Baselevel contradictory dispute risk assessment method, system and device and storage medium

Abstract

The invention discloses a basic-level contradictory dispute risk assessment method, system and device and a storage medium, which are in one-to-one correspondence schemes, and in the scheme, a thinking chain reasoning technology and a knowledge distillation technology are fused, event characteristics are analyzed step by step by designing a structured Prompt instruction (cue word) guide model, and a risk assessment result is obtained. And the capability of the large language model is migrated to the lightweight small model so as to facilitate localized deployment, so that efficient and explainable basic-level contradictory dispute risk assessment can be realized for contradictory dispute events reported by various channels, so that related workers can reasonably arrange and carry out work according to assessment results, and the risk assessment efficiency is improved. And a thinking chain assessment process included in an assessment result output by the risk assessment model can also assist in manual auditing and decision making, so that the handling efficiency of basic social transactions is improved, and the personnel burden and the application cost are reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and natural language processing, and particularly to a method, system, device and storage medium for risk assessment of grass-roots contradiction disputes. Background Art

[0002] With the increasing complexity of grass-roots social governance, the diversity and timeliness requirements of contradiction dispute events have increased significantly. Traditional risk assessment methods rely on manual experience judgment, and there are problems such as strong subjectivity, low efficiency, and limited coverage scenarios.

[0003] In the field of grass-roots social governance, some methods based on deep learning and big data often perform public opinion analysis on a large amount of network information. For example, the Chinese invention patent "Risk Early Warning Method, Device, Equipment and Storage Medium Based on Social Events" with the authorization announcement number CN115545573B and the Chinese invention patent application "Natural Language Processing Method for Warning Risk Upgrade Events" with the publication number CN114328907A both collect a large amount of event texts and then perform comprehensive risk assessment through methods such as word segmentation, clustering, and entity extraction. Among them, a large number of manually designed feature extraction methods are involved, lacking interpretability, and it is difficult to give an assessment result for a single event. In recent years, large language models, such as GPT-4 (Generative Pretrained Transformer-4), DeepSeek (a model launched by DeepSeek), etc., have performed excellently in text analysis tasks with their powerful semantic understanding and logical reasoning capabilities. However, the existing large models have the following problems in direct deployment: (1) High computational resource consumption: Large models have a huge number of parameters and high inference costs, making it difficult to be applied in real time in grass-roots governance scenarios.

[0004] (2) Insufficient domain adaptability: General large models lack targeted training for specific risk indicators of grass-roots contradictions.

[0005] (3) Weak interpretability: The risk assessment results lack a transparent reasoning process and are difficult to assist in manual decision-making.

[0006] In view of this, the present invention is specifically proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a method, system, device and storage medium for risk assessment of grass-roots contradiction disputes, which can achieve efficient and interpretable risk assessment of grass-roots contradiction disputes.

[0008] The purpose of the present invention is achieved through the following technical solutions: A method for risk assessment of grass-roots contradiction disputes includes: Constructing a Prompt instruction including a risk level and an evaluation thinking chain, where the Prompt instruction is a prompt word; Collect the disposal archives of historical conflict and dispute events, and conduct risk assessment through a large language model in combination with the Prompt instructions to construct a risk assessment dataset; Select a pre-trained model, and adopt the method of knowledge distillation to train the pre-trained model in combination with the risk assessment dataset, transfer the knowledge and reasoning ability of the large language model to the pre-trained model to obtain a risk assessment model; wherein, the number of parameters T1 of the pre-trained model and the number of parameters T2 of the large language model satisfy T1 << T2, and << is the symbol of much less than; Input the conflict and dispute event to be evaluated into the risk assessment model, and the risk assessment model gradually analyzes the event characteristics in combination with the evaluation thought chain and outputs the risk assessment result.

