Method and system for fusing emotional intensity and social influence to legal consultation risk scoring

Abstract

The application discloses a legal consultation risk scoring method and system fusing emotional intensity and social influence, and belongs to the technical field of artificial intelligence. The method comprises the following steps: constructing an emotional vocabulary library intensity table, and combining the emotional vocabulary library intensity table to calculate the emotional score of a consultation text; constructing a social influence intensity table, and combining the social influence intensity table to calculate the social influence score of the consultation text; and obtaining the risk score of the consultation text based on the emotional score and the social influence score. The application can early identify high-risk consultations, reasonably allocate processing resources, prevent potential legal crises and social events, meet the transparency requirements, and promote the intelligentization of legal services and the modernization of social governance.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a legal consultation risk scoring method and system that integrates emotional intensity and social influence. Background Technology

[0002] With the popularization of legal consultation services and the development of internet technology, more and more users are seeking legal assistance through online platforms. However, existing legal consultation systems have significant shortcomings in risk assessment, mainly in the following aspects:

[0003] First, traditional legal consultation systems lack the ability to quantitatively analyze the emotional dimension of consultation content. Current systems primarily rely on keyword matching and simple classification, failing to accurately identify the emotional intensity contained in the consultation text. For example, when users use words like "anger" or "despair," the system often only labels them as negative emotions, unable to quantify the degree of impact these emotions have on the case's risk level. This crude approach prevents the system from distinguishing the essential difference between "mild dissatisfaction" and "extreme anger," thus affecting the accuracy of risk assessment. Research shows a significant positive correlation between emotional intensity and case risk level; consultations with high emotional intensity often foreshadow higher legal risks and social impacts. However, existing systems lack standardized emotional vocabulary databases and intensity weighting tables, making it impossible to accurately assess this dimension.

[0004] Secondly, existing technologies have serious shortcomings in assessing social impact. The social impact of legal cases involves multiple dimensions, including public interest, livelihood concerns, and special groups. However, traditional systems can only perform simple case classification and cannot quantify the scope and extent of the potential social impact of a case. For example, cases involving "financial fraud by listed companies" are highly socially sensitive, but existing systems cannot automatically identify these keywords and assess their potential social impact. This lack of assessment capability prevents platforms from prioritizing cases that may cause significant social impact and from providing early warnings to relevant departments.

[0005] Third, existing risk assessment systems lack interpretability. Most systems employ "black box" machine learning models, which, while potentially achieving a certain level of accuracy, fail to clearly explain the basis and logic behind the scoring results to users and regulatory agencies. This lack of transparency not only reduces user trust but also makes it difficult for the systems to meet increasingly stringent algorithmic regulatory requirements. In the sensitive field of legal consulting, the decision-making process of scoring systems must be able to be reviewed and understood, and existing technologies are clearly inadequate in this regard.

[0006] Furthermore, existing systems often employ single-dimensional risk assessment methods, failing to comprehensively consider factors such as emotion, social impact, and legal entity status. This one-sidedness leads to risk assessment results that frequently deviate from reality. For example, a case may simultaneously possess high emotional intensity and high social impact, but existing systems may only consider one aspect, thus underestimating the overall risk.

[0007] Finally, existing technologies lack the ability to dynamically adapt. The legal environment and social concerns are constantly changing, and new sensitive terms and issues will continuously emerge, but the keyword databases and evaluation rules of traditional systems tend to remain fixed and cannot adapt to these changes. This rigidity leads to a decline in the effectiveness of system evaluations over time.

[0008] In summary, existing legal consultation risk assessment techniques suffer from several shortcomings, including inaccurate sentiment analysis, lack of social impact assessment, insufficient interpretability, lack of multi-dimensional comprehensive assessment capabilities, and poor dynamic adaptability. There is an urgent need for a new method and system that can comprehensively, accurately, and transparently assess legal consultation risks.

