modality and have difficulty processing voice consultations or evidence documents in image form, limiting their applicability in complex consultation scenarios and specific user groups.

[0005] Existing legal consultation systems often lack quantitative assessments and graded prompts of the inherent legal risks in user claims, resulting in users' lack of clear understanding of the legal risks of their own cases, which may lead to unrealistic expectations of the results of rights protection. Summary of the Invention

[0006] Purpose of the invention: To provide a legal consultation self-service system based on human-computer interaction, and further provide an autonomous consultation method based on the above-mentioned legal consultation self-service system based on human-computer interaction, so as to solve the above-mentioned problems existing in the prior art.

[0007] Technical solution: A self-service legal consultation system based on human-computer interaction, including seven components: a multimodal front-end interaction module, an intent recognition module, a dialogue state management module, a dynamic decision tree engine, a risk assessment and grading module, a result presentation and document generation module, and an artificial lawyer upgrade interface.

[0008] The multimodal front-end interaction module is used to receive text, voice and image information input by the user, convert the information of different modes into a format that can be processed by the system, and provide feedback in multimedia form; The intention recognition module is connected to the multimodal front-end interaction module and is used to analyze the text information input by the user and identify the user's legal consultation intention; The dialogue state management module is connected to the intention recognition module and is used to track and maintain the dialogue history and current status between the user and the system, record the information provided by the user, and guide the next step of the interaction process based on the recognized user intention and the set strategy; The dynamic decision tree engine is connected to the dialogue state management module and is used to store knowledge structures related to the legal consulting field, including preset legal questions, possible answer options, and jump rules based on user selections; The risk assessment and grading module is connected to the dynamic decision tree engine and is used to calculate the legal risk score of the case based on the key factual information provided by the user in the conversation, combined with the preset legal factor library and its weights, and divide the risk into different levels; The result presentation and document generation module is connected to the risk assessment and grading module and the dialogue state management module, and is used to generate a legal consulting report or suggestion based on the dialogue process and risk assessment results; The artificial lawyer upgrade interface is connected to the dialogue state management module and the risk assessment and grading module, and is used to transmit the desensitized consultation process data and assessment results to the artificial lawyer when preset conditions are met.

[0009] In a further embodiment, the intent and sentiment recognition module combines a Transformer model and sentiment analysis model optimized for the legal domain, and utilizes a rule-based corrector to improve recognition accuracy. The engine dynamically selects and presents the most relevant follow-up questions based on user-provided information, guiding users to provide key factual information. The results presentation and document generation module can generate editable legal document templates based on demand.

[0010] In a further embodiment, the multimodal front-end interaction module includes a voice processing unit, an image parsing unit, and a multimedia feedback unit. The branch rule library of the dynamic decision tree engine is divided by legal field and has the ability to receive external data through an API interface for online incremental updates.

[0011] In a further embodiment, the risk factor weight parameters of the risk assessment and grading module are jointly determined by legal experts and machine learning models.

[0012] In a further embodiment, the artificial lawyer upgrade interface includes a desensitizing mechanism for processing sensitive information before transmitting data.

[0013] A self-service consultation method for a legal consultation self-service system based on human-computer interaction, comprising the following steps: S1. Receive multimodal legal consultation information input by the user through text, voice or image; S2. Preprocessing the multimodal legal consultation information, including speech transcription and image key information recognition, and converting the processed information into a natural language text format; S3. performing intent recognition and analysis on the natural language text; S4. Based on the identified intent and emotion, combined with the current conversation state, the dynamic decision tree engine generates questions for the next interaction round. S5. Conduct multiple rounds of dynamic question-and-answer interactions with the user, and delve deeper along the branches of the dynamic decision tree based on user feedback to collect case-related factual information; S6. Based on the collected factual information, the risk assessment and grading module calculates the legal risk score of the case and grades it; S7. Generate a legal consulting report or recommendation based on the dialogue process, the risk assessment results, and the risk level, and generate relevant legal documents as needed; S8. If the risk level reaches the preset threshold or the user makes a request, the desensitized consultation data will be transmitted to the artificial lawyer through the artificial lawyer upgrade interface for subsequent processing.