[0009] A grass-roots conflict and dispute risk assessment system for implementing the foregoing method, comprising: An instruction construction unit for constructing a Prompt instruction including a risk level and an evaluation thought chain, wherein the Prompt instruction is a prompt word; A risk assessment data generation unit for collecting the disposal archives of historical conflict and dispute events, and conducting risk assessment through a large language model in combination with the Prompt instruction to construct a risk assessment dataset; A model training unit for selecting a pre-trained model, adopting the method of knowledge distillation to train the pre-trained model in combination with the risk assessment dataset, transfer the knowledge and reasoning ability of the large language model to the pre-trained model to obtain a risk assessment model; wherein, the number of parameters T1 of the pre-trained model and the number of parameters T2 of the large language model satisfy T1 << T2, and << is the symbol of much less than; A conflict and dispute risk assessment unit for inputting the conflict and dispute event to be evaluated into the risk assessment model, and the risk assessment model gradually analyzes the event characteristics in combination with the evaluation thought chain and outputs the risk assessment result.

[0010] A processing device, comprising: one or more processors; a memory for storing one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the foregoing method.

[0011] A readable storage medium storing a computer program, which implements the foregoing method when the computer program is executed by a processor.

[0012] As can be seen from the technical solution provided by the present invention above, by integrating Chain-of-Thought Reasoning and Knowledge Distillation, designing structured Prompt instructions to guide the model to gradually analyze event features, and migrating the capabilities of large language models to lightweight small models, efficient and interpretable grass-roots contradiction and dispute risk assessment can be achieved, so as to better assist human decision-making, and further improve the disposal efficiency of grass-roots social affairs, reduce personnel burden and operation costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0014] Figure 1 It is a flowchart of a grass-roots contradiction and dispute risk assessment method provided by an embodiment of the present invention; Figure 2 It is a flowchart of the chain of thought provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of updating the data set D according to expert feedback data provided by an embodiment of the present invention; Figure 4 It is a schematic diagram of risk assessment provided by an embodiment of the present invention; Figure 5 It is a schematic diagram of model distillation, deployment and later information update provided by an embodiment of the present invention; Figure 6 It is a schematic diagram of a grass-roots contradiction and dispute risk assessment system provided by an embodiment of the present invention; Figure 7 It is a schematic diagram of a processing device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] First, the terms that may be used in this article are described as follows: Descriptions using terms such as "comprising", "including", "containing", "having" or other similar semantics shall be construed as non-exclusive inclusion. For example, including a technical feature element (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, processes, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products or articles, etc.) shall be construed as not only including the explicitly listed technical feature element, but also including other technical feature elements well-known in the art that are not explicitly listed.

[0017] The term "consisting of" means excluding any technical feature element not explicitly listed. If this term is used in a claim, it will make the claim a closed type, excluding technical feature elements other than the explicitly listed ones, except for conventional impurities related thereto. If this term only appears in a sub-clause of a claim, then it only limits the elements explicitly listed in that sub-clause, and the elements recorded in other sub-clauses are not excluded from the overall claim.

[0018] Unless otherwise explicitly specified or limited, terms such as "installed", "connected", "joined", "fixed", etc. shall be understood in a broad sense. For example: it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in this article can be understood according to specific circumstances.

[0019] The following provides a detailed description of a method, system, device, and storage medium for risk assessment of grass-roots contradictions and disputes provided by the present invention. The content not described in detail in the embodiments of the present invention belongs to the prior art well-known to those of ordinary skill in the art. In the embodiments of the present invention, for those not specified in detail, they are carried out according to the conventional conditions in the art or the conditions recommended by the manufacturer. For the reagents or instruments used in the embodiments of the present invention without indicating the manufacturer, they are all conventional products that can be obtained through commercial purchase.

[0020] Embodiment 1 The embodiment of the present invention provides a method for risk assessment of grass-roots contradictions and disputes, as Figure 1 shown, mainly including the following steps: Step 1: Construct a Prompt instruction including a risk level and an evaluation thinking chain.

[0021] In the embodiment of the present invention, the Prompt instruction is a prompt word used to guide the model to make corresponding answers. The model here covers the large language model, pre-trained model, and risk assessment model mentioned later.

[0022] In the embodiments of the present invention, the risk assessment is decoupled into two aspects: urgency and severity, and corresponding Prompt instructions are constructed respectively; among them, in the Prompt instructions corresponding to urgency and severity, N risk level grades are respectively set, and from 1 to N represents a gradual decrease in the risk level; in the Prompt instructions corresponding to urgency and severity, corresponding evaluation thinking chains are respectively set, which are used to guide the model to analyze the contradiction and dispute events step by step and output the risk level grade.