[0009] Publicly available literature has conducted research on the identification of risks in legal texts. Patent publication number CN116227487A proposes a "Smart Review System for Risk Points in Legal Texts," but this system relies on a pre-set vocabulary database to extract vocabulary content, thereby identifying the presence of important sentences and designating these important sentences as risk points to analyze the suspiciousness, risk, or errors in that part of the legal text. However, this invention performs risk analysis on the entire article. While there are similarities in the use of rule-based keyword databases, this invention identifies the semantics of a social impact database, and the interpretation of identified risks incorporates content from multiple databases for joint explanation. Summary of the Invention

[0010] This invention discloses a legal consultation risk assessment method and system that integrates emotional intensity and social impact. It can identify high-risk consultations at an early stage, allocate resources reasonably, prevent potential legal crises and social events, meet transparency requirements, and promote the intelligentization of legal services and the modernization of social governance.

[0011] To achieve the above objectives, the technical solution of the present invention includes the following:

[0012] A legal counseling risk assessment method that integrates emotional intensity and social impact, the method comprising:

[0013] Construct an emotional lexicon strength table, and calculate the emotional score W of the consultation text based on the emotional lexicon strength table; wherein, the emotional lexicon strength table includes: emotional intensity weight, negative emotion example words, and positive emotion example words;

[0014] Construct a social influence intensity table, and calculate the social influence score S of the consultation text based on the social influence intensity table; wherein, the social influence intensity table includes: social influence intensity weights and example words;

[0015] Based on the emotional score W and the social influence score S, the risk score R of the consultation text is obtained.

[0016] Furthermore, by combining the sentiment lexicon strength table, the sentiment score W of the consultation text is calculated, including:

[0017] The RoBERTa-Large model was used to analyze the sentiment polarity of the consultation text, and the final output was the sentiment polarity coefficient P, P∈[-1,1];

[0018] Identify the sentiment words a contained in the consultation text i And retrieve the sentiment word 'a' from the sentiment lexicon strength table. i After obtaining the corresponding example words for positive or neutral emotions, we get the emotion word 'a'. i The corresponding emotional intensity weight α i ;

[0019] Based on the text sentiment polarity coefficient P and intensity weight α i The sentiment score W of the consultation text is obtained; if |W|>1, it is truncated so that |W|∈[0,1].

[0020] Furthermore, the sentiment score of the consultation text. n represents the number of sentiment words contained in the consultation text.

[0021] Furthermore, combining the social influence intensity scale, the social influence score S of the consultation text is calculated, including:

[0022] The consultation text and the social influence strength table are passed to the large language model to obtain the social influence strength weight w of the consultation text and the social influence value v corresponding to the consultation text.

[0023] Based on the social influence strength weight w and the social influence value v, the social influence score S of the consultation text is obtained.

[0024] Furthermore, based on the emotional score W and the social impact score S, the risk score R of the consultation text is obtained, including:

[0025] The sentiment value f(W) of the consultation text is calculated based on the sentiment score W.

[0026] The social impact value g(S) of the consultation text is calculated based on the social impact score S.

[0027] Based on the emotional score W and the social influence score S, the dynamic compensation value h(W,S) of the consultation text is calculated.

[0028] The risk score R of the consultation text is calculated based on the emotional value f(W), the social influence value g(S), and the dynamic compensation value h(W,S).

[0029] Furthermore, the sentiment value Where k1 represents the steepness of the emotional transition, and θ1 represents the classification threshold of the emotional intensity;

[0030] The social influence value Where k2 represents the steepness of the transition of the control influence, θ2 represents the influence threshold, a represents the baseline value of low influence, b represents the adjustment range of high influence, and a+b=1;

[0031] The dynamic compensation value Where c represents the compensation weight, and θ3 represents the index parameter, which is used to control the influence of the social influence score S on the compensation value h.

[0032] The risk score R = 100 × Min(1, f(W) × g(S) + h(W, S)).

[0033] Furthermore, after obtaining the risk score of the consultation text based on the emotional score W and the social influence score S, the method further includes:

[0034] If the risk score of the consultation text is greater than a risk threshold, a token-level attention matrix A for the consultation text is constructed using the Transformer model. i ;

[0035] Based on the token-level attention matrix A i The attention weights of each element are used to obtain candidate keywords from the consultation text;

[0036] Based on the emotional lexicon strength table and the social influence strength table, obtain the negative emotional intensity weight and social influence intensity weight of the candidate keyword.

