Method for carrying out risk data analysis based on LLM (Language Language Model)

Abstract

The invention provides a risk data analysis method based on an LLM large language model, risk data analysis is carried out based on the LLM large language model, data identification, classification and risk feature extraction can be carried out by utilizing the LLM large language model in the whole process, risk feature vector training learning is carried out on the LLM large language model, and the risk data analysis efficiency is improved. The defect that a large number of training samples are needed can be reduced; meanwhile, the LLM big language model cue words on all dimensions are constructed and input on one LLM big language model, the LLM big language model can continue to further carry out risk analysis on the classification and recognition result on all dimensions, and the risk analysis result of the risk data on all dimensions is rapidly and deeply mined and output. Therefore, the risk data can be intelligently processed by the LLM in the whole process, the risk types and all dimensions can be analyzed through one model, and the risk data analysis efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to a method for risk data analysis based on an LLM large language model, and a device, electronic device and computer-readable storage medium for risk data analysis based on an LLM large language model. Background Art

[0002] Risk data analysis methods mainly include web crawler technology, social media monitoring and risk data analysis. Although these methods play an important role in risk monitoring and analysis, they also have some shortcomings: First, the shortcomings of web crawler technology are mainly reflected in the data crawling process. Although it can automatically obtain a large amount of data from multiple sources and has a wide coverage, it may crawl some irrelevant information, which needs further screening and cleaning. In addition, it is more difficult to crawl dynamic web pages or content that requires login to access. Some websites may have anti-crawler mechanisms to limit data acquisition.

[0003] Secondly, the shortcomings of social media monitoring are mainly reflected in the amount of data and the authenticity of information. Although it can understand the public's emotions and opinions in real time, the amount of data is huge, and it is difficult to screen and analyze. At the same time, the authenticity and reliability of information on social media need to be verified. Different social media platforms have different user groups and characteristics, so comprehensive analysis is needed.

[0004] Finally, the disadvantages of risk data analysis are mainly reflected in complexity and time consumption. Although it can conduct in-depth analysis and mining of the collected data, identify hot issues and potential risks, the analysis process is relatively complex and time-consuming. In addition, the results of risk data analysis are easily affected by personal subjective factors, resulting in deviations in monitoring results. The interpretation of data still requires the judgment of professionals.

[0005] Therefore, although the traditional risk data analysis methods are simple to apply and can quickly implement risk analysis, they still have their own application shortcomings in actual applications and cannot ensure the accuracy and effectiveness of risk analysis. In addition, the traditional risk data analysis methods are not intelligent enough for risk data analysis and cannot achieve the effects of intelligent analysis and intelligent risk classification and identification and automated processing. Summary of the invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides the following technical solutions: On the one hand, a method for risk data analysis based on an LLM large language model is provided, the method being implemented by an electronic device, the method comprising: S1. Construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate LLM large language model-A for risk identification and classification; S2. Construct risk analysis rules and risk keywords on several analysis dimensions, and combine the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and input them into the LLM large language model-A in sequence; S3. Monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data; S4, using the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially perform risk analysis on the risk text data of this type, and output the risk analysis results on the corresponding analysis dimensions; S5. Add corresponding type labels to the risk analysis results in each analysis dimension, obtain a risk label analysis data set, and save it.

[0007] Preferably, the S1, constructing a risk feature vector based on risk big data and inputting it into a preset LLM large language model for model training, generating an LLM large language model-A for risk identification and classification, includes: Collecting the risk big data consisting of different types of risk data sets; Perform text processing on each risk data in the risk big data based on NLP technology to generate corresponding risk NLP data; Using a word embedding method, converting the risk NLP data into a text vector to obtain a corresponding risk feature vector; After the conversion is completed, a feature set constructed by a number of risk feature vectors is statistically obtained; Dividing the feature set into a training set and a validation set according to a preset ratio; Input the training set into a preset LLM large language model, and let the LLM large language model perform model training on each of the risk feature vectors in the training set to generate the LLM large language model-A; The recognition performance of the LLM large language model-A is verified using the verification set: If the verification is successful, the LLM large language model-A is deployed to the risk supervision system and put into use after parameter adjustment; If the verification fails, the above training steps are repeated to rebuild the LLM large language model-A.

[0008] Preferably, the S2, constructing risk analysis rules and risk keywords on several analysis dimensions, and combining the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and sequentially inputting them into the LLM large language model-A, includes: Collect user data analysis requirements on risk data in different analysis dimensions from the front end; The risk analysis rules and risk keywords in the data analysis requirements are extracted using the LLM large language model, and the risk analysis prompt words corresponding to the analysis dimension are automatically constructed; Deploy a distributed file system HDFS in the background to store the risk analysis results on each analysis dimension obtained in step S4; The risk analysis prompt words on each of the analysis dimensions are sent to the management node (Name Node) of the distributed file system HDFS, and the management node (Name Node) allocates corresponding storage nodes (Data Node) to the risk analysis prompt words on each of the analysis dimensions, and configures corresponding attribute values ​​containing the analysis dimensions at the storage nodes (Data Node).

