Clustering model-based event attribution method and device, equipment and storage medium
By adopting a clustering model-based event attribution method in the analysis of financial regulatory punishments, and using large language models and vectorized processing technology, the problems of low analysis efficiency and insufficient accuracy in the existing technology are solved, and more efficient and accurate cluster analysis of financial punishments are achieved.
Patent Information
- Application Number
- CN202411777804.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-03
- Publication Date
- 2025-03-28
AI Technical Summary
The prior art has problems of inefficient efficiency and insufficient accuracy in data analysis of financial regulatory penalties, especially when dealing with data diversity and long-tail reasons, resulting in inaccurate clustering results.
The event attribution method based on the clustering model is adopted, and event attribution data is processed through the large language model, key information and conceptual data are extracted, cluster analysis is performed after vector processing, clustering is calculated, clustering particle size parameters are calculated, and the output of the large language model and hierarchical clustering results are combined, and vector model and clustering model are optimized to improve analysis accuracy.
It improves the efficiency and accuracy of event attribution analysis, can better handle the diversity and long-tail reasons for financial punishment, and provide more accurate clustering results.
Smart Images
Figure CN119721223A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to an event attribution method, device, electronic device and storage medium based on a clustering model. Background Art
[0002] In the financial industry, the regulatory authorities' punishment of financial institutions' violations is an important means to maintain market order and protect consumer rights. With the rapid development of the financial market and the continuous emergence of financial innovations, regulatory authorities have continuously increased their supervision of financial institutions, and the complexity and diversity of penalty decisions have also increased. However, the existing technical analysis methods for financial regulatory penalty reasons have certain limitations in dealing with these complexities and diversity. For example, the data form diversity of financial penalty reasons and the difficulty in handling long-tail reasons. Penalty decisions usually contain multiple penalty reasons, and the granularity of the reasons varies. There are detailed descriptions of specific violation details and brief summaries, which makes it difficult for existing technologies to unify standards when processing data. Especially for those long-tail reasons that appear infrequently but are important, existing technologies often cannot effectively identify and process them, resulting in inaccurate clustering results. For another example, it is difficult to determine the clustering granularity of financial penalty reasons. When clustering penalty reasons, it is difficult for existing technologies to accurately determine the clustering granularity. Too fine clustering may lead to information redundancy, and too coarse clustering may lose key information, affecting the accuracy and effectiveness of the analysis. Summary of the invention
[0003] The main technical problem to be solved by the implementation methods of the present application is that the analysis efficiency and accuracy of the analysis results obtained by the prior art event attribution analysis need to be improved.
[0004] In order to solve the above technical problems, the first technical solution adopted by the implementation mode of the present application is: to provide an event attribution method based on a clustering model, including: obtaining event attribution data, processing the event attribution data through a preset large language model, and obtaining attribution cause key information data and attribution cause concept data; vectorizing the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector; calculating clustering granularity parameters, using a preset clustering model to cluster the event attribution feature vector according to the clustering granularity parameters to obtain an event attribution clustering result; inputting the event attribution clustering result into the preset large language model to obtain an event attribution classification label output by the large language model.
[0005] Optionally, the step of processing the event attribution data through a preset large language model to obtain attribution cause key information data and attribution cause concept data includes: splitting the event attribution data into sub-cause data through a preset text processing tool; processing the sub-cause data using a preset natural language processing model, extracting information features in the sub-cause data, and obtaining the attribution cause key information data; processing the attribution cause key information data through a regression prompt module of the preset large language model to obtain the attribution cause concept data.
[0006] Optionally, the clustering model is a hierarchical clustering model, and the step of calculating the clustering granularity parameter includes: obtaining a first sample pair consisting of two nodes of each merging operation during the historical hierarchical clustering processing of the hierarchical clustering model; processing the first sample pair through the semantic parsing module of the preset large language model to obtain the node clustering analysis result of the first sample pair; using the clustering level data of the first sample pair and the node clustering analysis result to calculate the precision and recall rate, as well as the weighted harmonic mean of the precision and the recall rate; and selecting the corresponding clustering level as the clustering granularity parameter according to the weighted harmonic mean and a preset consistency threshold range.