[0037] The multi-dimensional score of each candidate keyword is calculated by combining the attention weight, negative sentiment intensity weight, and social influence intensity weight.

[0038] Based on the multi-dimensional scoring of candidate keywords, the keywords of this consultation text are obtained.

[0039] Furthermore, after obtaining the risk score of the consultation text based on the emotional score W and the social influence score S, the method further includes:

[0040] Semantic analysis is performed on newly added consultation texts to identify out-of-vocabulary words in the texts. These out-of-vocabulary words are example words that do not appear in the social influence intensity table.

[0041] By calculating the semantic similarity between out-of-vocabulary words and example words, the social influence strength weight of the out-of-vocabulary word is determined, and the social influence strength table is updated.

[0042] A legal consultation risk scoring system that integrates emotional intensity and social impact, the system comprising:

[0043] The sentiment scoring module is used to construct a sentiment lexicon strength table and, in conjunction with this sentiment lexicon strength table, calculate the sentiment score W of the consultation text; wherein, the sentiment lexicon strength table includes: sentiment intensity weights, negative sentiment example words, and positive sentiment example words;

[0044] The influence scoring module is used to construct a social influence intensity table and, in conjunction with the social influence intensity table, calculate the social influence score S of the consultation text; wherein, the social influence intensity table includes: social influence intensity weights and example words;

[0045] The risk scoring module is used to obtain a risk score for the consultation text based on the emotional score W and the social influence score S.

[0046] An electronic device includes: a processor and a memory storing computer program instructions; the processor, when executing the computer program instructions, implements the legal consultation risk scoring method that integrates emotional intensity and social influence as described above.

[0047] Compared with the prior art, the present invention has at least the following beneficial effects.

[0048] 1. The accuracy of risk assessment has been significantly improved.

[0049] Compared with traditional methods, this invention significantly improves the accuracy of legal consultation risk assessment by quantifying emotional intensity and classifying social impact. Specifically, it can distinguish the emotional difference between "anger" (1.0) and "dissatisfaction" (0.5), and identify the difference in social impact between "vaccine safety" (1.0) and "property dispute" (0.2).

[0050] The improved accuracy stems from three technological innovations: 1) refined emotion quantification replaces simple polarity classification; 2) the introduction of the social impact dimension fills the assessment blind spots of traditional systems; and 3) multi-dimensional integration avoids one-sided judgments. For example, for a particular consultation, a traditional system might assess based solely on emotion or solely on case type, while this invention can consider both and their interaction simultaneously, providing a more accurate comprehensive assessment.

[0051] 2. Enhanced early warning capabilities for high-risk cases.

[0052] The tiered assessment system of this invention is particularly beneficial for the early identification and intervention of high-risk cases. The system can detect warning signals in consultations, such as highly emotionally charged vocabulary ("murder") combined with topics of high social impact, automatically raising the risk level and triggering an early warning mechanism. In practical applications, this capability enables the platform to identify consultations that may escalate into extreme events or public crises in advance, creating conditions for timely intervention.

[0053] The effectiveness of early warning is reflected in: 1) timely identification of high-weighted words; 2) superimposed detection of multiple high-risk indicators; and 3) configurable warning thresholds. For example, the system can be set to automatically issue a warning when both emotional and social impact are high, even if the legal entity dimension score is not high. This flexibility is crucial for preventing serious incidents.

[0054] 3. Improved system transparency and user trust.

[0055] The interpretability framework of this invention significantly improves the transparency and user acceptance of the system. Unlike traditional "black box" models, this invention clearly demonstrates the basis for risk scoring, including emotional intensity and social impact. This transparency brings multiple benefits: users are more likely to accept the assessment results; lawyers can understand the nature of cases more quickly; and platform administrators can effectively review system decisions.

[0056] Explainability is achieved through the following techniques: 1) rule-based extraction of rating elements; 2) natural language-generated explanation templates; and 3) interactive display of rating details. This transparency reduces rating disputes and significantly improves the system's usability and credibility.

[0057] 4. Optimize resource allocation efficiency.