[0009] Preferably, the step S3, monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data, includes: activating a risk monitoring system and monitoring risk data in real time through the risk monitoring system; After the corresponding risk data is monitored, the risk data is collected and sent to the pre-deployed LLM large language model-A, and the LLM large language model-A monitors the risk feature vector of the risk data and outputs the corresponding risk type, and at the same time parses and outputs the risk text data corresponding to the risk feature vector; The risk text data is cached in a background database.

[0010] Preferably, the S4, through the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially performs risk analysis on the risk text data of this type, and outputs the risk analysis results on the corresponding analysis dimensions, including: Sending a risk analysis request to the management node (Name Node); The management node (Name Node) responds to the risk analysis request, sequentially retrieves the risk analysis prompt words configured at each storage node (Data Node) and feeds them back to the LLM large language model-A; Based on the risk analysis prompt words, the risk text data is sequentially analyzed by the LLM large language model-A, and the risk analysis results on the corresponding analysis dimensions are output; The risk analysis result is fed back to the management node (Name Node), and the management node (NameNode) stores the risk analysis result in the corresponding storage node (Data Node).

[0011] On the other hand, a device for performing risk data analysis based on an LLM large language model is provided, wherein the device for performing risk data analysis based on an LLM large language model is used to implement the method for performing risk data analysis based on an LLM large language model, and the device comprises: The large language model training module is used to construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate the LLM large language model-A for risk identification and classification; A prompt word construction module is used to construct risk analysis rules and risk keywords on several analysis dimensions, and to construct risk analysis prompt words on corresponding analysis dimensions in combination with the risk analysis rules and the risk keywords, and to input the prompt words into the LLM large language model-A in sequence; A risk monitoring module, used to monitor risk data through the LLM large language model-A, identify and classify the risk type of the monitored risk data and the corresponding type of risk text data; A risk analysis module, used to perform risk analysis on the risk text data of this type in turn based on the risk analysis prompt words on each analysis dimension through the LLM large language model-A, and output the risk analysis results on the corresponding analysis dimensions; The risk labeling storage module is used to label the risk analysis results in each analysis dimension with corresponding type labels, obtain and save the risk label analysis data set.

[0012] On the other hand, an electronic device is provided, comprising: a processor; a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the methods for risk data analysis based on the LLM large language model is implemented.

[0013] On the other hand, a computer-readable storage medium is provided, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement any of the above-mentioned methods for risk data analysis based on the LLM large language model.

[0014] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least: The present invention performs risk data analysis based on the LLM large language model, and can use the LLM large language model to perform data identification, classification and risk feature extraction throughout the process. By training and learning the risk feature vector of the LLM large language model, the defect of requiring a large number of training samples can be reduced; at the same time, the LLM large language model prompt words in various dimensions are constructed and input on an LLM large language model, and the LLM large language model can continue to further perform risk analysis on the classification and recognition results in various dimensions, and quickly and deeply mine and output the risk analysis results of this type of risk data in various dimensions. Therefore, the risk data can be intelligently processed by the LLM large language model throughout the process, and risk type and various dimensional analysis can be realized through one model, greatly improving the efficiency of risk data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0016] Figure 1 It is a flow chart of a method for risk data analysis based on the LLM large language model provided by an embodiment of the present invention; Figure 2 is a schematic diagram of model training provided by an embodiment of the present invention; Figure 3 is a schematic diagram of risk data analysis based on prompt words provided by an embodiment of the present invention; Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0018] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0019] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0020] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are consistent.

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0022] The LLM large language model used in this embodiment can be an LLM large language model developed by the user based on GPT-3 / 4, such as a pre-trained large language model, such as GPT-3, GPT-4, etc. These models have been pre-trained on large-scale text data and have powerful language understanding and generation capabilities. Or an LLM large language model (such as Wenxin Yiyan) called by a third party through the system background API. This embodiment does not limit it.

[0023] The embodiment of the present invention provides a method for risk data analysis based on the LLM large language model, which can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart of the method for risk data analysis based on the LLM large language model is shown. The processing flow of the method may include the following steps: S1. Construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate LLM large language model-A for risk identification and classification; S2. Construct risk analysis rules and risk keywords on several analysis dimensions, and combine the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and input them into the LLM large language model-A in sequence; S3. Monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data; S4, using the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially perform risk analysis on the risk text data of this type, and output the risk analysis results on the corresponding analysis dimensions; S5. Add corresponding type labels to the risk analysis results in each analysis dimension, obtain a risk label analysis data set, and save it.

[0024] ‌LLM large language model is a trained application model. After feature matching learning, it can be put into use for risk data analysis‌: After the preprocessed risk data is input into the LLM large language model, the LLM large language model can conduct in-depth analysis of the risk data based on the defined risk keywords and analysis rules. It can identify the occurrence of keywords, understand the semantics and sentiment tendencies of the text, and extract key risk information related to the risk.

[0025] Further risk location and classification can be performed later: based on the analysis results of the LLM large language model, key risk information can be located. This can be achieved by identifying the clustering of keywords, the distribution of sentiment tendencies, etc., so as to determine the hot spots and focus of risks. The located risks can be classified in various dimensions, such as positive risks, negative risks, neutral risks, etc., in order to better understand and deal with different types of risks.