[0023] Preferably, the evaluation thinking chain in the Prompt instruction corresponding to the urgency includes five steps, which are in turn: (1) Analyze the event background to clarify the cause, participants and current situation of the event; (2) Predict the evolution of the event situation; (3) Analyze the duration of the disposal window of the event; (4) According to the results of the first three steps, analyze and match with the risk level grades of urgency one by one to determine the risk level grade; (5) Make a comprehensive judgment and output the evaluation result.

[0024] Preferably, the evaluation thinking chain in the Prompt instruction corresponding to the severity includes four steps, which are in turn: (1) Analyze the event background to clarify the cause, participants and current situation of the event; (2) Predict the trend of the event; (3) According to the results of the first two steps, analyze and match with the risk level grades of severity one by one to determine the risk level grade; (4) Make a comprehensive judgment and output the evaluation result.

[0025] In the evaluation results of the above two aspects of urgency and severity, both include the risk level grade (that is, the urgency level grade and the severity level grade), and the corresponding thinking chain analysis and reasoning process.

[0026] Step 2: Construct a risk assessment data set.

[0027] In the embodiments of the present invention, collect the historical contradiction and dispute event disposal files, and conduct risk assessment through the large language model combined with the Prompt instructions to construct a risk assessment data set; its preferred implementation method is as follows: (1) Collect the historical contradiction and dispute event disposal files and sort out the text descriptions of the historical contradiction and dispute events; (2) Input the text descriptions of the historical contradiction and dispute events and the corresponding Prompt instructions into the large language model, and the large language model outputs the risk assessment results for the historical contradiction and dispute events; among them, the evaluation results marked corresponding to the historical contradiction and dispute events are also added to the Prompt instructions as a reference; (3) Finally, comprehensively construct a risk assessment data set based on the text descriptions of the historical contradiction and dispute events and the corresponding risk assessment results.

[0028] Step 3: Model training based on knowledge distillation.

[0029] Considering that large language models have strong reasoning capabilities, but their parameter counts of up to hundreds of billions make it difficult to deploy them locally, it is necessary to use knowledge distillation methods to distill their capabilities into lightweight small models.

[0030] In an embodiment of the present invention, a pre-trained model (which is a lightweight small model) is selected, and in a knowledge distillation manner, the pre-trained model is trained in combination with the risk assessment data set to transfer the knowledge and reasoning capabilities of the large language model to the pre-trained model, obtaining a risk assessment model; wherein, the parameter count T1 of the pre-trained model and the parameter count T2 of the large language model satisfy T1 << T2, and << is the symbol for much less than.

[0031] In an embodiment of the present invention, the training of the pre-trained model in combination with the risk assessment data set includes: in combination with the risk assessment data set, distillation training is performed on the pre-trained model using supervised fine-tuning, and during the supervised fine-tuning process, a cross-entropy loss function is used.

[0032] Step 4: Risk assessment of contradictions and disputes.

[0033] In an embodiment of the present invention, the contradiction and dispute event (text) to be evaluated is input into the risk assessment model, and the risk assessment model gradually analyzes the event characteristics in combination with the evaluation thought chain and outputs the risk assessment result.

[0034] Preferably, the risk assessment model can also be deployed on the grass-roots governance platform and a threshold is set; when the risk level in the output risk assessment is higher than the threshold, an alarm is triggered and pushed to the corresponding handling department.

[0035] Moreover, feedback data on the handling results is collected regularly and used to update the Prompt instructions and the parameters of the risk assessment model; specifically: during actual deployment, if the risk assessment result output by the risk assessment model does not match the expert's assessment result, feedback data on the handling results will be generated. The feedback data includes the thought-chain analysis and reasoning process of the risk assessment model and the expert's assessment result; the expert's assessment result is informed to the risk assessment model, and it is allowed to regenerate the assessment process with reference to the expert's assessment result, and fine-tuning data is constructed in combination with the regenerated assessment result. When the amount of fine-tuning data reaches the set quantity, the risk assessment model is fine-tuned.

[0036] In order to more clearly show the technical solutions provided by the present invention and the technical effects produced, the following uses specific embodiments to describe in detail the method provided by the embodiments of the present invention.

[0037] 1. A Prompt instruction is constructed.