[0007] Optionally, before the step of inputting the event attribution clustering results into the preset large language model, it also includes: calculating the cosine distance between each event attribution and each clustering category in the event attribution clustering results to obtain the classification probability of each event attribution belonging to each clustering category; setting at least one clustering category corresponding to each event attribution according to a preset classification probability threshold range; normalizing the classification probability, and using the normalized classification probability to calculate the entropy value of each event attribution; removing the event attribution with the largest entropy value in each clustering category in the event attribution clustering results.
[0008] Optionally, after the step of using the normalized classification probability to calculate the entropy value of each event attribution, the method further comprises: selecting the cluster category with the highest entropy value as the anchor category from the event attribution clustering results; calculating the similarity distance between the cluster categories, and selecting the cluster category with the smallest similarity distance to the anchor category as the cluster category to be judged; randomly selecting one event attribution from each of the cluster categories to be judged to form a fine-tuning event attribution data pair; analyzing and judging the event attribution with the highest similarity between the fine-tuning event attribution data pair and the anchor category through the preset large language model, and updating the result of the analysis and judgment to the fine-tuning event attribution data pair; using the anchor category and the fine-tuning event attribution data pair to generate a hard negative example triple; using the hard negative example triple to perform comparative learning, and adjusting the parameters of the vector model according to the result of the comparative learning.
[0009] Optionally, after the step of obtaining the event attribution classification label output by the large language model, it includes: sending the event attribution data and the event attribution classification label to a manual business expert, receiving adjustment result data from the manual business expert; updating the event attribution classification label according to the adjustment result data, and optimizing the vector model and the clustering model.
[0010] Optionally, the step of vectorizing the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector includes: extracting the penalty cause key feature vector and the penalty cause probability feature vector corresponding to the attribution cause key information data and the attribution cause concept data respectively through a preset vector model; performing vector splicing processing on the penalty cause key feature vector and the penalty cause probability feature vector to obtain the event attribution feature vector.
[0011] In order to solve the above technical problems, the second technical solution adopted by the implementation method of the present application is: to provide an event attribution device based on a clustering model, including: a data preprocessing module, used to obtain event attribution data, and process the event attribution data through a preset large language model to obtain attribution cause key information data and attribution cause concept data; a data vectorization module, used to vectorize the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector; a data clustering module, used to calculate clustering granularity parameters, and use a preset clustering model to cluster the event attribution feature vector according to the clustering granularity parameters to obtain an event attribution clustering result; a classification label module, used to input the event attribution clustering result into the preset large language model to obtain the event attribution classification label output by the large language model.
[0012] To solve the above technical problems, the third technical solution adopted in the implementation manner of the present application is: to provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the event attribution method based on the clustering model as described above.
[0013] In order to solve the above technical problems, the fourth technical solution adopted in the implementation mode of the present application is: providing a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by an electronic device, the electronic device executes the event attribution method based on the clustering model as described above.
[0014] Different from the related art, this application unifies the event attribution granularity by splitting and rewriting event attribution, explores the deep meaning of event attribution, provides more feature data for event attribution clustering, and introduces a large language model for synchronous judgment, which is compared with the results of hierarchical clustering to obtain the optimal clustering granularity, thereby improving the accuracy of event attribution based on clustering models. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] One or more embodiments are exemplarily described by corresponding drawings, which do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and the figures in the drawings do not constitute proportional limitations unless otherwise stated.
[0016] Figure 1 It is a schematic diagram of the operating environment of the event attribution method based on the clustering model provided in an embodiment of the present application.
[0017] Figure 2 It is a schematic diagram of the execution flow of the event attribution method based on the clustering model provided in an embodiment of the present application.
[0018] Figure 3 It is a schematic diagram of the execution flow of calculating clustering granularity parameters in the event attribution method based on the clustering model provided in an embodiment of the present application.
[0019] Figure 4 It is a schematic diagram of the execution flow of adjusting clustering categories in the event attribution method based on the clustering model provided in an embodiment of the present application.
[0020] Figure 5 It is a schematic diagram of the system structure of an event attribution device based on a clustering model provided in an embodiment of the present application.