[0026] In the present invention, risk data is classified and analyzed based on the training features and prompt words on the same LLM large language model, thereby reducing the complexity of model application. Specifically: First, a risk feature vector is constructed based on risk big data and input into the preset LLM large language model for model training to generate LLM large language model-A for risk identification and classification. This enables LLM large language model-A to identify the text vector of risk data and the risk type of the corresponding risk data and the corresponding type of risk text data (the text vector features of the identified risk data can be converted and output as the corresponding risk text data).

[0027] Secondly, construct risk analysis rules and risk keywords on several analysis dimensions, and combine the risk analysis rules and the risk keywords to construct risk analysis prompt words on the corresponding analysis dimensions and input them into the LLM large language model-A in sequence. The LLM large language model-A can perform risk analysis on the risk text data of this type in sequence based on the risk analysis prompt words on each analysis dimension, and output the risk analysis results on the corresponding analysis dimensions. Therefore, based on the prompt words, the LLM large language model-A can perform online analysis on the risk text data previously classified, identified and output, and based on the keywords and analysis logic in the prompt words, intelligently and automatically analyze the risk text data, and output the corresponding risk analysis results.

[0028] Finally, the analysis results in each dimension are saved to obtain risk analysis data in different analysis dimensions, so that users can grasp the development trend and status of risk data from different dimensions.

[0029] Therefore, the advantages of the present invention of simultaneously performing feature training application on a large LLM language model and performing model data analysis based on prompt words mainly include the following points: 1. Unified model architecture and knowledge base: Performing feature training applications and data analysis on the same LLM large language model can ensure the consistency of the model architecture and knowledge base. This helps the model maintain consistency when understanding and processing data and improves the accuracy of analysis. 2. Efficient feature learning and expression capabilities: The LLM large language model has learned a wealth of language rules and knowledge information through training on a large-scale corpus. In feature training applications, the model can efficiently extract text features and provide strong support for data analysis. At the same time, data analysis based on prompt words can make full use of the model's expressive power to generate fluent and natural text output and improve analysis results. 3. Flexible adaptability and generalization ability: The LLM large language model has strong adaptability and generalization capabilities. In feature training applications, the model can adjust parameters according to specific tasks to adapt to different data features and scenario requirements. At the same time, data analysis based on prompt words can also flexibly respond to various complex analysis tasks and provide personalized solutions. 4. Reduce development and maintenance costs: Using the same LLM large language model for feature training applications and data analysis can avoid repeated development of multiple models and reduce development and maintenance costs. At the same time, the unified model architecture also helps to simplify the data processing process and improve analysis efficiency. 5. Promote knowledge sharing and collaborative work: Working on the same LLM large language model can promote knowledge sharing and collaborative work. The learning results in feature training applications can be directly applied to data analysis tasks, and the results of data analysis can also provide feedback and guidance for feature training, achieving mutual promotion and common improvement of models and applications.

[0030] The application principle of the LLM large language model (abbreviated as "LLM" or "large model" or "model") can be further applied and understood in combination with its development background and principles. like Figure 2As shown, preferably, the S1, constructing a risk feature vector based on risk big data and inputting it into a preset LLM large language model for model training, generating an LLM large language model-A for risk identification and classification, includes: Collecting the risk big data consisting of different types of risk data sets; Perform text processing on each risk data in the risk big data based on NLP technology to generate corresponding risk NLP data; Using a word embedding method, converting the risk NLP data into a text vector to obtain a corresponding risk feature vector; After the conversion is completed, a feature set constructed by a number of risk feature vectors is statistically obtained; Dividing the feature set into a training set and a validation set according to a preset ratio; Input the training set into a preset LLM large language model, and let the LLM large language model perform model training on each of the risk feature vectors in the training set to generate the LLM large language model-A; The recognition performance of the LLM large language model-A is verified using the verification set: If the verification is successful, the LLM large language model-A is deployed to the risk supervision system and put into use after parameter adjustment; If the verification fails, the above training steps are repeated to rebuild the LLM large language model-A.

[0031] The way to collect risk big data can be: Collect risk data from multiple platforms, which can include posts on social media, news reports, forum discussions, etc. Obtain a large amount of risk data from social media, news websites, forums and other platforms through crawler technology or API interface. Preprocess the collected data, including removing noise data, segmentation, removing stop words and normalization to convert text data into a structured data format, laying the foundation for subsequent analysis. Preprocess the collected data, including cleaning data, removing noise, unifying formats, etc., to ensure the accuracy and consistency of the data.

[0032] In order to let the model learn the text features of different types of risk data, this paper adopts the text vector conversion method to extract and convert the risk feature vectors in the risk data, so as to construct the feature set for model training. Subsequently, the model is allowed to learn the features of the vector features (the various text features represented by the risk feature vectors) in the divided training set, so as to train and generate an optimized model that can identify different feature vectors: LLM Large Language Model-A. At this time, the trained LLM Large Language Model-A can complete the recognition of the corresponding type of risk text features, so it can be used for risk supervision and automatically classify risk data. Different types of risk data sets, such as personal user risk data, collective risk data, official risk data, etc. in social news risks, or risk data on various formats and time periods, etc., can be specifically classified by users / enterprises / institutions.