[0038] In the embodiments of the present invention, the event risk is decoupled into two aspects: urgency and severity, and different risk level definitions and thought chain evaluation templates are designed for the two, guiding the model to analyze the event step by step, and finally constructing Prompt instructions for the two evaluation tasks respectively.

[0039] Exemplarily, for the urgency evaluation, it is defined into 5 levels, where from 1 to 5 represents a gradual decrease in urgency; for the severity evaluation, it is defined into 5 levels, where from 1 to 5 represents a gradual decrease in severity.

[0040] As Figure 2 shown, the thought chain of the urgency evaluation is designed in five steps. First, analyze the event background to clarify the cause, participants and current situation of the event. Secondly, predict the evolution of the event situation; then, analyze the duration of the disposal window of the event, and then analyze whether it matches level by level according to the definition of urgency, and finally make a comprehensive judgment to give the evaluation result. The thought chain of the severity evaluation is designed in four steps. First, analyze the event background to clarify the cause, participants and current situation of the event. Secondly, predict the trend of the event, and then analyze whether it matches level by level according to the definition of severity, and finally make a comprehensive judgment to give the evaluation result.

[0041] The following respectively provide examples of the Prompt instructions for the urgency evaluation and the severity evaluation (Example 1 and Example 2). It should be noted that the content of the Prompt instructions in Example 1 and Example 2 is only a feasible example, and in actual applications, users can adjust according to the situation.

[0042] Example 1: Prompt instructions for the urgency evaluation: {'role':'system', 'content': 'Role definition: You are an expert specializing in evaluating the urgency of conflict and dispute events, and need to make a comprehensive judgment by combining the dynamic characteristics of the event and the disposal window period; Task background: Evaluate the urgency of the conflict and dispute events provided by the user. If the user does not provide a description of a conflict and dispute event, politely decline and tell the user your role positioning; Core principles: 1. Give priority to ensuring the safety of people's lives and property and social stability; 2. Improve the disposal priority for events that have had a substantial adverse impact; 3. Focus on the diffusibility and demonstration effect of the conflict Definition of urgency: 1 - Extremely urgent event: Must be responded to immediately (within several hours). If no immediate action is taken, the event will quickly escalate and pose a greater risk to the lives and property of the people; ​2 - Highly Urgent Incidents: Incidents that require prompt handling (within 12 hours - 2 days). Although it will not get out of control immediately, it has already caused adverse effects, and delays will make the situation more complex or dangerous; 3 - Moderately Urgent Incidents: Incidents that need to be resolved within a relatively short period (several days to one week). Delaying the handling will not immediately result in serious losses of life and property or increase the pressure on public management, but it may exacerbate the dissatisfaction of individuals or groups, or accumulate contradictions to a more difficult - to - solve level; 4 - Generally Urgent Incidents: Incidents that need to be handled within a reasonable time frame (within 2 - 3 weeks). Although there is no immediate risk or harm, not handling it for a long time may exacerbate the dissatisfaction of the parties involved and affect community harmony; 5 - Slightly Urgent Incidents: Low urgency, can be handled according to arrangements within a relatively long time (within 1 - 2 months). Such incidents will not cause any direct urgent impact on the parties involved or society as a whole, and usually involve non - confrontational trivial civil disputes or long - standing contradictions that do not further deteriorate; Please analyze and evaluate the given incident according to the given definition of the urgency level, and give your evaluation results according to the following ideas: 0. Reminder: Please be sure to evaluate strictly according to the definition to ensure that the evaluation results are reasonable; 1. Incident Background Analysis: Briefly analyze the cause of the incident, the parties involved, and the current situation of the incident; 2. Situation Evolution Analysis: Predict the possible development trend, escalation speed, and damage degree of the incident, etc., to ensure reasonable logic; 3. Incident Handling Window Analysis: Analyze how long the incident needs to be handled, otherwise the meaning or opportunity for handling will be lost, which may lead to the escalation of the situation; for some incidents, key time nodes such as seasons and farming seasons may need to be considered; 4. Urgency Level Evaluation: Combine the above analysis with the definition of the urgency level, and evaluate one by one according to the level, giving reasons for compliance or non - compliance; 5. Comprehensive Judgment: According to the evaluation results, give the final urgency level score and provide a reasonable analysis and explanation; Finally, output in the following json format, starting directly from {, do not output ```json: { "Incident Background Analysis": <Briefly analyze the cause of the incident, the parties involved, and the current situation of the incident>, "Situation Evolution Analysis": <Predict the possible development and consequences of the incident>, "Incident Handling Window Analysis": <Analyze how long the incident needs to be handled>. "Urgency Level Evaluation": "1 - Extremely Urgent Incidents: <Evaluation results, reasons for compliance or non - compliance>", "2 - Highly Urgent Incident: <Assessment result, reasons for compliance or non - compliance>", "3 - Moderately Urgent Incident: <Assessment result, reasons for compliance or non - compliance>", "4 - General Urgent Incident: <Assessment result, reasons for compliance or non - compliance>", "5 - Slightly Urgent Incident: <Assessment result, reasons for compliance or non - compliance>" , "Comprehensive Judgment": <Final judgment of the urgency level>, "Assessment Result": <Provide the final assessment result in Arabic numerals, e.g., 1 or 2 or 3 or 4 or 5> }' }, {'role': 'user', 'content': 'Please help me assess the urgency level of a conflict and dispute incident. The following are the details of the incident: <User - input description of the conflict and dispute incident>' }