[0058] The multi-dimensional scoring system of this invention enables legal consultation platforms to achieve precise matching and optimized allocation of resources. Based on risk level, the platform can implement differentiated processing strategies: high-risk consultations are prioritized for allocation to experienced expert lawyers with shortened response times; medium-risk consultations are processed according to standard procedures; and low-risk consultations can be partially automated or directed to self-service resources.

[0059] Efficiency optimization benefits from: 1) accurate risk classification; 2) configurable resource allocation rules; and 3) deep integration with workflow systems. The system not only assesses risks but also automatically triggers corresponding workflows based on the risk assessment results, achieving closed-loop management of assessment and handling.

[0060] 5. Adaptability and sustainable development capabilities.

[0061] The dynamic adaptation mechanism of this invention endows the system with long-term sustainable development capabilities. Unlike traditional static systems, this invention can adapt to changes in the following ways: 1) regularly updating the vocabulary and weights; 2) learning new patterns from new cases; and 3) adjusting sensitivity based on social hotspots. For example, when new livelihood issues (such as "violations of the double reduction policy") emerge, the administrator can quickly add them to the high social impact vocabulary to ensure the system maintains the accuracy of its assessments.

[0062] This adaptability is achieved through a two-pronged approach: technically, the system provides user-friendly automated learning tools; and managerially, a system for regular evaluation and updates is established.

[0063] 6. Social benefits and public value.

[0064] Beyond its technological advantages, this invention also offers significant social benefits: early identification of high-risk legal advice helps prevent extreme events and social crises; prioritizing cases involving public interests and special groups promotes social fairness and justice; and transparent risk assessment enhances public trust in legal AI systems. While these benefits are difficult to quantify, they are of great importance to the construction of a society governed by the rule of law.

[0065] In particular, the system's sensitive identification of cases involving vulnerable groups (such as "discrimination against people with disabilities") reflects the humanistic care and social responsibility of technology. Attached Figure Description

[0066] Figure 1 Time series diagram of multi-dimensional risk scoring system.

[0067] Figure 2 Risk source analysis flowchart. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0069] The legal consultation risk scoring method of this invention, such as Figure 1 The process includes the following steps.

[0070] Step S1: Text sentiment quantification.

[0071] The input is a consultation text T. The RoBERTa-Large model is used to analyze the sentiment polarity of the text and finally output the sentiment coefficient P, where P∈[-1,1]. The closer the P value is to 1, the more significant the positive sentiment, and the closer the P value is to -1, the more significant the negative sentiment.

[0072] The emotional terms and their weights used in this invention are shown in Table 1:

[0073] Table 1. Strength of the Emotional Lexicography

[0074]

[0075] Assuming the output sentiment polarity coefficient is P, the list of sentiment words identified in the consultation text is {a1, a2, ..., a...}. n For each emotion word a i Refer to the intensity weight α in Table 1 i Adjusting the polarity fraction: If |W|>1, then it needs to be truncated so that the weighted sentiment score W∈[-1,1].

[0076] For example: The text is "What pains me is that the other person has been deceiving me, which makes me feel irritable and confused" (High + Medium + Low + Low, α1 = 1.0, α2 = 0.5, α3 = 0.2, α4 = 0.2). If P = -0.6, then...

[0077] Step S2: Quantify social influence.

[0078] The social influence contained in the text is judged based on the GPT model. The input content is the consultation text T and the social influence intensity table as shown in Table 2 below.

[0079] Table 2. Social Influence Intensity Table

[0080]

[0081] The consultation text T is integrated with the social influence intensity table into a single prompt word:

[0082]

[0083] The GPT model is used to determine two things: 1) the strength of the text semantics in the social influence strength table, so as to determine the social influence strength weight w of this text T; 2) the magnitude of the social impact v that the consultation case will lead to, v∈[0,1], the closer the v value is to 1, the greater the social impact.

[0084] The formula for calculating S is: S = w·v, S∈[0,1].

[0085] Step S3: Risk value prediction.

[0086] The formula structure of the emotional-social influence gating fusion risk scoring model is: R = 100 × Min(1, f(W) × g(S) + h(W, S)).