[0033] If the classification dimension is the risk development trend or status type, you can further collect corresponding risk data sets from sentiment classification results, topic distribution and other types.

[0034] Therefore, risk big data mainly limits the monitoring direction of the model on risk data from a horizontal perspective (that is, from the general direction of risk data types). The following direction of risk analysis based on prompt words in various dimensions is to let the model conduct internal analysis and disassembly of risk data, and conduct in-depth mining and analysis of various types of risk data from a vertical perspective. From various dimensions such as risk overview, keyword frequency, emotional tendency distribution, hot topics, etc., the risk data is instantiated and analyzed in text semantics, so as to grasp the development nature of risks and study how risks are formed, how they develop and how they are distributed.

[0035] Natural language processing (NLP) techniques, such as word embedding or sentence embedding, are used to convert preprocessed text data into vector representations in a high-dimensional vector space. These vectors can capture the semantic information in the text, making similar texts closer in the vector space, which is convenient for subsequent analysis and classification.

[0036] Select a suitable large language model architecture, such as Transformer, and train the model based on the extracted risk vectors; during the training process, adjust the model parameters so that the model can accurately understand and generate text data related to the risk. After the training is completed, the large language model will have the ability to automatically identify, classify and analyze risk data. Deploy the trained large language model to the risk supervision system to monitor and analyze new risk data in real time or regularly. Through the sentiment classification results, topic distribution and other information output by the model, hot topics, sentiment tendencies and potential risks in the risk can be discovered in a timely manner.

[0037] The method for processing risk big data based on NLP technology and generating risk feature vectors may include the following steps: ‌Risk big data collection and preliminary processing‌: First, collect a large amount of risk data from various sources (such as social media, news sites, forums, etc.).

[0038] The collected data is preliminarily processed, including removing irrelevant data, duplicate data, and obvious noise to ensure the purity and validity of the data.

[0039] ‌Text processing and NLP technology application‌: Use natural language processing (NLP) technology to perform in-depth text processing on the risk data after preliminary processing. This includes basic NLP tasks such as word segmentation, part-of-speech tagging, named entity recognition, syntactic analysis, as well as advanced NLP tasks such as sentiment analysis and topic identification.

[0040] Through NLP technology, risk data is converted into structured NLP data to facilitate subsequent analysis and processing.

[0041] ‌Word embedding model selection and training‌: Select an appropriate word embedding model, such as Word2Vec, GloVe, FastText, or BERT, and train the model based on the characteristics of the risk data and the needs of subsequent analysis.

[0042] If the risk data is large in scale and contains rich contextual information, you can consider using a pre-trained word embedding model and fine-tune it on this basis to better suit specific risk analysis tasks.

[0043] ‌Text vector conversion and risk feature vector generation‌: Using the trained word embedding model, each word or phrase in the risk NLP data is converted into a corresponding word vector.

[0044] For the entire risk text, a risk feature vector can be generated by averaging word vectors, weighted averaging, or using a specific text representation method (such as sentence embedding). This feature vector can capture the semantic information and key features in the risk text.

[0045] ‌Application of risk feature vector‌: The generated risk feature vector can be used for a variety of risk analysis tasks, such as risk classification, sentiment analysis, topic identification, cluster analysis, etc.

[0046] By calculating the similarity or distance between different risk feature vectors, it is also possible to compare and analyze risk data, helping users better understand the evolution and trends of risks.

[0047] For model training and verification, you can refer to the training and verification methods of traditional AI models, such as: ‌Feature set partitioning‌: First, the processed and generated feature set (i.e., risk feature vector set) is divided into a training set and a validation set according to a preset ratio (e.g., 8:2). This ratio can be adjusted according to actual conditions, but it is generally recommended that the training set accounts for a larger proportion so that the model can fully learn; the validation set accounts for a smaller proportion to evaluate the performance of the model.

[0048] ‌Model Training‌: Input the divided training set into the preset LLM (Large Language Model). Here, LLM can be a pre-trained large language model, such as BERT, GPT, etc., or a custom-built model.

[0049] Let the LLM large language model train the risk feature vectors in the training set. During the training process, the model will learn how to predict or generate corresponding outputs (such as sentiment classification, topic identification, etc.) based on the input risk feature vectors.

[0050] Through continuous iterative training, the model will gradually adjust its parameters so that its performance on the training set gets better and better, that is, it can more accurately predict or generate outputs corresponding to the input risk feature vector.

[0051] Generate LLM large language model-A: After sufficient training, an LLM large language model optimized for risk feature vectors was obtained, called LLM large language model-A.

[0052] ‌Identification performance verification‌: Use the previously divided validation set to verify the recognition performance of the LLM large language model-A. Input the risk feature vector in the validation set into the model and observe the prediction or generation results output by the model.

[0053] By comparing the model output with the true label or expected output in the validation set, the recognition performance of the model can be evaluated. Common evaluation indicators include accuracy, recall, F1 score, etc. This is verified by the administrator.