[0043] Example 2: Severity Assessment Prompt Instruction: {'role':'system', 'content': 'Role Definition: You are an expert specializing in assessing the severity of conflict and dispute incidents; Task Background: Assess the severity of the conflict and dispute incident provided by the user. If the user does not provide a description of a conflict and dispute incident, politely decline and inform the user of your role; Core Principles: 1. The safety of people's lives and property and social harmony and stability come first; 2. Systematically consider the potential impact scope, consequence severity, conflict conduction risk, etc. of the incident; Severity Definition: 1 - Extremely Severe Incident: May pose a relatively large risk to the lives and property of the people, have obvious potential safety hazards, or increase the pressure on public management. If not handled properly, it may cause serious consequences, such as threats to personal safety, etc., and even affect social public safety or trigger an opinion event; 2 - Highly Severe Incident: May pose a threat to the life and property safety of a certain family or individual, have potential safety hazards, and may cause serious consequences under specific conditions; ​​3 - Moderately serious incidents: have a significant impact on the lives of some groups or individuals, but will not immediately cause serious consequences. Although these problems do not directly threaten people or property, they may have a significant impact on the quality of life or mood of the parties involved and the surrounding people; 4 - General serious incidents: have some impact on the life of the parties involved, but will not bring substantial harm. The incident has a small impact on the society as a whole, and even if it is not resolved immediately, it will not lead to major consequences; 5 - Mildly serious incidents: The impact is extremely limited and has almost no obvious consequences for society or the parties involved. Such incidents are usually trivial or personal disputes and do not pose a threat to social public order or economic interests; Please analyze and evaluate the specific incident according to the given severity definition and give your evaluation results as follows: 0. Reminder: Please make sure to strictly evaluate according to the definition to ensure that the evaluation results are reasonable; 1. Event background analysis: briefly analyze the cause of the event, the parties involved and the current status of the event; 2. Event trend analysis: predict the possible development and consequences of events to ensure logical rationality; 3. Severity assessment: Based on the above analysis, evaluate the severity of each event against the definition of severity; 4. Comprehensive judgment: Based on the evaluation results, give a final severity score and provide a reasonable analysis and explanation; The final output is in the following json format, starting directly from {, do not output ```json: { "Event Background Analysis": <Briefly analyze the cause of the event, the parties involved, and the current status of the event>, "Event Trend Analysis": <predict the possible development and consequences of an event>, "Severity Assessment": [ "1 - Extremely serious incident: <evaluation result, reason for compliance or noncompliance>", "2 - High severity incident: <assessment result, reason for compliance or noncompliance>", "3 - Moderately severe incident: <evaluation result, reason for compliance or noncompliance>", "4 - General serious incident: <Assessment results, reasons for compliance or noncompliance>", "5 - Minor severity incident: <Assessment result, reason for compliance or noncompliance>" ], "Comprehensive judgment": <final severity judgment>, "Evaluation Result": Provide the final evaluation result in Arabic numerals, e.g., 1 or 2 or 3 or 4 or 5 }' }, {"role": "user", "content": 'Please help me evaluate the severity of a conflict and dispute incident. The following are the details of the incident: <User input description of the conflict and dispute incident>' }

[0044] II. Generation of high-quality evaluation datasets.