[0087] Specifically as follows:

[0088]

[0089] (1) Emotion gating function:

[0090] The sentiment gating function f(W) maps the sentiment value W∈[-1,1] to the interval [0,1] using the sigmoid function, and its formula is as follows:

[0091]

[0092] The parameter k1 = 10 controls the steepness of the emotional transition. A higher k1 makes the function transition rapidly near the threshold θ1, ensuring that the emotional intensity is close to 1 (high emotion) when |W| ≥ 0.7, and close to 0 (low emotion) when |W| ≤ 0.3.

[0093] The parameter θ1 = 0.4 represents the classification threshold for sentiment intensity. By adjusting θ1, the starting point for high sentiment can be controlled. For example, when θ1 = 0.4, f(W) ≥ 0.62 when |W| ≥ 0.5, and f(W) ≥ 0.88 when |W| ≥ 0.7. This setting ensures that the f(W) value for medium sentiment (|W| ∈ [0.3, 0.7) is significantly higher than that for low sentiment, but slightly lower than that for high sentiment.

[0094] In social media analytics, high sentiment (such as anger or excitement) typically corresponds to extreme ratings (|W|≥0.7), while moderate sentiment (such as neutral comments) corresponds to moderate ratings. This function avoids the jump problem caused by hard thresholds through the smoothness of the sigmoid function.

[0095] (2) Influence Gating Function:

[0096] The influence gating function g(S) is a translated sigmoid function, and its formula is:

[0097]

[0098] The parameter a = 0.6 is the baseline value for low influence. When S ≈ 0 (extremely low influence), g(S) ≈ 0.6, to avoid completely ignoring the contribution of low-influence content.

[0099] The parameter b = 0.4 represents the adjustment range for high influence. When S ≈ 1 (extremely high influence), g(S) ≈ a + b = 1.

[0100] The parameter k2 = 12 controls the steepness of the influence transition. It ensures that g(S) ≈ 1 when S ≥ 0.8, and g(S) ≈ 0.6 when S ≤ 0.3.

[0101] The parameter θ2 = 0.5 is the influence threshold. When S ≥ 0.6, g(S) ≥ 0.8, which clearly distinguishes between medium and high influence.

[0102] In terms of social influence, a high level of influence significantly increases the risk value, while a low level of influence retains the basic weight. This function achieves this through parameters a and b.

[0103] (3) Dynamic compensation item:

[0104] The compensation term is used to prevent R from reaching zero during periods of low emotion; its formula is as follows:

[0105]

[0106] The parameter c = 0.1 is the compensation weight. A smaller c ensures that the compensation term only slightly increases R, avoiding excessive interference in high-emotional scenes.

[0107] 1-f(W): Emotional deficiency. When f(W)≈0 (low emotion), the compensation term is the largest; when f(W)≈1, the compensation term approaches 0.

[0108] parameter The non-linear adjustment of influence weakens the compensatory effect of low influence (S≈0) but retains the contribution of medium to high influence. Content with low sentiment but high influence still requires a certain risk value, which is determined by S. 1 / 3 Compensation for achieving this goal.

[0109] (4) Feasibility explanation:

[0110] The sentiment gating f(W) explicitly classifies high, medium, and low sentiment through the Sigmoid function, with parameters k1 and θ1 controlling the strictness of the classification.

[0111] Influence g(S): The minimum value for low influence and the saturation value for high influence are achieved by translating the Sigmoid function. The parameters a, b, and θ2 adjust the boundary point.

[0112] Compensation term h(W,S): Through low weight and non-linear design, it ensures that low-emotion content is not ignored, but does not interfere with high-emotion scenes.

[0113] The feasibility of the emotional-social influence gating risk scoring model is demonstrated by combining high, medium, and low levels of W and S parameters. The results are shown in Table 3.

[0114] Table 3 Results of the Legal Consulting Risk Assessment Model

[0115]

[0116] High risk is defined as R≥80, medium risk as 60≤R<80, low risk as R<60, and R∈[0,100].