[0054] If the model performs well on the validation set, that is, all evaluation indicators reach the expected level, then it can be considered that the LLM Large Language Model-A has good recognition performance and can be used in actual risk analysis tasks.

[0055] If the model performs poorly on the validation set, you may need to analyze the reasons and further optimize the model. This may include adjusting the model structure, increasing training data, improving feature extraction methods, etc.

[0056] After the model performance reaches the expected level, the LLM Large Language Model-A can be deployed to the actual risk analysis system (the application system deployed on the user's backend server) to automate and efficiently process and analyze new risk data.

[0057] Through the above steps, the feature set can be divided into a training set and a validation set, and the LLM large language model-A for risk feature vectors can be trained, and its recognition performance can be verified to ensure the reliability and effectiveness of the model in practical applications.

[0058] like Figure 3 As shown, preferably, the S2, constructing risk analysis rules and risk keywords on several analysis dimensions, and combining the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and sequentially inputting them into the LLM large language model-A, includes: Collect user data analysis requirements on risk data in different analysis dimensions from the front end; The risk analysis rules and risk keywords in the data analysis requirements are extracted using the LLM large language model, and the risk analysis prompt words corresponding to the analysis dimension are automatically constructed; Deploy a distributed file system HDFS in the background to store the risk analysis results on each analysis dimension obtained in step S4; The risk analysis prompt words on each of the analysis dimensions are sent to the management node (Name Node) of the distributed file system HDFS, and the management node (Name Node) allocates corresponding storage nodes (Data Node) to the risk analysis prompt words on each of the analysis dimensions, and configures corresponding attribute values ​​containing the analysis dimensions at the storage nodes (Data Node).

[0059] In this embodiment, the data monitoring and feedback between the front-end and the back-end can be understood in combination with the existing architecture of the front-end and back-end systems. Specifically, the data analysis requirements for each dimension input by the front-end (prompt words for data extraction can also be constructed), the model analyzes and extracts the risk analysis rules and risk keywords for the corresponding dimensions, and constructs prompt words for the corresponding dimensions in combination with the rules and keywords. The prompt words for each dimension can be as follows: 1. Conventional dimensions: The dimensions of risk analysis that may be performed on risk data mainly include time dimension, space dimension, emotion dimension, theme dimension, user dimension, content dimension and channel dimension. Risk analysis prompt words for corresponding dimensions are constructed respectively: ‌Time dimension‌: Observe the changes in the discussion popularity of a specific topic or keyword over a period of time, identify peak and trough periods, and conduct periodic research and sudden event tracking. This dimension helps to grasp the development trend and timeliness of risks.

[0060] ‌Spatial dimension‌: Using IP addresses or user-reported location information, we can draw a risk map to show the differences in attention paid to the topic in different regions and make cross-regional comparisons. This helps us understand the geographical distribution and regional characteristics of risks.

[0061] ‌Emotional dimension‌: Classify text into positive, negative or neutral emotion categories, quantify the overall public opinion atmosphere, and can also measure emotion intensity and track emotion fluctuations. This dimension can reveal the public's emotional attitudes and emotional changes towards specific events or topics.

[0062] ‌Topic dimension‌: Automatically extract the words that best represent the core of the current discussion from massive texts, build a topic model, and identify hot topics. This helps to grasp the core issues and discussion focus of risks. ‌User dimension‌: key person positioning, community division, and interaction pattern analysis‌. Through this dimension, we can understand the characteristics and behavior patterns of different user groups, as well as their role in risk communication.

[0063] ‌Content dimension‌: distinguish between various forms of content such as text, pictures, and videos, evaluate the effectiveness of different types of materials in dissemination, and conduct quality assessments and statistics on the ratio of original to reprinted content. This helps to understand the diversity and innovation of risky content.

[0064] ‌Channel dimension‌: examine the characteristics of different social media platforms, conduct cross-validation and channel preference analysis‌. This dimension helps to formulate targeted risk communication strategies and improve the effectiveness of risk communication.

[0065] 2: In-depth analysis dimensions: Analyze from the following in-depth analysis dimensions and construct corresponding in-depth analysis prompt words. The in-depth analysis dimensions include relationship, damage, complexity, continuity and relevance: ‌Relationship‌: Determine the relationship between the incident and the unit, determine whether it is the main responsible department and the difficulty of response. This helps to clarify the responsible entity and response strategy for risk response.

[0066] ‌Degree of harm‌: Assess the magnitude of the negative effects of the risk, including the degree of questioning and accountability‌. This dimension helps to understand the potential impact of the risk on an organization or individual and develop appropriate response measures.

[0067] Complexity: Sort out the context of the incident, identify the main contradictions, and quickly determine the information coordination department to summarize the situation and judge the online information. This helps to grasp the complexity and difficulty of the risk and formulate an effective treatment plan.

[0068] ‌Persistence‌: Considers the process required to handle the event itself and the duration of risk attention. This dimension helps predict the duration and development trend of the risk and provides guidance for long-term response.

[0069] ‌Relevance‌: Analyze whether the incident will make the public associate or compare it with similar situations in the past. This helps to understand the chain reactions and related impacts that the risk may cause, and provide a basis for comprehensive response.