[0021] Figure 6 It is a schematic diagram of the hardware structure of an electronic device that executes an event attribution method based on a clustering model provided in an embodiment of the present application. DETAILED DESCRIPTION
[0022] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0023] It should be noted that, if there is no conflict, the various features in the embodiments of the present application can be combined with each other, all within the scope of protection of the present application. In addition, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flow chart, in some cases, the steps shown or described can be performed in a different order than the module division in the device schematic diagram or the order in the flow chart.
[0024] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.
[0025] To facilitate understanding of this embodiment, firstly, a method for event attribution based on a clustering model disclosed in the embodiment of the present application is introduced in detail. Figure 1 , Figure 1 Schematic diagram of the operating environment of the event attribution method based on the clustering model provided in the embodiment of the present application, such as Figure 1 As shown, the execution subject of the event attribution method based on the clustering model provided in the embodiment of the present application is generally an electronic device with certain computing capabilities, such as a computer device. In some possible implementations, the event attribution method based on the clustering model can be implemented by a processor calling a computer-readable instruction stored in a memory. Figure 1 The computer device in the term can be a server. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. It can be understood that Figure 1 The number of computer devices in the figure is only for reference and can be expanded to any number according to actual needs.
[0026] Please continue reading Figure 2 , Figure 2 is a schematic diagram of the execution flow of the event attribution method based on the clustering model provided in the embodiment of the present application, such as Figure 2 As shown, the following steps are included:
[0027] S1. Obtain event attribution data, process the event attribution data through a preset large language model, and obtain attribution cause key information data and attribution cause concept data.
[0028] Among them, a large language model refers to a model that can predict text sequences obtained by training with large-scale text data based on deep learning technology. These models usually have a huge amount of parameters and a strong learning ability. Large language models usually need to be pre-trained on a large-scale text corpus to learn the language patterns, structures and semantic information in the text data, which can be text data in the financial field in the embodiment of the present application. The typical structure of a large language model is based on a deep neural network, such as a recurrent neural network (RNN), a long short-term memory network (LSTM), a Transformer, etc., which can capture long-term dependencies and semantic information in text data.
[0029] As an optional implementation, the process of obtaining the attribution cause key information data and the attribution cause concept data in the above step S1 may include the following sub-steps S11 to S13.
[0030] S11. Split the event attribution data into sub-cause data using a preset text processing tool.
[0031] S12. Use the preset natural language processing model to process the sub-cause data, extract the information features in the sub-cause data, and obtain the key information data of the attribution cause.
[0032] Among them, the process of natural language processing model processing data usually involves text data input, feature extraction, modeling and output prediction. Through feature extraction, the model can better capture the semantic and grammatical information of text data and improve the model's ability to understand text data. By extracting representative features, the model can better generalize to unseen data and reduce the risk of overfitting.
[0033] S13. Process the key information data of the attribution cause through the regression prompt module of the preset large language model to obtain the conceptual data of the attribution cause.
[0034] S2. The attribution cause key information data and the attribution cause concept data are vectorized through a preset vector model to obtain an event attribution feature vector.
[0035] Specifically, the above step S2 can first extract the key feature vector of the penalty cause and the probability feature vector of the penalty cause corresponding to the key information data of the attribution cause and the concept data of the attribution cause respectively through a preset vector model. Then, the key feature vector of the penalty cause and the probability feature vector of the penalty cause are vector-concatenated to obtain the event attribution feature vector.
[0036] S3. Calculate clustering granularity parameters, use a preset clustering model to cluster the event attribution feature vectors according to the clustering granularity parameters, and obtain event attribution clustering results.
[0037] Please continue reading Figure 3 , Figure 3 is a schematic diagram of the execution flow of calculating clustering granularity parameters in the event attribution method based on the clustering model provided in the embodiment of the present application, such as Figure 3 As shown, the above clustering model is a hierarchical clustering model, and the process of calculating the clustering granularity parameter in the above step S3 may include the following steps S31 to S34.
[0038] S31. Obtaining a first sample pair consisting of two nodes in each merging operation during the historical hierarchical clustering process of the hierarchical clustering model.
[0039] S32: Process the first sample pair through the semantic parsing module of the preset large language model to obtain a node clustering analysis result of the first sample pair.
[0040] S33. Calculate the precision and recall, and the weighted harmonic mean of the precision and recall using the clustering level data of the first sample pair and the node clustering analysis result.