[0014] In a further embodiment, in step S6, the risk assessment is calculated using a weighted scoring formula, which is based on preset legal element factors and their weights.

[0015] In a further embodiment, the weighted scoring formula is: ; in, is the weight of the i-th risk factor, is the quantitative value of the i-th risk factor provided by the user, and n is the number of risk factors involved in the assessment.

[0016] In a further embodiment, in step S8, when manual intervention is performed, the desensitization process includes using regular expressions to match and replace user identity sensitive information, and encrypting and transmitting other sensitive fields.

[0017] In a further embodiment, in step S5, the dynamic decision tree engine supports online updating of branching rules.

[0018] Beneficial effects: The present invention relates to a legal consultation self-service system and method based on human-computer interaction, which has the following beneficial effects: Dynamic path optimization: Through the dynamic decision tree engine, the system can adjust the consultation path based on the user's real-time feedback, avoiding the repetition and redundancy of static questions and answers. Tests show that consultation efficiency can be improved by about 40%.

[0019] Risk quantification and transparency: The introduction of a risk assessment and grading module enables users to intuitively and quantitatively understand the legal risk level of their own cases and the likelihood of successful rights protection, reducing user expectation deviations. User satisfaction surveys can reach approximately 89%.

[0020] Multimodal interaction: Supports multiple input methods such as text, voice, and image, greatly expanding the system's applicable scenarios and improving user convenience, especially for users who are not good at text input. The usage rate of elderly users can be increased by about 35%.

[0021] Seamless human interaction: In high-risk or complex cases, the system can quickly identify and automatically transmit desensitized key information to human lawyers, achieving seamless and efficient human service intervention. The average processing time for high-risk cases can be reduced by approximately 60% compared to traditional systems.

[0022] Dynamic update of knowledge base: The decision tree rule base and risk factor base support online updates to ensure the timeliness and accuracy of the system's legal knowledge and the ability to quickly adapt to changes in laws and regulations.

[0023] Intent recognition analysis: Able to identify user intentions and emotions and trigger soothing words to provide a more humane consulting experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the system architecture of the present invention.

[0025] Figure 2 This is a schematic diagram of the dynamic decision tree of the present invention, and takes the "traffic accident" consultation scenario as an example to illustrate the branch logic and node jump conditions.

[0026] Figure 3 This is a risk assessment flow chart of the present invention.

[0027] Figure 4 This is a flow chart of the legal consulting method of the present invention.

[0028] Figure 5 This is a schematic diagram of the data transmission flow of the artificial lawyer upgrade interface of the present invention. DETAILED DESCRIPTION

[0029] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art are not described to avoid confusion with the present invention.

[0030] The present invention relates to a self-service legal consultation system based on human-computer interaction, which mainly includes: Multimodal front-end interaction module: This module receives text, voice, and image information input by users, converts the information in different modalities into a format that the system can process, and provides multimedia feedback. Specifically, it includes a voice processing unit, an image analysis unit, and a multimedia feedback unit.

[0031] Intent and Sentiment Recognition Module: This module analyzes natural language input (from text input or speech transcription) to identify the user's legal consultation intent and current emotional state. This module combines a natural language processing model optimized for the legal domain with a sentiment analysis model, and uses a rule-based corrector to improve recognition accuracy.

[0032] Dialogue state management module: used to track and maintain the conversation history and current status between the user and the system, record the various information provided by the user, and guide the next step of the interaction process based on the identified user intentions and emotions, as well as the set strategies.

[0033] Dynamic Decision Tree Engine: This engine stores knowledge structures related to legal consulting, including pre-set legal questions, possible answers, and user-selected branching rules. Based on the information provided by the user during the conversation, the engine dynamically selects and presents the most relevant follow-up questions, guiding the user to provide key facts. It also supports online updating of branching rules and multi-conditional parallel branching.