[0070] Conventional risk analysis and in-depth analysis are complementary and interrelated. The risk analysis dimension provides a basic classification and sorting of risk data, providing a basis and foundation for in-depth analysis; while the in-depth analysis dimension conducts a deeper and more comprehensive analysis and mining on this basis, revealing the inherent laws and potential risks of risk data.

[0071] Therefore, the risk analysis prompt words corresponding to the analysis dimensions constructed can enable the model to perform risk analysis based on the above dimensions and obtain richer and more comprehensive risk analysis results.

[0072] In order to manage and save the risk analysis results in various dimensions, the present invention adopts distributed management and storage technology. A distributed file system HDFS is deployed in the background to store the risk analysis results in various analysis dimensions.

[0073] The distributed file system HDFS has a management node (Name Node) and several storage nodes (Data Node). The management node (Name Node) can manage each storage node (Data Node), for example, assigning corresponding storage nodes (Data Node) and corresponding storage addresses and spaces to the risk analysis prompt words on each analysis dimension, and can also store the analysis results of each node prompt words output by the subsequent model accordingly, so as to facilitate risk data classification management and preservation. After the management node (Name Node) assigns corresponding storage nodes (Data Node) to the risk analysis prompt words on each analysis dimension, it is necessary to configure the corresponding attribute values ​​(also types, so as to facilitate the storage of risk text data output by the model identification and prediction) containing the analysis dimension at the storage node (Data Node). Through these attribute values, it is convenient to locate and manage (call or write) the data belonging to each node, and to store the risk analysis results on each dimension in a distributed manner, so as to facilitate risk analysis and viewing, and avoid the confusion of risk supervision management caused by centralized storage.

[0074] The application principle of the distributed file system HDFS will not be elaborated here.

[0075] Preferably, the step S3, monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data, includes: activating a risk monitoring system and monitoring risk data in real time through the risk monitoring system; After the corresponding risk data is monitored, the risk data is collected and sent to the pre-deployed LLM large language model-A, and the LLM large language model-A monitors the risk feature vector of the risk data and outputs the corresponding risk type, and at the same time parses and outputs the risk text data corresponding to the risk feature vector; The risk text data is cached in a background database.

[0076] The risk monitoring system can capture risk data in business data by referring to the risk data capture and supervision rules configured in the background. Therefore, after the system is activated, the system can monitor risk data in real time, and import the captured risk data into the model trained by the present invention. The model first identifies the type of current risk data based on the training characteristics, and then caches the risk data in the background. The subsequent model will further perform risk analysis on the risk data. After the analysis, the analysis results are stored in the corresponding storage node. After the model identifies the risk type of the risk data, the risk feature vector corresponding to the risk data can also be output at the same time. By parsing the risk feature vector of the risk data, the corresponding risk text data can be generated. Therefore, the model can be used to quickly generate the risk feature vector of the currently monitored risk data, and the model parses and outputs the risk text data of the current risk data for the next step of analysis (the subsequent model completes the risk analysis on each dimension and deletes the background cache data).

[0077] Preferably, the S4, through the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially performs risk analysis on the risk text data of this type, and outputs the risk analysis results on the corresponding analysis dimensions, including: Sending a risk analysis request to the management node (Name Node); The management node (Name Node) responds to the risk analysis request, sequentially retrieves the risk analysis prompt words configured at each storage node (Data Node) and feeds them back to the LLM large language model-A; Based on the risk analysis prompt words, the risk text data is sequentially analyzed by the LLM large language model-A, and the risk analysis results on the corresponding analysis dimensions are output; The risk analysis result is fed back to the management node (Name Node), and the management node (NameNode) stores the risk analysis result in the corresponding storage node (Data Node).

[0078] In the model, the model can further analyze the risk text data based on the identified and parsed output. The model can request prompt words on each dimension from HDFS in turn, and the management node retrieves the prompt words configured at each storage node and feeds them back to the model. The model performs risk analysis on the risk text data of the current risk data based on the prompt words configured at each storage node.

[0079] The model can automatically and intelligently analyze the risk text data based on the prompt words, and output the analysis results that match the rules and keywords of the corresponding dimensions in the prompt words. According to the defined risk keywords and analysis rules, the LLM large language model is guided to conduct in-depth analysis of risk data. The model can identify the occurrence of keywords, understand the semantics and sentiment of the text, and extract key risk information related to the risk. The analysis rules and keywords on the corresponding dimensions are determined by the previous construction configuration.

[0080] The risk analysis is performed through the LLM large language model-A and the results on the corresponding dimensions are output. The risk analysis process can be referred to as follows: 1. Risk analysis based on prompt words The LLM Large Language Model-A analyzes the risk text data one by one based on the risk analysis prompt words input. It identifies the content related to the prompt words in the text and interprets and classifies them based on the knowledge and patterns learned during its training process.

[0081] 2. Output risk analysis results For each risk analysis dimension, LLM Large Language Model-A will output the corresponding risk analysis results. These results can be quantitative (such as the proportion of positive emotions in the emotional dimension, the list of hot topics in the subject dimension, etc.), or qualitative (such as the positioning of key figures in the user dimension, the evaluation of the original and reprinted ratio in the content dimension, etc.).

