Error table data interpretable repair method and device and electronic equipment

Abstract

The invention discloses an interpretable error table data repairing method and device and electronic equipment. The method comprises the steps that an error table data information matrix, a table data error mask matrix and an error table data semantic embedding matrix are obtained; according to the three matrixes, a plurality of tuples with the strongest association degree with each error data are retrieved; in the error table data information matrix, sampling a plurality of representative error tuples for repairing; generating a repair thinking chain of each representative error data by using a large language model according to repair and retrieval results of the representative error data; generating a plurality of rule instances for each repair rule by using a large language model according to the repair thinking chain of the representative error data; applying all rule instances, generating a plurality of repair candidate results for each piece of error data, and voting to select each piece of error data to select a final result; and for error data which is not repaired by the rule, according to a retrieval result, obtaining a repair result and a repair thinking chain by using a large language model.

Description

Technical Field

[0001] The present application relates to the field of table data repair, and in particular to a method and device for interpretably repairing erroneous table data, and an electronic device. Background Art

[0002] Traditional methods for repairing table error data mainly rely on preset data rules, such as functional dependencies, conditional functional dependencies, and negated constraints, which are usually formulated by domain experts based on their in-depth understanding of data characteristics and business logic. However, such methods are highly dependent on expert knowledge, requiring users to not only fully understand the data to be repaired, but also be familiar with the configuration of the data repair system to reasonably set and adjust the rules. In recent years, with the development of machine learning technology, some emerging methods have automatically completed data repair tasks by training models, avoiding users from manually providing specific rules, simplifying the repair process and improving the level of automation. However, these machine learning-based methods usually rely on large-scale training data, which is difficult to implement when data is scarce or difficult to label, and the models, especially deep learning models, lack sufficient interpretability, making the repair process and results difficult for users to understand and trust. In addition, these methods perform poorly when dealing with undefined or novel error patterns, because the models may not be able to effectively identify and handle anomalies beyond the scope of training data. Summary of the invention

[0003] In order to overcome the above technical problems, the present application implements and provides an interpretable repair method and device for erroneous table data, and an electronic device to ensure the fairness of the prediction.

[0004] According to a first aspect of an embodiment of the present application, a method for interpretably repairing erroneous table data is provided, comprising: S1: Obtain the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S2: Retrieving a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S3: In the error table data information matrix, a number of representative error tuples are sampled for repair; S4: according to the repair and retrieval results of the representative error data, a repair thought chain for each representative error data is generated using a large language model; S5: generating multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; S6: Apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; S7: For erroneous data that are not repaired by the rules, a large language model is used to obtain the repair results and the repair thinking chain based on its retrieval results.

[0005] According to a second aspect of an embodiment of the present application, there is provided a device for interpreting and repairing erroneous table data, comprising: An acquisition module, used for acquiring an error table data information matrix, a table data error mask matrix and an error table data semantic embedding matrix; A retrieval module, used to retrieve a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; A sampling and repairing module, used for sampling a number of representative error tuples in the error table data information matrix for repair; A repair thought chain generation module, used to generate a repair thought chain for each representative error data using a large language model according to the repair result and retrieval result of the representative error data; A repair rule generation module, used to generate multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; The rule repair module is used to apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; The large language model repair module is used to obtain the repair results and repair thought chains using the large language model based on the retrieval results of the erroneous data that has not been repaired by the rules.