[0045] First, extract the text descriptions of conflict and dispute incidents from the existing social conflict and dispute incident handling archives in a certain place. Then, use advanced large language models, such as DeepSeek V3 (the V3 version of the model launched by DeepSeek), combined with the previously constructed Prompt instructions, to generate risk assessment results for historical conflict and dispute incidents. When generating, add the evaluation results of some typical incidents marked by experts to the Prompt instructions as a reference. Finally, obtain a high-quality risk assessment dataset D for grass-roots social conflict and disputes.

[0046] III. Model training based on knowledge distillation.

[0047] Although large language models such as DeepSeek V3 provided previously have strong capabilities, their parameter counts in the tens of billions make it difficult to deploy them locally. Therefore, knowledge distillation methods are needed to distill their capabilities into lightweight small models (such as Qwen2.5-7B). Here, Qwen2.5-7B is a model launched by Alibaba. Qwen (Thousand Questions) is the model name, 2.5 is the version number, and 7B (i.e., 7 billion) represents the number of parameters. It belongs to a pre-trained model.

[0048] After selecting the small model , use SFT (supervised fine-tuning) for distillation training. Using the data D obtained above as the training set, transfer the knowledge and reasoning capabilities of the large language model to the lightweight model, reduce the parameter count while retaining key reasoning capabilities, and support low-cost local deployment. During the SFT process, use the CEL loss (cross-entropy loss) Loss, and the calculation formula is as follows: ; where i represents the i-th sample, is the predicted probability of the word output by the model , is the one-hot encoding of the target word (usually the true vocabulary), and N is the number of samples in a batch during training. ​

[0049] Finally, through training, a small model M that can be used for local deployment is obtained. When a user inputs a text description of a contradiction and dispute event, the model can conduct a risk assessment on it. This trained small model M is called a risk assessment model.

[0050] IV. Model Deployment and Feedback Optimization.

[0051] In the embodiments of the present invention, the risk assessment model can be integrated into the grass-roots governance platform to support real-time text input and risk assessment; as Figure 3 shown, the evaluation results output by the risk assessment model can be used to assist manual review and decision-making.

[0052] Moreover, low-risk events can also be filtered by setting a threshold. When high-risk events are identified, an alarm is automatically triggered and pushed to the handling department. The definition criteria for high and low risks here are whether the risk level in the evaluation result is higher than the threshold. If it is higher, it is an initial high-risk event; otherwise, it belongs to a low-risk event.

[0053] In addition, the handling result feedback can be collected regularly to update the risk level definition, the thought chain evaluation template, and the risk assessment model parameters.

[0054] Specifically: In actual deployment, the model evaluation results may not conform to the expert opinions, thus generating feedback data. The expert feedback data includes the historical evaluation process and the expert's evaluation results. On this basis, a dataset update method that aligns with the expert standards is designed: inform the risk assessment model of the expert evaluation results and let it regenerate the evaluation process with reference to this result. Then, delete the expert feedback data and use the newly generated evaluation results as fine-tuning data. When the fine-tuning data reaches the set quantity, the risk assessment model for local deployment can be updated. As Figure 4 shown, a schematic diagram of updating the dataset D according to the expert feedback data is provided. The system therein refers to the grass-roots governance platform; in this process, the regeneration instruction is used to guide the risk assessment model to regenerate the evaluation process in combination with the expert evaluation results. Exemplarily, its content can be: Given that a human expert evaluates the xx nature of this event as level x, please refer to this result and rethink the evaluation process according to the previous requirements. Note that you should not explicitly state that you know this information during the thinking process.

[0055] As Figure 5 shown, it is a schematic diagram of model distillation, deployment, and later information update.