[0082] 3. Integrate and present analysis results Finally, the risk analysis results of each dimension are integrated to form a comprehensive risk analysis report. This report can include detailed analysis results of each dimension, as well as comprehensive judgments and recommendations based on these results. The report can be in the form of text, charts, dashboards, etc., to facilitate user understanding and use.

[0083] Suppose there is a risky text data: "After a company's new product was released, a large number of positive reviews appeared on social media, but some users also expressed dissatisfaction with certain features of the new product." ‌Time dimension‌: LLM Large Language Model-A may output the time point or time period when the risk text data appears.

[0084] ‌Emotional dimension‌: The model may identify two emotions: "positive evaluation" and "dissatisfaction", and give the proportion of positive and negative emotions.

[0085] ‌Topic Dimension‌: The model might identify “new product launch” as the topic of this text.

[0086] ‌User dimension‌: The model might try to identify groups of users who express positive and negative sentiments.

[0087] ‌Content dimension‌: The model may evaluate whether the text is original or reposted, as well as the quality of its content.

[0088] ‌Channel dimension‌: If the text data contains source information, the model may be able to identify on which social media platform the text was posted.

[0089] In summary, through the LLM large language model-A and risk analysis prompt words, a comprehensive and multi-dimensional analysis of risk text data can be performed, and the risk analysis results on the corresponding dimensions can be output. The risk analysis results output by the model for the prompt words on each analysis dimension can be returned to the management node by the model and stored in the storage node corresponding to the dimension data according to its attributes.

[0090] Therefore, each analysis result can be saved according to the conditions of dimensional analysis, which is convenient for users to distribute and view risk supervision and analysis results in various dimensions, and manage risk monitoring data in an orderly manner.

[0091] S5. Add corresponding type labels to the risk analysis results in each analysis dimension, obtain a risk label analysis data set, and save it.

[0092] By adding corresponding type labels, risk analysis results of the same type but different analysis dimensions can be displayed in the same way. In this way, even in different storage nodes, it is possible to distinguish which risk results are which types of risk data based on labels, which is convenient for users to obtain the result type and source when analyzing.

[0093] On the other hand, a device for performing risk data analysis based on an LLM large language model is provided, wherein the device for performing risk data analysis based on an LLM large language model is used to implement the method for performing risk data analysis based on an LLM large language model, and the device comprises: The large language model training module is used to construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate the LLM large language model-A for risk identification and classification; A prompt word construction module is used to construct risk analysis rules and risk keywords on several analysis dimensions, and to construct risk analysis prompt words on corresponding analysis dimensions in combination with the risk analysis rules and the risk keywords, and to input the prompt words into the LLM large language model-A in sequence; A risk monitoring module, used to monitor risk data through the LLM large language model-A, identify and classify the risk type of the monitored risk data and the corresponding type of risk text data; A risk analysis module, used to perform risk analysis on the risk text data of this type in turn based on the risk analysis prompt words on each analysis dimension through the LLM large language model-A, and output the risk analysis results on the corresponding analysis dimensions; The risk labeling storage module is used to label the risk analysis results in each analysis dimension with corresponding type labels, obtain and save the risk label analysis data set.

[0094] Please understand the functions and interactions of the above modules in conjunction with the corresponding steps in the above solution. They will not be described here.

[0095] Figure 4 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 4 As shown, the electronic device may include the above-mentioned device for performing risk data analysis based on the LLM large language model. Optionally, the electronic device 410 may include a first processor 2001 .

[0096] Optionally, the electronic device 410 may further include a memory 2002 and a transceiver 2003 .

[0097] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0098] Combine the following Figure 4 The components of the electronic device 410 are described in detail: The first processor 2001 is the control center of the electronic device 410, and may be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPUs), or may be application specific integrated circuits (ASICs), or may be one or more integrated circuits configured to implement the embodiments of the present invention, such as one or more microprocessors (digital signal processors, DSPs), or one or more field programmable gate arrays (field programmable gate arrays, FPGAs).

[0099] Optionally, the first processor 2001 can execute various functions of the electronic device 410 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0100] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 4 CPU0 and CPU1 are shown in FIG.

[0101] In a specific implementation, as an embodiment, the electronic device 410 may also include multiple processors, such as Figure 4 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0102] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0103] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently and access the first processor 2001 through the interface circuit ( Figure 4 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0104] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0105] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 4 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0106] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 4 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0107] It should be noted that Figure 4 The structure of the electronic device 410 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure recognition device may include more or fewer components than those shown in the figure, or combine certain components, or arrange the components differently.

[0108] In addition, the technical effects of the electronic device 410 can refer to the technical effects of the method for risk data analysis based on the LLM large language model described in the above method embodiment, and will not be repeated here.