[0034] Risk assessment and grading module: Based on the key factual information provided by the user in the conversation, combined with the preset legal factor library and its weights, the legal risk score of the case is calculated and the risks are divided into different levels.

[0035] Result presentation and document generation module: Generates legal consulting reports or recommendations based on the dialogue process and risk assessment results, and can generate editable legal document templates (such as evidence lists, lawyer's letter drafts, etc.) as needed, while providing a dynamic visual display of risk levels.

[0036] Artificial lawyer upgrade interface: When the system determines that the case risk is too high, the complexity exceeds the system's processing capabilities, or the user explicitly requires it, the desensitized consultation process data and evaluation results can be seamlessly transmitted to the artificial lawyer in the background, enabling rapid intervention of artificial services.

[0037] Furthermore, the branch rule base of the dynamic decision tree engine is divided according to legal fields and has the ability to receive external data through an API interface for online incremental updates.

[0038] Furthermore, the risk factor weight parameters of the risk assessment and grading module are jointly determined by legal experts and machine learning models to improve the professionalism and accuracy of the assessment.

[0039] Based on the above-mentioned legal consultation self-service system based on human-computer interaction, the present invention proposes a self-service consultation method for legal consultation self-service system based on human-computer interaction, and the specific steps are as follows: Step 1: receiving multimodal legal consultation information input by the user through text, voice or image; Step 2: pre-processing the multimodal legal consultation information, including speech transcription and image key information recognition, and converting the processed information into a natural language text format; Step 3: performing intent recognition and analysis on the natural language text; Step 4: Based on the identified intent and emotion, combined with the current conversation state, the dynamic decision tree engine generates questions for the next interaction round; Step 5: Conduct multiple rounds of dynamic question-and-answer interactions with the user, and delve deeper into the branches of the dynamic decision tree based on user feedback to collect case-related factual information; Step 6: Based on the collected factual information, the risk assessment and grading module calculates the legal risk score of the case and grades it; Step 7: Generate a legal consultation report or recommendation based on the dialogue process, the risk assessment results, and the risk level, and generate relevant legal documents as needed; Step eight: If the risk level reaches the preset threshold or the user makes a request, the desensitized consultation data will be transmitted to the artificial lawyer through the artificial lawyer upgrade interface for subsequent processing.

[0040] Furthermore, in step six, the risk assessment is calculated using a weighted scoring formula, which is based on preset legal element factors and their weights.

[0041] Furthermore, in step eight, when manual intervention is performed, the desensitization process includes using regular expressions to match and replace user identity sensitive information (such as ID number, phone number, etc.), and encrypting and transmitting other sensitive fields.

[0042] In a further preferred embodiment, the present invention provides a self-service legal consultation system based on human-computer interaction, which primarily includes a multimodal front-end interaction module, an intent and emotion recognition module, a dialogue state management module, a dynamic decision tree engine, a risk assessment and grading module, a result presentation and document generation module, and a human lawyer upgrade interface. These modules are connected via an internal bus or network to enable information exchange and functional collaboration.

[0043] The multimodal front-end interaction module is responsible for direct user interaction, receiving text input, processing voice input, and image input. The speech processing unit utilizes advanced end-to-end speech recognition models (such as DeepSpeech or its improved models) to support recognition in multiple dialects and transcribe user speech into text in real time with a low error rate (e.g., less than 5% word error rate in specific application scenarios). The image parsing unit utilizes object detection and recognition technologies (such as YOLOv5 or Transformer-based vision models) combined with optical character recognition (OCR) to perform structured parsing of user-uploaded evidence documents (such as contract scans, bill photos, and chat log screenshots), identifying and extracting key fields (e.g., party names, signing dates, amounts, and disputed clauses in contracts; amount, date, and payee in bills). The multimedia feedback unit 130 is responsible for presenting the system's responses in a user-friendly multimedia format. This can be achieved through text, TTS (Text-to-Speech) synthesis of legal advice, or by visually displaying pre-stored flowcharts (e.g., a labor arbitration application flowchart or a statute of limitations calculation flowchart).