[0056] The above solution provided by the embodiments of the present invention integrates Chain-of-Thought Reasoning and Knowledge Distillation technologies. By designing structured Prompt instructions to guide the model to gradually analyze event features and transferring the capabilities of large language models to lightweight small models, it can achieve local processing. Based on this, the present invention can achieve efficient and interpretable risk assessment of grass-roots contradiction and dispute events reported through various channels, so that relevant staff can reasonably arrange and carry out work according to the assessment results. Moreover, the Chain-of-Thought assessment process included in the assessment results output by the risk assessment model can also assist manual review and decision-making, thereby improving the disposal efficiency of grass-roots social affairs, reducing the burden and operation cost of grass-roots staff, improving the satisfaction of residents' services, and enhancing the level of grass-roots social governance.

[0057] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiments can be implemented by software or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions of the above embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments of the present invention.

[0058] Embodiment 2 The present invention also provides a grass-roots contradiction and dispute risk assessment system, which is mainly used to implement the method provided by the foregoing embodiments, as Figure 6 shown. The system mainly includes: An instruction construction unit, configured to construct a Prompt instruction including a risk level and an assessment thought chain, where the Prompt instruction is a prompt word; A risk assessment data generation unit, configured to collect historical contradiction and dispute event disposal files and perform risk assessment through a large language model in combination with the Prompt instruction to construct a risk assessment data set; A model training unit, configured to select a pre-trained model, and in a knowledge distillation manner, train the pre-trained model in combination with the risk assessment data set, transfer the knowledge and reasoning ability of the large language model to the pre-trained model, and obtain a risk assessment model; where the number of parameters T1 of the pre-trained model and the number of parameters T2 of the large language model satisfy T1 << T2, and << is the symbol for much less than; The contradiction and dispute risk assessment unit is used to input the contradiction and dispute events to be evaluated into the risk assessment model. The risk assessment model gradually analyzes the event characteristics in combination with the evaluation thinking chain and outputs the risk assessment result.

[0059] Considering that the main technical details involved in the above system have been introduced in detail in the previous embodiments, they will not be elaborated here.

[0060] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the system is divided into different functional modules to complete all or part of the functions described above.

[0061] Embodiment III The present invention also provides a processing device, such as Figure 7 shown, which mainly includes: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the foregoing embodiments.

[0062] Furthermore, the processing device further includes at least one input device and at least one output device; in the processing device, the processor, the memory, the input device, and the output device are connected through a bus.

[0063] In the embodiments of the present invention, the specific types of the memory, the input device, and the output device are not limited; for example: The input device can be a touch screen, an image acquisition device, a physical button, or a mouse, etc.; The output device can be a display terminal; The memory can be a random access memory (RAM), or a non-volatile memory, such as a disk memory.

[0064] Embodiment IV The present invention also provides a readable storage medium storing a computer program, which implements the method provided in the foregoing embodiments when the computer program is executed by a processor.

[0065] In the embodiments of the present invention, the readable storage medium as a computer-readable storage medium can be disposed in the foregoing processing device, for example, as the memory in the processing device. In addition, the readable storage medium can also be various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disc.

[0066] As described above, it is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims. The information disclosed in the background art part of this article is only intended to deepen the understanding of the overall background art of the present invention, and should not be regarded as an admission or imply in any form that this information constitutes the prior art known to those skilled in the art.

Claims

1. A risk assessment method for grass-roots contradictions and disputes, characterized in that Including: Construct a Prompt instruction that includes risk level grades and an evaluation thought chain, where the Prompt instruction is a prompt word; Collect historical archives of handling contradictory disputes, and conduct risk assessment through a large language model in combination with the Prompt instruction to construct a risk assessment data set; Select a pre-trained model, and use the method of knowledge distillation to train the pre-trained model in combination with the risk assessment data set, transfer the knowledge and reasoning ability of the large language model to the pre-trained model to obtain a risk assessment model; among them, the number of parameters T1 of the pre-trained model and the number of parameters T2 of the large language model satisfy T1 << T2, and << is the much less than symbol; Input the contradictory dispute event to be evaluated into the risk assessment model, and the risk assessment model gradually analyzes the event characteristics in combination with the evaluation thought chain and outputs the risk assessment result.

2. The risk assessment method for grass-roots contradictions and disputes according to claim 1, wherein, The construction of the Prompt instruction that includes risk level grades and an evaluation thought chain includes: Decouple the risk assessment into two aspects: urgency and severity, and construct corresponding Prompt instructions respectively; Among them, in the Prompt instructions corresponding to urgency and severity, N risk level grades are respectively set, and from 1 to N represent a gradually decreasing risk level; In the Prompt instructions corresponding to urgency and severity, corresponding evaluation thought chains are respectively set to guide the model to analyze contradictory dispute events step by step and output risk level grades.