[0041] Among them, the weighted harmonic mean in step S33 is the F-beta value, which is the weighted harmonic mean of precision and recall, and is usually used to evaluate the performance of classification models. The F-beta value ranges from 0 to 1. The higher the value, the better the model performs in terms of precision and recall. A value of 1 indicates that the balance between precision and recall is optimal, which can help analyze the performance of the model in different categories and provide useful information in model optimization and comparison.
[0042] S34. According to the weighted harmonic mean and a preset consistency threshold range, a corresponding clustering level is selected as a clustering granularity parameter.
[0043] S4. Input the event attribution clustering result into the preset large language model to obtain the event attribution classification label output by the large language model.
[0044] Please continue reading Figure 4 , Figure 4 is a schematic diagram of the execution flow of adjusting the clustering category in the event attribution method based on the clustering model provided in the embodiment of the present application, such as Figure 4 As shown, the above clustering model is a hierarchical clustering model. Before the step of inputting the event attribution clustering result into the preset large language model in the above step S4, the following steps S41 to S44 may be included.
[0045] S41. Calculate the cosine distance between each event attribution and each cluster category in the event attribution clustering result to obtain the classification probability that each event attribution belongs to each cluster category.
[0046] Among them, cosine distance is a measurement method to measure the similarity between the directions of two vectors. It is often used in fields such as text mining, recommendation systems, and clustering. Cosine distance measures the angle between two vectors in multidimensional space, rather than the absolute size of the vectors or the Euclidean distance. The value range of cosine distance is between 0 and 1. The closer the value is to 0, the less similar the two vectors are, and the closer the value is to 1, the more similar the two vectors are.
[0047] S42. Setting at least one clustering category corresponding to each event attribution according to a preset classification probability threshold range.
[0048] S43. Normalize the classification probability and use the normalized classification probability to calculate the entropy value of each event attribution.
[0049] Among them, entropy is an important concept in information theory, which is used to measure the uncertainty or information content of a random variable. The higher the entropy value, the greater the uncertainty of the random variable, and vice versa. In information theory, entropy can be used to measure the average amount of information or the measure of uncertainty, such as Shannon Entropy. Shannon Entropy is the most common form of entropy in information theory and is used to measure the uncertainty of a random variable.
[0050] S44, removing the event attribution with the largest entropy value in each cluster category in the event attribution clustering result. The largest entropy value indicates that the uncertainty in the cluster category is the highest, that is, the classification accuracy is the worst, so the event attribution with the largest entropy value is removed.
[0051] As a preferred implementation, after the above step S43, the above vector model is further optimized according to the calculated entropy value, that is, the following steps S431 to S436 may also be included.
[0052] S431. Select the cluster category with the highest entropy value from the event attribution clustering results as the anchor category.
[0053] S432: Calculate the similarity distances between cluster categories, and select the cluster category with the smallest similarity distance to the anchor category as the cluster category to be determined.
[0054] S433. Randomly select one event attribution from each cluster category to be determined to form a fine-tuned event attribution data pair.
[0055] S434, analyzing and determining the event attribution with the highest similarity between the fine-tuned event attribution data pair and the anchor point category through the preset large language model, and updating the result of the analysis and determination to the fine-tuned event attribution data pair.
[0056] S435. Generate hard negative example triplets using anchor categories and fine-tuned event attribution data.
[0057] Among them, hard negative example triples refer to a special case in triple loss, where negative examples are considered to be "difficult to distinguish" or "closer to positive examples". When training a neural network, the introduction of hard negative example triplets can help improve the performance and convergence speed of the model during training. More complex loss function designs can be used, such as loss functions with Margin, to ensure that the distance between positive examples and hard negative examples is greater than a certain boundary value between the positive examples and easy negative examples. In each round of training, hard negative example triplets are used to calculate the loss, and the model parameters are updated through the back-propagation algorithm.
[0058] S436. Use the hard negative example triples for contrastive learning, and adjust the parameters of the vector model according to the results of the contrastive learning.
[0059] As a preferred implementation, after the above step S4, the event attribution data and the event attribution classification label may be first sent to the manual business expert, and the adjustment result data of the manual business expert may be received. Then, the event attribution classification label may be updated according to the adjustment result data, and the vector model and the clustering model may be optimized.