[0044] The intent and emotion recognition module 200 receives processed text information from the multimodal front-end interaction module 100. Based on a Transformer model (such as the BERT-legal model) pre-trained on a large amount of legal text and fine-tuned for a specific legal consulting domain, the intent recognition unit 210 can accurately identify the user's legal consultation intent (e.g., "employment contract termination," "property division dispute," "intellectual property infringement," etc.), achieving an accuracy rate exceeding 92% in specific domains. The sentiment analysis unit 220 analyzes the text using a psychological model (such as the Valence-Arousal-Dominance (VAD) model) to identify the user's current emotional state (e.g., anxiety, anger, frustration, etc.), enabling the dialogue state management module 300 to adopt different interaction strategies. The rule corrector 230 includes a set of post-processing rules based on legal domain dictionaries, grammatical rules, and contextual information to correct potential misjudgments made by the intent and emotion recognition model. For example, regular expressions and domain dictionaries are used to accurately match and correct easily confusing professional terms such as "contract dispute."

[0045] The conversation state management module runs throughout the entire consultation process, using a dialogue management framework (such as Rasa) to maintain the conversation state and record all key information provided by the user, including user attributes (such as location, occupation, and identity), case-related facts (such as the time and location of the incident, the people involved, and the evidence), and completed question and answer rounds. The policy engine dynamically adjusts the system's interaction strategy based on the output of the intent and emotion recognition module and the current conversation state. For example, if it detects a high level of user anxiety, the policy engine will prioritize soothing language and may even trigger an escalation interface with a human lawyer in advance.

[0046] The dynamic decision tree engine is a core component of this invention. It stores a highly structured legal consultation knowledge base organized in the form of a decision tree. The branching rule base constructs a tree-like structure based on different legal fields (such as labor law, marriage and family law, intellectual property law, and tort liability law). Each node represents a legal question or information that requires the user to provide, and the lines connecting the nodes represent jump conditions based on the user's answer or information provided. For example, in a "traffic accident" consultation scenario, the decision tree might first ask, "Are there any casualties?" and jump to different branches based on the answer (yes / no). If the answer is "yes," further questions might be asked, such as "Extent of casualties?", "Will you call the police?", and "Amount of medical expenses?" Parallel branching strategies can be employed, simultaneously asking "Will you call the police?" and "Amount of medical expenses" to gather information more quickly. The dynamic decision tree engine features an online update mechanism. It receives information such as new regulations, judicial interpretations, and typical case studies from the legal database or manually entered through an API interface, automatically triggering incremental updates to the rule base. Hot reloading is supported, ensuring the timeliness and accuracy of consultation content.

[0047] The risk assessment and grading module receives key factual information collected by the dynamic decision tree engine. The risk factor library stores quantitative factors related to case risks in different legal fields (for example, the amount of contract breach, the completeness of the chain of evidence, whether the statute of limitations has expired, the solvency of the other party, etc.), and contains more than legal element factors. The weight allocator assigns a weight value to each risk factor. These weights are set by senior legal experts based on experience, and can be trained and optimized on historical case data through machine learning models to reflect the actual impact of each factor on the final result. The scoring calculation unit assigns a value to each relevant risk factor based on the factual information provided by the user (usually quantified as a value between 0-1, for example, the full score for evidence sufficiency is 1, no evidence is assigned a value of 0, and partial evidence is assigned a value of 0.5), and then applies the weighted scoring formula to calculate the overall legal risk score (Risk Score): ; in, is the weight of the i-th risk factor, is the quantitative value of the i-th risk factor provided by the user, and n is the number of risk factors involved in the assessment. Figure 3 As shown, the grading rule 540 divides the case risk into different levels according to the calculated total risk score: for example, high risk (RiskScore ≥80, usually requiring manual lawyer intervention), medium risk (60≤RiskScore<80, the system can provide detailed rights protection step suggestions and document templates), and low risk (RiskScore<60, the system can directly generate self-service solutions and simple legal documents).