[0117] Step S4 Risk Source Analysis

[0118] Risk source analysis is a key interpretative module of the multi-dimensional risk scoring system for legal consultation, automatically triggered when the system detects high-risk consultation text (R≥80). Its core functions include: (1) High-risk sentence location: identifying the sentence with the highest risk contribution in the text; (2) Keyword extraction: marking the core words that specifically lead to high risk; (3) Interpretable output: generating a human-understandable source analysis report explaining the basis of the risk. The flowchart of this risk source analysis is as follows: Figure 2 As shown.

[0119] (1) High-risk trigger judgment: When the risk score R ≥ 80, continue the subsequent source tracing steps;

[0120]

[0121] (2) Text preprocessing: The original consultation text T (string) is segmented into sentences by punctuation using regular expressions, short sentences (<5 characters) are filtered out, and the sentence list is output: TS = [ts1, ts2, ..., ts n ].

[0122] (3) Statement-level risk quantification: Transformer model + custom risk perception layer

[0123] Risk score calculation: RiskScore(ts) i ) = λ·emotional score(ts) i )+(1-λ)·Social Influence Score(ts) i ), where λ = 0.6, λ ∈ [0, 1]

[0124] Example:

[0125] <![CDATA[s1]]> 0.95 0.9 0.93 <![CDATA[s2]]> 0.30 0.2 0.26

[0126] (4) High-risk statement screening: The threshold rule is when RiskScore(ts) is... i If ≥0.7, then retain.

[0127] (5) Keyword extraction and scoring:

[0128] 1) Attention weight extraction: Use the Transformer model to obtain the token-level attention matrix A i ;

[0129] 2) Candidate word selection: Select A i Tokens with a weight ≥ 0.5;

[0130] 3) Multi-dimensional scoring:

[0131] By synergistically combining attention weight, social impact intensity, and negative emotion intensity, the interpretation scoring system can cover known legal risks, dynamically adapt to emerging expressions, and eliminate low-risk interference. The formula is as follows:

[0132] KeyScore(ω)=0.4·AW(ω)+0.4·SI(ω)+0.2·NE(ω)

[0133] Where AW(ω) is the attention weight, which is the output value of the attention mechanism of the Transformer model. The model generates an attention matrix for the input text and extracts the maximum attention value for each word (token): AW(ω) = max(AttentionMatrix[ω,:]), AW(ω)∈[0,1], and the larger the value, the more the model pays attention to the word.

[0134] SI(ω) represents the intensity of social impact, and the vocabulary and weights it is based on are shown in Table 2.

[0135] NE(ω) represents the intensity of negative emotion, i.e., the intensity of negative emotion carried by a word, used to measure emotional risk. The vocabulary and weights used to determine this are the negative emotion content listed in Table 1.

[0136] (6) Final explanatory output:

[0137] Returned and displayed in JSON format:

[0138]

[0139] S4.1 Dynamic Vocabulary Update Mechanism

[0140] This system aims to address the problem of traditional legal risk assessment systems having static social influence vocabularies that cannot adapt to changing social issues. Run weekly, the mechanism uses the Legal-BERT model to automatically capture new words, combined with manual review, to ensure the system can dynamically identify emerging legal risk terms (such as "violations of the double reduction policy" and "AI face-swapping fraud") and scientifically adjust their weights and classifications.

[0141] The Legal-BERT model was used to perform semantic analysis on newly added consultation texts each week. The model scans the text using a sliding window (512 tokens), identifies out-of-vocabulary words, and calculates their semantic similarity with high, medium, and low-weight words in the existing social influence vocabulary. For example, when the new word "metaverse virtual property dispute" is detected, the model associates it with the existing word "virtual property dispute" (similarity 0.78) to generate a candidate word list.

[0142] Administrators assess the legal risk relevance of new words through dual review, confirming weights (1.0, 0.5, 0.2) and placing eligible words in their corresponding positions for later risk calculation. The new words are then injected into the test environment, and 1000 historical cases are used to verify score changes. Simultaneously, manual verification checks are performed to ensure that changes in the high-risk case recognition rate and false positive rate remain within specified thresholds; if these conditions are met, the verification is successful.