[0006] According to a third aspect of an embodiment of the present application, there is provided an electronic device, including: one or more sensors; one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0007] The technical solution provided by the embodiments of the present application may have the following beneficial effects: The present invention designs a context-related tuple retrieval method for table error data repair tasks, which can effectively identify and extract context information closely related to error data, and improve the accuracy of large language model repair data in a retrieval-enhanced manner; by sampling representative error tuples and manually repairing them by users, combined with the repair thought chain generated by the large language model, the large language model can learn the repair strategy of typical errors in the data set, thereby reducing the "illusion" problem of the large language model when facing data in a specific field, and further improving the accuracy of repair; by automatically generating data repair rules, some frequently used repair strategies are abstracted into data rules, avoiding frequent calls to large language models, thereby improving the efficiency of data repair and reducing costs; the interpretability of data repair is achieved in two ways: on the one hand, using data repair rules to provide clear repair basis; on the other hand, for error data that is not repaired by the rules, the repair reasons are elaborated in detail in the form of thought chains. The prediction accuracy of this method is 42% higher than that of the current optimal table error data repair algorithm in terms of accuracy and 2.8 times higher in efficiency. It should be understood that the above general description and the detailed description below are only exemplary and explanatory, and cannot limit this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0009] Figure 1 It is a flowchart of an interpretable repair method for erroneous table data provided in an embodiment of the present application.

[0010] Figure 2 It is a block diagram of an interpretable repair device for error table data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0012] The method for interpretable repair of erroneous tabular data provided in this application can be applied to interpretable repair of erroneous tabular data in almost all fields, such as medical data and traffic data, etc. The following is a detailed explanation using medical data as an example.

[0013] Figure 1is a flowchart of an interpretable method for removing error table data provided by the present application according to an exemplary implementation example, such as Figure 1 As shown, the method may include the following steps: S1: Obtain the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S11: Acquire table data information and table data error information collected by the sensor; Specifically, medical sensors can be used to collect patients' health data, environmental sensors can be used to monitor environmental data, and industrial sensors can be used to collect equipment's operating status and fault information. This implementation uses the use of sensors to collect patients' health data as an example for specific explanation. A patient's health data can have multiple fields, such as a patient's blood pressure, blood sugar level, heart rate, weight, height and other feature fields; errors may exist in some health data fields due to incorrect or missing information filled in by the doctor, failure of the measuring equipment and other reasons. S12: Construct an error table data information matrix and a table data error mask matrix based on the table data information and the table data error information; Specifically, for an error table containing N tuples, each tuple containing d feature fields, the error table data information matrix can be expressed as , where each tuple It can be expressed as The error information of these data will be stored in a table data error mask matrix, which can be expressed as ,in The error status corresponding to a single tuple, The value range is {0, 1}, and ;when No. When the dimension data is wrong, ;otherwise .

[0014] S13: semantically embedding each element in the error table data information matrix through a semantic embedding method, thereby generating an error table data semantic embedding matrix.

[0015] Specifically, the semantic embedding model can use the pre-trained Sentence-BERT model to semantically embed the data of each cell in the table, convert it into a high-dimensional normalized vector, and obtain the semantic embedding matrix of the error table data, which is expressed as ,in Is a tuple The model can extract the semantic features of cell data and map them into a unified vector space, thereby capturing the deep semantic information of the data.

[0016] S2: Retrieving a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S21: for each column to be repaired, calculate the normalized mutual information value between it and the remaining columns, and regard the columns whose normalized mutual information values ​​are greater than or equal to the specified threshold as relevant columns; Specifically, for each , the context-dependent tuple retrieval module calculates The normalized mutual information value with the rest of the columns is Any column in , and The normalized mutual information of is expressed as: in yes and The joint probability distribution of and They are and The marginal probability distribution of . For any column ,like If the value of is greater than or equal to a specified threshold, such as 0.5, it is selected as a relevant column. The present application filters irrelevant columns in the table data by normalized mutual information, thereby reducing the amount of calculation in the subsequent retrieval process and enhancing the accuracy of the retrieval.

[0017] S22: Based on the weighted cosine similarity in the relevant column, measure the similarity between the remaining tuples and the tuple to be repaired, and preliminarily sort the tuples according to the similarity, and retrieve several tuples with the highest ranking; Specifically, for Any tuple in ( ), which is related to The weighted cosine similarity is expressed as: in, and They are tuples and No. the error status of the column element, It is the first List, is the column currently waiting to be repaired. Yes and The normalized mutual information value of and They are tuples and No. Semantic embedding of column elements.