[0060] The event attribution method based on the clustering model provided in the embodiment of the present application obtains event attribution data, processes the event attribution data through a preset large language model, and obtains attribution cause key information data and attribution cause concept data; vectorizes the attribution cause key information data and attribution cause concept data through a preset vector model to obtain an event attribution feature vector; calculates clustering granularity parameters, uses a preset clustering model to cluster the event attribution feature vector according to the clustering granularity parameters, and obtains an event attribution clustering result; inputs the event attribution clustering result into a preset large language model to obtain an event attribution classification label output by the large language model, realizes the splitting and rewriting of event attribution, unifies the event attribution granularity, explores the deep meaning of event attribution, provides more feature data for event attribution clustering, introduces a large language model for synchronous judgment, compares with the result of hierarchical clustering, and then obtains the optimal clustering granularity, and selects a suitable data sample according to the entropy value to input into the large language model for clustering analysis, thereby improving the accuracy of event attribution based on the clustering model and reducing the analysis cost.
[0061] Please continue reading Figure 5 , Figure 5 is a schematic diagram of the system structure of an event attribution device based on a clustering model provided in an embodiment of the present application, such as Figure 5 As shown, the event attribution device 50 based on the clustering model includes: a data preprocessing module 51, a data vectorization module 52, a data clustering module 53 and a classification label module 54.
[0062] The data preprocessing module 51 is used to obtain event attribution data, and process the event attribution data through a preset large language model to obtain attribution cause key information data and attribution cause concept data.
[0063] The data vectorization module 52 is used to perform vectorization processing on the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector.
[0064] The data clustering module 53 is used to calculate clustering granularity parameters, and use a preset clustering model to perform clustering processing on the event attribution feature vectors according to the clustering granularity parameters to obtain event attribution clustering results.
[0065] The classification label module 54 is used to input the event attribution clustering result into the preset large language model to obtain the event attribution classification label output by the large language model.
[0066] As an optional implementation, the data preprocessing module 51 is specifically used to split the event attribution data into sub-cause data through a preset text processing tool; use a preset natural language processing model to process the sub-cause data, extract information features in the sub-cause data, and obtain the attribution cause key information data; process the attribution cause key information data through the regression prompt module of the preset large language model to obtain the attribution cause concept data.
[0067] As an optional implementation, the data clustering module 53 is also specifically used to obtain a first sample pair consisting of two nodes in each merging operation during the historical hierarchical clustering processing of the hierarchical clustering model; process the first sample pair through the semantic parsing module of the preset large language model to obtain the node clustering analysis result of the first sample pair; use the clustering level data of the first sample pair and the node clustering analysis result to calculate the precision and recall rate, as well as the weighted harmonic mean of the precision and the recall rate; and select the corresponding clustering level as the clustering granularity parameter according to the weighted harmonic mean and a preset consistency threshold range.
[0068] As an optional implementation, the classification label module 54 is also specifically used to calculate the cosine distance between each event attribution and each cluster category in the event attribution clustering result to obtain the classification probability that each event attribution belongs to each cluster category; set at least one cluster category corresponding to each event attribution according to a preset classification probability threshold range; normalize the classification probability, and use the normalized classification probability to calculate the entropy value of each event attribution; remove the event attribution with the largest entropy value in each cluster category in the event attribution clustering result.
[0069] As an optional implementation, the classification label module 54 is also specifically used to calculate the cosine distance between each event attribution and each cluster category in the event attribution clustering result to obtain the classification probability that each event attribution belongs to each cluster category; set at least one cluster category corresponding to each event attribution according to a preset classification probability threshold range; normalize the classification probability, and use the normalized classification probability to calculate the entropy value of each event attribution; remove the event attribution with the largest entropy value in each cluster category in the event attribution clustering result.
[0070] As an optional implementation, the classification label module 54 is also specifically used to select the cluster category with the highest entropy value from the event attribution clustering results as the anchor category; calculate the similarity distance between the cluster categories, and select the cluster category with the smallest similarity distance to the anchor category as the cluster category to be judged; randomly select an event attribution from each of the cluster categories to be judged to form a fine-tuning event attribution data pair; parse and judge the event attribution with the highest similarity between the fine-tuning event attribution data pair and the anchor category through the preset large language model, and update the result of the parsing and judgment to the fine-tuning event attribution data pair; use the anchor category and the fine-tuning event attribution data pair to generate a hard negative example triple; use the hard negative example triple to perform comparative learning, and adjust the parameters of the vector model according to the result of the comparative learning.