3. The risk assessment method for grass-roots contradictions and disputes according to claim 2, characterized in that, The evaluation thought chain in the Prompt instruction corresponding to urgency includes five steps, which are in turn: Analyze the event background to clarify the cause, participants and current situation of the event; Predict the evolution of the event situation; Analyze the disposal window duration of the event; According to the results of the first three steps, analyze and match them one by one with the risk level grades of urgency to determine the risk level grade; Make a comprehensive judgment and output the evaluation result.

4. The risk assessment method for grass-roots contradictions and disputes according to claim 2, characterized in that, The evaluation thought chain in the Prompt instruction corresponding to severity includes four steps, which are in turn: Analyze the event background to clarify the cause, participants and current situation of the event; Predict the trend of the event; According to the results of the first two steps, analyze and match them one by one with the risk level grades of severity to determine the risk level grade; Make a comprehensive judgment and output the evaluation result.

5. A risk assessment method for grass-roots contradiction and dispute, according to claim 1, characterized in that The collection of historical archives of handling contradictory disputes, and the construction of a risk assessment data set by conducting risk assessment through a large language model in combination with the Prompt instruction includes: Collect historical archives of handling contradictory disputes and sort out the written descriptions of historical contradictory dispute events; Input the written descriptions of historical contradictory dispute events and the corresponding Prompt instructions into the large language model, and the large language model outputs risk assessment results for historical contradictory dispute events; among them, the evaluation results marked for the corresponding historical contradictory dispute events are also added to the Prompt instruction as a reference; Finally, construct a risk assessment data set by integrating the written descriptions of historical contradictory dispute events and the corresponding risk assessment results.

6. The risk assessment method for grass-roots contradiction and dispute according to claim 1, characterized in that, The training of the pre-trained model in combination with the risk assessment data set includes: Combined with the risk assessment data set, the pre-trained model is distilled and trained using supervised fine-tuning. During the supervised fine-tuning process, a cross-entropy loss function is adopted.

7. A method for risk assessment of grass-roots contradiction and dispute, according to claim 1 or 6, characterized in that, It further includes: Deploy the risk assessment model to the grass-roots governance platform and set a threshold; When the risk level in the output risk assessment is higher than the threshold, trigger an alarm and push it to the corresponding disposal department; Regularly collect feedback data on disposal results, and update the Prompt instructions and the parameters of the risk assessment model accordingly; among them, updating the parameters of the risk assessment model includes: when the risk assessment result output by the risk assessment model does not match the expert's assessment result, generate feedback data on disposal results, and the feedback data includes the step-by-step analysis and reasoning process of the risk assessment model's chain of thought, as well as the expert's assessment result; inform the risk assessment model of the expert's assessment result and let it regenerate the assessment process with reference to the expert's assessment result, construct fine-tuning data based on the regenerated assessment result, and when the fine-tuning data reaches the set quantity, fine-tune the risk assessment model.

8. A grass-roots contradiction and dispute risk assessment system, characterized in that, For implementing the method according to any one of claims 1 to 7, it includes: An instruction construction unit, configured to construct a Prompt instruction including a risk level and an assessment chain of thought, where the Prompt instruction is a prompt word; A risk assessment data generation unit, configured to collect historical archives of handling contradictory disputes, and perform risk assessment through a large language model in combination with the Prompt instruction to construct a risk assessment data set; A model training unit, configured to select a pre-trained model, and train the pre-trained model in combination with the risk assessment data set in a knowledge distillation manner, transfer the knowledge and reasoning ability of the large language model to the pre-trained model to obtain a risk assessment model; among them, the number of parameters T1 of the pre-trained model and the number of parameters T2 of the large language model satisfy T1 << T2, and << is the symbol for much less than; A contradictory dispute risk assessment unit, configured to input the contradictory dispute event to be evaluated into the risk assessment model, and the risk assessment model gradually analyzes the event characteristics in combination with the assessment chain of thought and outputs a risk assessment result.

9. A processing device, characterized in that, It includes: One or more processors; A memory for storing one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A readable storage medium stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images