[0018] The weighted cosine similarity is used to calculate the similarity between the tuple to be repaired and other tuples and sort them, and retrieve several tuples with the highest similarity. This application considers the data error state, the normalized mutual information value between the column where the data is located and the column to be repaired, and the semantic embedding of the data to design the weighted cosine similarity, uses the error state weight to reduce the impact of erroneous data on retrieval, adjusts the weight according to the mutual information value between columns, ensures the dominant position of important columns in the similarity calculation, captures more complex data associations through semantic embedding, and accurately measures the similarity between tuples, so that each tuple is accurately sorted according to the similarity with the tuple to be repaired.

[0019] S23: The preliminarily sorted tuples are further sorted according to data quality to ultimately determine the tuples with the highest ranking, that is, the tuples with the strongest correlation with the erroneous data.

[0020] Specifically, the tuples with the same weighted cosine similarity are sorted again, and the top few tuples are selected. The sorting basis is: Determine whether the value of the column to be repaired is correct. If the value of the column to be repaired in the tuple is correct, the tuple is selected first. If the values ​​of the columns to be repaired of these tuples are the same (that is, both are correct or both are incorrect), the tuple containing more correct data is preferred.

[0021] After sorting, the top tuples are retrieved as context-related tuples of the error data. Tuples with the same weighted cosine similarity are re-sorted based on data quality to improve the accuracy of sorting and further help retrieve the most relevant context information for the tuple to be repaired.

[0022] S3: In the error table data information matrix, sample several representative error tuples for repair; this step may include the following sub-steps: S31: using clustering algorithm to group tuples, the user specifies the number of clusters, and divides all tuples into several clusters; Specifically, the K-Means algorithm may be used to cluster all tuples, wherein the number of clusters is specified by the user, and all tuples are divided into several clusters.

[0023] S32: adopt a greedy algorithm to select a tuple from each cluster as a representative error tuple, so that the selected representative error tuple can cover as many error columns as possible and contain more error data; Specifically, a greedy algorithm is used to select a representative tuple from each cluster, and the priority of selection is: Cover as many error columns as possible; Cover as many error cells as possible.

[0024] This application designs a greedy algorithm so that the selected representative tuples can cover as many error columns as possible and contain as much error data as possible, so that the representative tuples can cover more comprehensive error types, thereby helping to generate more comprehensive error repair strategies and improve error repair accuracy.

[0025] S33: The selected representative error tuple is handed over to the user for manual repair to obtain the corresponding correct result.

[0026] Specifically, the representative error tuples finally selected will be manually corrected by the user to obtain the correct values ​​of these tuples.

[0027] S4: Based on the repair and retrieval results of the representative error data, a repair thought chain for each representative error data is generated using a large language model; this step may include the following sub-steps: S41: Based on the representative error tuples repaired by the user, a repair thought chain is constructed for each error column to generate prompt words; Specifically, the repair thought chain generates prompt words mainly including the following parts: i. Task description: Introduce the goal of the repair thought chain generation task, which is to generate a thought chain for error repair.

[0028] ii. Common examples: Generate examples by randomly introducing four common error types (such as missing values, spelling errors, formatting issues, and constraint violations) on a dataset, such as the Chicago Food Dataset. These examples serve as contextual demonstrations to guide the large language model to automatically generate repair thought chains.

[0029] iii. Representative tuple information: Provides detailed information about the representative tuple, including the error tuple and error value.

[0030] iv. Retrieved related tuples: Through the context-dependent tuple retriever, the tuple information related to the representative tuple is retrieved as the reference context for repair.

[0031] v. Correction results provided by the user: records the user's correction information on the representative tuple error column to guide the generation of the repair thinking chain.

[0032] S42: Generate prompt words using the above-mentioned repair thought chain, and call the large language model to obtain the repair thought chain for each error data.