[0071] As an optional implementation, the event attribution device 50 based on the clustering model also includes a manual adjustment module, which is used to send the event attribution data and the event attribution classification label to a manual business expert, and receive the adjustment result data of the manual business expert; update the event attribution classification label according to the adjustment result data, and optimize the vector model and the clustering model.
[0072] As an optional implementation, the data vectorization module 52 is specifically used to extract the penalty cause key feature vector and penalty cause probability feature vector corresponding to the attribution cause key information data and the attribution cause concept data respectively through a preset vector model; perform vector splicing processing on the penalty cause key feature vector and the penalty cause probability feature vector to obtain the event attribution feature vector.
[0073] It should be noted that the above-mentioned event attribution device based on the clustering model can execute the event attribution method based on the clustering model provided in the embodiment of the present application, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not fully described in the embodiment of the event attribution device based on the clustering model, please refer to the event attribution method based on the clustering model provided in the embodiment of the present application.
[0074] Figure 6 is a schematic diagram of the hardware structure of an electronic device 600 for executing an event attribution method based on a clustering model provided in an embodiment of the present application, such as Figure 6 As shown, the electronic device 600 includes:
[0075] One or more processors 610 and memory 620, Figure 6 A processor 610 is taken as an example.
[0076] The processor 610 and the memory 620 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.
[0077] The memory 620, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the event attribution method based on the clustering model in the embodiment of the present application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 620, that is, implementing the event attribution method based on the clustering model in the above method embodiment.
[0078] The memory 620 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of an event attribution device based on a clustering model, etc. In addition, the memory 620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include a memory remotely arranged relative to the processor 610, and these remote memories may be connected to the event attribution device based on the clustering model via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0079] The one or more modules are stored in the memory 620, and when executed by the one or more processors 610, execute the event attribution method based on the clustering model in any of the above method embodiments, for example, execute the above described Figure 2 Steps S1 to S4 of the method, Figure 3 Steps S31 to S34 of the method, Figure 4 Steps S41 to S44 of the method are implemented Figure 5 The functions of modules 51-54 in.
[0080] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.
[0081] The embodiment of the present application provides a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors, such as Figure 6 A processor 610 in the embodiment may enable the one or more processors to execute the event attribution method based on the clustering model in any of the above method embodiments, for example, to execute the above described Figure 2 Steps S1 to S4 of the method, Figure 3 Steps S31 to S34 of the method, Figure 4 Steps S41 to S44 of the method are implemented Figure 5 The functions of modules 51-54 in.
[0082] The present application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, wherein the computer program includes program instructions. When the program instructions are executed by the electronic device, the electronic device is enabled to execute the event attribution method based on the clustering model in any of the above method embodiments, for example, to execute the above described event attribution method. Figure 2 Steps S1 to S4 of the method, Figure 3 Steps S31 to S34 of the method, Figure 4 Steps S41 to S44 of the method are implemented Figure 5 The functions of modules 51-54 in.
[0083] The device embodiments described above are merely illustrative, wherein 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 modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0084] Through the description of the above implementation methods, ordinary technicians in this field can clearly understand that each implementation method can be implemented by means of software plus a general hardware platform, and of course, it can also be implemented by hardware. Ordinary technicians in this field can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.
[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Under the concept of the present application, the technical features in the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other changes in different aspects of the present application as described above, which are not provided in detail for the sake of simplicity. Although the present application has been described in detail with reference to the aforementioned embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present application.
Claims
1. An event attribution method based on a clustering model, characterized in that: include: Acquire event attribution data, and process the event attribution data through a preset large language model to obtain attribution cause key information data and attribution cause concept data; Vectorizing the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector; Calculating clustering granularity parameters, using a preset clustering model to perform clustering processing on the event attribution feature vector according to the clustering granularity parameters, and obtaining event attribution clustering results; The event attribution clustering result is input into the preset large language model to obtain the event attribution classification label output by the large language model.