[0143] In summary, this invention addresses the superficiality of existing sentiment analysis technologies by constructing a tiered sentiment lexicon and employing a scientific weighting scheme (high 1.0, medium 0.5, low 0.2). The system can accurately distinguish between different intensities of emotion, such as "anger" (1.0), "dissatisfaction" (0.5), and "hesitation" (0.2), providing precise numerical data for risk assessment. For example, when words like "murder" (1.0) appear in consultation texts, the system automatically identifies their high-risk attributes, rather than simply labeling them as "negative emotions."

[0144] This invention addresses the lack of social impact assessment by employing a systematic social impact vocabulary (high 1.0, medium 0.5, low 0.2) and classification system (public interest, government involvement, etc.). The system can automatically identify high-social-impact cases such as "vaccine safety issues" (1.0) and distinguish them from ordinary cases like "property fee disputes" (0.2). The assessment framework also considers the direction of impact (such as the protection of special groups), ensuring that sensitive cases receive sufficient attention.

[0145] This invention accurately quantifies risk through an emotion-social influence gating risk scoring model. Risk value prediction utilizes a weighted integration of multiple dimensions, including emotion intensity and social influence, to address the limitations of single-dimensional assessments. This model simultaneously considers all dimensions and their interactions, enabling the identification of "double high-risk" cases (high emotion + high social influence). Source tracing analysis, based on a combination of rules and models, addresses the "black box" problem. The system clearly displays the scoring criteria, such as "high-risk rating stems from: 1) emotion-related vocabulary; 2) social influence-related vocabulary." This interpretability not only enhances user trust but also facilitates regulatory review.

[0146] This invention addresses the rigidity of static models by employing a scalable social influence vocabulary adjustment mechanism. System administrators can add new keywords or adjust weights based on social changes reflected in consultation questions, ensuring the system remains up-to-date.

[0147] This invention addresses the ambiguity of qualitative assessments through precise mathematical modeling and a standardized scoring process. The system outputs clear risk scores (e.g., 0-100 points) and levels (high, medium, low), facilitating the development of differentiated response strategies and system performance evaluation.

[0148] The ultimate goal of this invention is to provide a scientific, comprehensive, and accurate legal consultation risk assessment tool that can identify high-risk consultations early, allocate resources rationally, prevent potential legal crises and social events, while meeting transparency requirements and promoting the intelligentization of legal services and the modernization of social governance.

[0149] Although the present invention has been described through the above preferred embodiments, those skilled in the art will understand that various modifications in form and detail can be made therein without departing from the spirit and scope of the invention. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A legal consultation risk assessment method that integrates emotional intensity and social influence, characterized in that, The method includes: A sentiment lexicon strength table is constructed, and the sentiment score of the consultation text is calculated based on this sentiment lexicon strength table. The emotional vocabulary strength table includes: emotional intensity weights, negative emotion example words, and positive emotion example words. Construct a social influence intensity scale and, in conjunction with this scale, calculate the social influence score of the consultation text. The social influence intensity table includes: social influence intensity weights and example words. Based on sentiment score Social Influence Score The risk score of the consultation text was obtained. ; In this process, the sentiment score of the consultation text is calculated by combining the sentiment lexicon strength table. ,include: The RoBERTa-Large model was used to analyze the sentiment polarity of the consultation text, and the sentiment polarity coefficient of the text was finally output. , ; Identify sentiment words contained in the consultation text And retrieve the sentiment word from the sentiment lexicon strength table. After obtaining the corresponding example words for positive or neutral emotions, you will get the corresponding emotion word. Corresponding emotional intensity weight ; Based on the text sentiment polarity coefficient and intensity weight The emotional score of the consultation text was obtained. ;in, This indicates the number of sentiment words contained in the consultation text. Then truncate it so that ; Among them, based on sentiment scoring Social Influence Score The risk score of the consultation text was obtained. ,include: Based on sentiment score To calculate the sentiment value of the consultation text. , It indicates the steepness of the emotional transition. The classification threshold representing the intensity of emotion; Based on social influence score To calculate the social impact value of the consultation text. , This indicates the steepness of the transition in controlling influence. Indicates the threshold of influence. A benchmark value indicating low influence. Indicates the range of regulation with high influence. ; Based on sentiment score Social Influence Score Calculate the dynamic compensation value of the consultation text. , Indicates the compensation weight. This represents the exponential parameter. Used to control social influence scores For compensation value The form of influence; Based on sentiment value Social Influence Value and dynamic compensation value Calculate the risk score of the consultation text. .