[0048] The result presentation and document generation module generates a legal consulting report based on the summary of the entire consulting process and the risk assessment results. The report template engine calls the preset report template according to the risk level. The report content includes the case summary, risk assessment results, legal basis (automatically embedding relevant legal clauses, such as Article 38 of the Labor Contract Law), rights protection suggestions or self-service solutions. For medium and low-risk cases, the module can also generate editable legal document templates (such as collection letters, draft notices of termination of labor contracts, and lists of evidence), which users can edit or download online. The dynamic visualization unit can visualize the case risks in the form of heat maps, charts, etc., marking key risk points, so that users can understand the assessment results more intuitively.

[0049] The manual lawyer upgrade interface is a bridge between the system and manual services. This interface is triggered when specific conditions are met (such as the risk level is determined to be high risk, the user repeatedly expresses that they do not understand the system response, the user explicitly requests manual service, or the conversation state management module determines that manual intervention is required based on the user's emotions). Before transmitting data, the desensitization mechanism uses regular expressions to match and replace sensitive personal information entered by the user, such as ID number, phone number, home address, etc., and encrypts other sensitive but retained information (such as bank account number, contract amount). The context synchronization unit packages the complete conversation history, collected factual information, risk assessment results, and preliminary reports generated by the system into a structured file (such as JSON format), and pushes it to the back-end manual lawyer work platform through a secure channel. It may also send a text message notification to the user at the same time to inform him that the manual lawyer is about to intervene.

[0050] The method of the present invention is further described below with reference to a method flow chart and specific examples.

[0051] A self-service method for legal consultation based on human-computer interaction includes the following main steps: Step S1: User initiates a consultation. The user initiates a consultation through the system's webpage or app portal. The user can enter a question in the text box, speak by clicking the voice input button, or upload a file (such as a scanned contract or image). For example, a user uploads a scanned contract and voice-describes, "The company has been in arrears with wages for three months."

[0052] Step S2: Multimodal Data Processing. The multimodal front-end interaction module receives user input. The speech processing unit transcribes the speech into text: "The company has been in arrears with wages for three months." The image parsing unit performs OCR on the scanned contract, extracting key fields such as contract type, parties involved, signing date, and salary standards, and then structures this information. Text input is processed directly as text.

[0053] Step S3: Intent and emotion recognition. The intent and emotion recognition module receives the text message. The intent recognition unit analyzes the phrase "Company has delayed wages for three months" and determines the user's intent is "Labor dispute - delayed wages." The emotion analysis unit analyzes the user's tone of voice or the sentiment of the text and detects the emotion "anxiety."

[0054] Step S4: Conversation Status Update and Strategy Development. The conversation status management module updates the conversation status, records the user's intent and emotions, and develops the next interaction strategy based on the strategy engine. For example, in response to anxiety, the system may first send soothing words.

[0055] Step S5: Dynamic decision tree-driven interaction. Based on the user's intent, "Labor dispute - wage arrears," the dynamic decision tree engine selects the corresponding decision tree branch from the branch rule library. Based on the current state and collected information, the engine generates the next most critical question and presents it to the user via the multimedia feedback unit. For example, the system might ask, "Do you have a written labor contract with the company?" or "Please upload your pay slip." The user answers the question or uploads a file, and this information reenters the S2-S4 loop, driving the decision tree along a specific branch.

[0056] Step S6: Real-time Risk Assessment. During the conversation or after sufficient key information has been collected, the risk assessment and grading module begins its work. The scoring calculation unit uses the user-provided factual information, such as contract type, salary payment status, and evidence retention, to access the risk factor library and weight allocator to calculate a risk score. For example, if the user answers "verbal agreement" and "no salary slip," the quantitative value of the risk factor "sufficiency of written evidence" may be very low (e.g., 0.2), significantly affecting the overall risk score. The grading rules map the score to a risk level.