2. The event attribution method based on clustering model according to claim 1, characterized in that: The step of processing the event attribution data through a preset large language model to obtain attribution cause key information data and attribution cause concept data includes: Splitting the event attribution data into sub-cause data using a preset text processing tool; Process the sub-cause data using a preset natural language processing model, extract information features in the sub-cause data, and obtain key information data of the attribution cause; The attribution cause key information data is processed by the regression prompt module of the preset large language model to obtain the attribution cause concept data.
3. The event attribution method based on clustering model according to claim 1, characterized in that: The clustering model is a hierarchical clustering model, and the step of calculating the clustering granularity parameter includes: Obtaining a first sample pair consisting of two nodes in each merging operation during the historical hierarchical clustering processing of the hierarchical clustering model; Processing the first sample pair by using the semantic parsing module of the preset large language model to obtain a node clustering analysis result of the first sample pair; Calculating the precision and recall, and the weighted harmonic mean of the precision and the recall using the clustering level data of the first sample pair and the node clustering analysis result; According to the weighted harmonic mean and a preset consistency threshold range, a corresponding clustering level is selected as the clustering granularity parameter.
4. The event attribution method based on clustering model according to claim 1, characterized in that: Before the step of inputting the event attribution clustering result into the preset large language model, the method further includes: Calculating the cosine distance between each event attribution and each cluster category in the event attribution clustering result to obtain the classification probability that each event attribution belongs to each cluster category; Setting at least one clustering category corresponding to each event attribution according to a preset classification probability threshold range; Normalizing the classification probability, and using the normalized classification probability to calculate the entropy value of each event attribution; The event attribution with the largest entropy value in each cluster category in the event attribution clustering result is removed.
5. The event attribution method based on clustering model according to claim 4 is characterized in that: After the step of calculating the entropy value of each event attribution using the normalized classification probability, the method further comprises: Selecting the clustering category with the highest entropy value from the event attribution clustering results as the anchor category; Calculating the similarity distances between the cluster categories, and selecting the cluster category with the smallest similarity distance to the anchor point category as the cluster category to be determined; Randomly extract one event attribution from each of the cluster categories to be determined to form a fine-tuned event attribution data pair; Analyzing and determining the event attribution of the fine-tuned event attribution data pair with the highest similarity to the anchor point category through the preset large language model, and updating the result of the analysis and determination to the fine-tuned event attribution data pair; Generate a hard negative example triple using the anchor category and the fine-tuned event attribution data pair; The hard negative example triples are used for contrastive learning, and the parameters of the vector model are adjusted according to the result of the contrastive learning.
6. The event attribution method based on clustering model according to claim 1, characterized in that: After the step of obtaining the event attribution classification label output by the large language model, the method further includes: Sending the event attribution data and the event attribution classification label to a human business expert, and receiving adjustment result data from the human business expert; The event attribution classification label is updated according to the adjustment result data, and the vector model and the clustering model are optimized.
7. The event attribution method based on clustering model according to claim 1, characterized in that: The step of vectorizing the attribution cause key information data and the attribution cause concept data through a preset vector model to obtain an event attribution feature vector includes: Extracting the penalty cause key feature vector and the penalty cause probability feature vector corresponding to the attribution cause key information data and the attribution cause concept data respectively through a preset vector model; The key feature vector of the penalty reason and the probability feature vector of the penalty reason are subjected to vector concatenation processing to obtain the event attribution feature vector.
8. An event attribution device based on a clustering model, characterized in that: include: A data preprocessing module is used to obtain event attribution data, process the event attribution data through a preset large language model, and obtain attribution cause key information data and attribution cause concept data; A data vectorization module, used for performing vectorization processing on the key information data of the attribution cause and the conceptual data of the attribution cause through a preset vector model to obtain an event attribution feature vector; A data clustering module, used to calculate clustering granularity parameters, and use a preset clustering model to perform clustering processing on the event attribution feature vector according to the clustering granularity parameters to obtain an event attribution clustering result; A classification label module is used to input the event attribution clustering result into the preset large language model to obtain the event attribution classification label output by the large language model.
9. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the event attribution method based on the clustering model described in any one of claims 1-7.
10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by an electronic device, the electronic device executes the event attribution method based on a clustering model as described in any one of claims 1 to 7.