2. The method according to claim 1, characterized in that, Calculate the social impact score of the consultation text by combining the social impact strength scale. ,include: The consultation text and social influence strength table are passed to the large language model to obtain the social influence strength weight of the consultation text. And the social impact value corresponding to the consultation text. ; Based on the weight of social influence and the social impact value The social impact score of the consultation text was obtained. .

3. The method according to any one of claims 1 to 2, characterized in that, Based on sentiment score Social Influence Score After obtaining the risk score of the consultation text, the method further includes: If the risk score of the consultation text is greater than a risk threshold, a token-level attention matrix for the consultation text is constructed using a Transformer model. ; Token-level attention matrix The attention weights of each element are used to obtain candidate keywords from the consultation text; Based on the emotional lexicon strength table and the social influence strength table, obtain the negative emotional intensity weight and social influence intensity weight of the candidate keyword. The multi-dimensional score of each candidate keyword is calculated by combining the attention weight, negative sentiment intensity weight, and social influence intensity weight. Based on the multi-dimensional scoring of candidate keywords, the keywords of this consultation text are obtained.

4. The method according to any one of claims 1 to 2, characterized in that, Based on sentiment score Social Influence Score After obtaining the risk score of the consultation text, the method further includes: Semantic analysis is performed on newly added consultation texts to identify out-of-vocabulary words in the texts. These out-of-vocabulary words are example words that do not appear in the social influence intensity table. By calculating the semantic similarity between out-of-vocabulary words and example words, the social influence strength weight of the out-of-vocabulary word is determined, and the social influence strength table is updated.

5. A legal consultation risk scoring system that integrates emotional intensity and social influence, characterized in that, The system includes: The sentiment scoring module is used to construct a sentiment lexicon strength table and, in conjunction with this sentiment lexicon strength table, calculate the sentiment score of the consultation text. The emotional vocabulary strength table includes: emotional intensity weights, negative emotion example words, and positive emotion example words. The influence scoring module is used to construct a social influence strength table and, in conjunction with this table, calculate the social influence score of the consultation text. The social influence intensity table includes: social influence intensity weights and example words. The risk scoring module is used for sentiment-based scoring. Social Influence Score The risk score of the consultation text was obtained; In this process, the sentiment score of the consultation text is calculated by combining the sentiment lexicon strength table. ,include: The RoBERTa-Large model was used to analyze the sentiment polarity of the consultation text, and the sentiment polarity coefficient of the text was finally output. , ; Identify sentiment words contained in the consultation text And retrieve the sentiment word from the sentiment lexicon strength table. After obtaining the corresponding example words for positive or neutral emotions, you will get the corresponding emotion word. Corresponding emotional intensity weight ; Based on the text sentiment polarity coefficient and intensity weight The emotional score of the consultation text was obtained. ;in, This indicates the number of sentiment words contained in the consultation text. Then truncate it so that ; Among them, based on sentiment scoring Social Influence Score The risk score of the consultation text was obtained. ,include: Based on sentiment score To calculate the sentiment value of the consultation text. , It indicates the steepness of the emotional transition. The classification threshold representing the intensity of emotion; Based on social influence score To calculate the social impact value of the consultation text. , This indicates the steepness of the transition in controlling influence. Indicates the threshold of influence. A benchmark value indicating low influence. Indicates the range of regulation with high influence. ; Based on sentiment score Social Influence Score Calculate the dynamic compensation value of the consultation text. , Indicates the compensation weight. This represents the exponential parameter. Used to control social influence scores For compensation value The form of influence; Based on sentiment value Social Influence Value and dynamic compensation value Calculate the risk score of the consultation text. .

6. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the legal consultation risk scoring method that integrates emotional intensity and social influence as described in any one of claims 1-4.

Images