[0057] Step S7: Generate results and report. The result presentation and document generation module generates a legal advisory report based on the dialogue process and risk assessment results. The report content includes the risk level, risk point prompts based on risk factor analysis, legal basis (citing relevant legal provisions), and rights protection suggestions. For example, if the assessment result is medium risk, the report may include a detailed flowchart of the labor arbitration application steps. If the assessment result is low risk, a self-service solution may be directly provided, such as the steps and template for sending a collection letter. The dynamic visualization unit may generate a case risk heat map, marking "insufficient evidence" as the main risk point.

[0058] Step S8: Determine and execute human intervention. The conversation state management module or the risk assessment and grading module determines whether human lawyer intervention is necessary. Judgment criteria include the risk level reaching a high threshold, the user's explicit request, the system's inability to understand the user's question (multiple rounds without matching a decision tree branch), or the policy engine's determination based on user sentiment that human intervention is necessary. If these conditions are met, the human lawyer escalation interface is triggered. The desensitization mechanism desensitizes and encrypts the consultation data. The context synchronization unit sends the processed data (including conversation history, key information, and risk assessment reports) to the backend human lawyer and notifies the user.

[0059] Example 1: Labor Dispute Consultation Users upload a scanned copy of their employment contract through the app, and the system uses image analysis to extract key information, such as "3-month probation period" and "8,000 yuan monthly salary." The user also uses voice to describe, "The company said they failed the probation period and are firing me, but they didn't notify me in advance."

[0060] The intent and emotion recognition module identifies the intent as "labor dispute - illegal termination of labor contract", and the emotion analysis unit detects the emotion of "anger".

[0061] The conversation state management module records intent and sentiment. The dynamic decision tree engine enters the "illegal termination of labor contract" branch and first asks: "What is the form of the company's notice of termination of labor contract?" The user answers, "Verbal notification." This triggers the "Non-written notification" branch of the decision tree. The system then asks, "Did the company inform you of the specific reason? What was the reason?" The user replied: "They said they failed the probation period, but the contract doesn't specify the specific assessment criteria." As factual information is gathered, the risk assessment and grading module conducts an assessment based on risk factors such as "Termination Notice Form (Oral)", "Termination Reason (Probationary Period Failure)", and "Whether the Contract Contains Assessment Standards". Assume the assessment result is a RiskScore of 82, which is classified as high risk.

[0062] The system immediately triggers the manual lawyer upgrade interface, sending the desensitized conversation data, extracted contract information, and high-risk assessment results to the back-end lawyer. Simultaneously, the results presentation and document generation module generates a preliminary report, informing the user of the high-risk level and advising immediate professional legal assistance. The report also states that a manual lawyer will contact the user shortly.

[0063] Example 2: Online update of decision tree Suppose the country has issued a new "Electronic Evidence Regulations", which has made important revisions to the rules for accepting electronic evidence such as electronic chat records and blockchain evidence.

[0064] System administrators upload the new regulations to the dynamic decision tree engine's online update mechanism via an API. This mechanism automatically parses the new regulations and adds or modifies nodes and rules under the "Evidence Type" decision tree branch. For example, under the "Evidence Type" branch, nodes are added regarding issues like "Proof of Authenticity of Electronic Chat Records" and "Validity of Blockchain Evidence," and the corresponding risk assessment factors and weights for evidence effectiveness are updated.

[0065] When users consult regarding electronic evidence, the updated decision tree will guide them to provide information about the method of obtaining electronic evidence, storage media, integrity, etc. The risk assessment module will more accurately assess the risk of accepting electronic evidence based on the updated rules.