[0109] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0110] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0111] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0112] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0113] In the present invention, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0114] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0115] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0117] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0118] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0119] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0120] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0121] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for risk data analysis based on the LLM large language model, characterized in that: The method comprises: S1. Construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate LLM large language model-A for risk identification and classification; S2. Construct risk analysis rules and risk keywords on several analysis dimensions, and combine the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and input them into the LLM large language model-A in sequence; S3. Monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data; S4, using the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially perform risk analysis on the risk text data of this type, and output the risk analysis results on the corresponding analysis dimensions; S5. Add corresponding type labels to the risk analysis results in each analysis dimension, obtain a risk label analysis data set, and save it.

2. The method for risk data analysis based on the LLM large language model according to claim 1 is characterized in that: S1, constructing a risk feature vector based on risk big data and inputting it into a preset LLM large language model for model training, generating an LLM large language model-A for risk identification and classification, including: Collecting the risk big data consisting of different types of risk data sets; Perform text processing on each risk data in the risk big data based on NLP technology to generate corresponding risk NLP data; Using a word embedding method, converting the risk NLP data into a text vector to obtain a corresponding risk feature vector; After the conversion is completed, a feature set constructed by a number of risk feature vectors is statistically obtained; Dividing the feature set into a training set and a validation set according to a preset ratio; Input the training set into a preset LLM large language model, and let the LLM large language model perform model training on each of the risk feature vectors in the training set to generate the LLM large language model-A; The recognition performance of the LLM large language model-A is verified using the verification set: If the verification is successful, the LLM large language model-A is deployed to the risk supervision system and put into use after parameter adjustment; If the verification fails, the above training steps are repeated to rebuild the LLM large language model-A.

3. The method for risk data analysis based on the LLM large language model according to claim 1 is characterized in that: S2, constructing risk analysis rules and risk keywords on several analysis dimensions, and combining the risk analysis rules and the risk keywords to construct risk analysis prompt words on corresponding analysis dimensions and input them into the LLM large language model-A in sequence, including: Collect user data analysis requirements on risk data in different analysis dimensions from the front end; The risk analysis rules and risk keywords in the data analysis requirements are extracted using the LLM large language model, and the risk analysis prompt words corresponding to the analysis dimension are automatically constructed; Deploy a distributed file system HDFS in the background to store the risk analysis results on each analysis dimension obtained in step S4; The risk analysis prompt words on each of the analysis dimensions are sent to the management node (Name Node) of the distributed file system HDFS, and the management node (Name Node) allocates corresponding storage nodes (Data Node) to the risk analysis prompt words on each of the analysis dimensions, and configures corresponding attribute values ​​containing the analysis dimensions at the storage nodes (Data Node).

4. The method for risk data analysis based on the LLM large language model according to claim 3 is characterized in that: The step S3, monitoring risk data through the LLM large language model-A, identifying and classifying the risk type of the monitored risk data and the corresponding type of risk text data, includes: activating a risk monitoring system and monitoring risk data in real time through the risk monitoring system; After the corresponding risk data is monitored, the risk data is collected and sent to the pre-deployed LLM large language model-A, and the LLM large language model-A monitors the risk feature vector of the risk data and outputs the corresponding risk type, and at the same time parses and outputs the risk text data corresponding to the risk feature vector; The risk text data is cached in a background database.

5. The method for risk data analysis based on the LLM large language model according to claim 4 is characterized in that: S4, using the LLM large language model-A, based on the risk analysis prompt words on each analysis dimension, sequentially performs risk analysis on the risk text data of this type, and outputs the risk analysis results on the corresponding analysis dimensions, including: Sending a risk analysis request to the management node (Name Node); The management node (Name Node) responds to the risk analysis request, sequentially retrieves the risk analysis prompt words configured at each storage node (Data Node) and feeds them back to the LLM large language model-A; The risk analysis is performed on the risk text data in sequence based on the risk analysis prompt words through the LLM large language model-A, and the risk analysis results on the corresponding analysis dimensions are output; The risk analysis result is fed back to the management node (Name Node), and the management node (Name Node) stores the risk analysis result in the corresponding storage node (Data Node).

6. A device for risk data analysis based on an LLM large language model, the device for risk data analysis based on an LLM large language model is used to implement the method for risk data analysis based on an LLM large language model as claimed in any one of claims 1 to 5, characterized in that: The device comprises: The large language model training module is used to construct risk feature vectors based on risk big data and input them into the preset LLM large language model for model training to generate the LLM large language model-A for risk identification and classification; A prompt word construction module is used to construct risk analysis rules and risk keywords on several analysis dimensions, and to construct risk analysis prompt words on corresponding analysis dimensions in combination with the risk analysis rules and the risk keywords, and to input the prompt words into the LLM large language model-A in sequence; A risk monitoring module, used to monitor risk data through the LLM large language model-A, identify and classify the risk type of the monitored risk data and the corresponding type of risk text data; A risk analysis module, used to perform risk analysis on the risk text data of this type in turn based on the risk analysis prompt words on each analysis dimension through the LLM large language model-A, and output the risk analysis results on the corresponding analysis dimensions; The risk labeling storage module is used to label the risk analysis results in each analysis dimension with corresponding type labels, obtain and save the risk label analysis data set.

7. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program codes, which can be called by a processor to execute the method according to any one of claims 1 to 5.

Images