[0066] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A self-service legal consultation system based on human-computer interaction, characterized by include: The multimodal front-end interaction module receives text, voice, and image information input by the user, converts the information in different modalities into a format that the system can process, and provides multimedia feedback; An intention recognition module, connected to the multimodal front-end interaction module, for analyzing the text information input by the user and identifying the user's legal consultation intention; The dialogue state management module is connected to the intention recognition module and is used to track and maintain the dialogue history and current state between the user and the system, record the information provided by the user, and guide the next step of the interaction process based on the recognized user intention and the set strategy; A dynamic decision tree engine, connected to the dialogue state management module, is used to store knowledge structures related to the legal consulting field, including preset legal questions, possible answer options, and jump rules based on user selections; A risk assessment and grading module, connected to the dynamic decision tree engine, is used to calculate the legal risk score of the case based on the key factual information provided by the user in the conversation, combined with a preset legal factor library and its weights, and to classify the risk into different levels; A result presentation and document generation module, connected to the risk assessment and grading module and the dialogue state management module, for generating a legal consulting report or recommendation based on the dialogue process and risk assessment results; The artificial lawyer upgrade interface connects the dialogue state management module and the risk assessment and grading module, and is used to transmit the desensitized consultation process data and assessment results to the artificial lawyer when preset conditions are met.

2. The self-service legal consultation system based on human-computer interaction according to claim 1, characterized in that: The intent and emotion recognition module combines a Transformer model and a sentiment analysis model optimized for the legal field, and improves recognition accuracy through a rule corrector; The engine dynamically selects and presents the most relevant follow-up questions based on the information provided by the user, guiding the user to provide key factual information; The result presentation and document generation module can generate corresponding editable legal document templates according to needs.

3. The self-service legal consultation system based on human-computer interaction according to claim 1 is characterized by: The multimodal front-end interaction module includes a voice processing unit, an image analysis unit and a multimedia feedback unit; The branch rule base of the dynamic decision tree engine is divided according to legal fields and has the ability to receive external data through an API interface for online incremental updates.

4. The self-service legal consultation system based on human-computer interaction according to claim 1, characterized in that: The risk factor weight parameters of the risk assessment and grading module are jointly determined by legal experts and machine learning models.

5. The self-service legal consultation system based on human-computer interaction according to claim 3 is characterized by: The manual lawyer upgrade interface includes a desensitizing mechanism for processing sensitive information before transmitting data.

6. The self-service consultation method of the legal consultation self-service system based on human-computer interaction according to any one of claims 1 to 5, characterized in that The following steps are involved: S1. Receive multimodal legal consultation information input by the user through text, voice or image; S2. Preprocessing the multimodal legal consultation information, including speech transcription and image key information recognition, and converting the processed information into a natural language text format; S3. performing intent recognition and analysis on the natural language text; S4. Based on the identified intent and emotion, combined with the current conversation state, the dynamic decision tree engine generates questions for the next interaction round. S5. Conduct multiple rounds of dynamic question-and-answer interactions with the user, and delve deeper along the branches of the dynamic decision tree based on user feedback to collect case-related factual information; S6. Based on the collected factual information, the risk assessment and grading module calculates the legal risk score of the case and grades it; S7. Generate a legal consulting report or recommendation based on the dialogue process, the risk assessment results, and the risk level, and generate relevant legal documents as needed; S8. If the risk level reaches a preset threshold or the user makes a request, the desensitized consultation data will be transmitted to the artificial lawyer through the artificial lawyer upgrade interface for subsequent processing.

7. The self-service consultation method of the legal consultation self-service system based on human-computer interaction according to claim 6, characterized in that: In step S6, the risk assessment is calculated using a weighted scoring formula, which is based on preset legal element factors and their weights.

8. The self-service consultation method of the legal consultation self-service system based on human-computer interaction according to claim 7 is characterized in that: The weighted scoring formula is: ； in, is the weight of the i-th risk factor, is the quantitative value of the i-th risk factor provided by the user, and n is the number of risk factors involved in the assessment.

9. The self-service consultation method of the legal consultation self-service system based on human-computer interaction according to claim 7, characterized in that: In step S8, when manual intervention occurs, the desensitization process includes using regular expressions to match and replace user identity sensitive information, and encrypting and transmitting other sensitive fields.

10. The self-service consultation method of the legal consultation self-service system based on human-computer interaction according to claim 6, characterized in that: In step S5, the dynamic decision tree engine supports online updating of branch rules.

Images