[0033] Specifically, by using the above-mentioned repair thought chain generation prompt words, the large language model is called to automatically generate a repair thought chain for each error data. This application uses a large language model to automatically generate a repair thought chain for representative error data, reveals the repair strategy for representative error data, reduces the cost of manually generating repair thought chains and potential inconsistencies, and provides guidance for subsequent rule generation and error repair.

[0034] S5: Generate multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; this step may include the following sub-steps: S51: for each designated repair rule, using a prompt word paraphrase tool, based on some representative tuples and their repair thought chains, generate multiple different versions of repair rule generation prompt words; Specifically, half of the tuples are randomly selected from the representative tuple set of dirty data as the training set for generating repair rules. Then, for each rule type in the repair rule description set, the prompt word paraphrase tool is used to reconstruct the specified task description and generate multiple different versions of the repair rule generation task description to guide the large language model to generate possible repair rule candidate sets. On this basis, the generated task description is combined with the representative tuple repair workflow to form the repair rule generation prompt word.

[0035] S52: Generate prompt words using the above repair rules, and call the large language model to generate specific instances of the repair rules; Specifically, by using the above-mentioned repair rule to generate prompt words, the large language model is called to automatically generate a repair rule instance.

[0036] S53: Verify each rule on all representative tuples and all correct tuples, and only retain the rule instances that pass the verification.

[0037] Specifically, all the annotated representative tuples and the correct tuples detected are selected as the verification set, and those repair rules that can correctly repair all dirty data values ​​without destroying the correct data are retained to form the final set of repair rules. This application uses a large language model to extract repair rules from the repair process of representative tuples, thereby reducing the number of calls to the large language model in the subsequent repair process, reducing the computational cost, and providing an explanation for the error repair results in the form of repair rules.

[0038] S6: Apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; this step includes the following sub-steps: S61: For each column to be repaired, firstly execute all verified rule instances in parallel to generate multiple interpretable repair results for each erroneous data; Specifically, for each column to be repaired, all verified rule instances are first executed in parallel to generate multiple interpretable repair results for each erroneous data.

[0039] S62: Selecting a repair result supported by most rules from among the multiple interpretable repair results as a final repair result for each erroneous data.

[0040] Specifically, among the multiple interpretable repair results, the repair result supported by the majority of rules is selected as the final repair result for each erroneous data. The final repair result is selected by a voting strategy, which effectively reduces the risk of erroneous repair caused by a single rule deviation and improves the reliability of the repair result.

[0041] S7: For the erroneous data that is not repaired by the rules, based on the retrieval results, the large language model is used to obtain the repair results and the repair thought chain. This step may include the following sub-steps: S71: Construct an interpretable data repair prompt word for each error data that is not repaired by the rule based on the representative error tuple and its repair thinking chain; Specifically, the interpretable data repair prompt word includes the following five components: Task description: Introduce the target task of generating repair results and their detailed repair thought chain; ii. General repair examples: Based on the four dirty tuples with random errors in the Chicago food dataset, four general repair examples are constructed. Each example includes a dirty tuple, a dirty value, and a retrieved tuple, and outputs the repair result and the corresponding repair thought chain; iii. Representative repair example: For each representative tuple containing the target dirty column, an example is constructed, with the input being the representative tuple, the dirty value, and the retrieved tuple, and the output being the repair result and the reasoning thought chain generated by the thought chain generation prompt words; iv. Error tuple: Provides the tuple containing the dirty value in the target dirty column and its corresponding dirty value; v. Context-dependent tuples: The k tuples that are most similar to the error tuple are selected through the context-dependent tuple retrieval tool.

[0042] S72: Use the above-mentioned explainable data repair prompt words and call the large language model to generate a repair result and a repair thought chain for each erroneous data.

[0043] According to the explainable data repair prompt words, the large language model is called to generate the repair results of all dirty tuples and the corresponding repair thought chain. The repair result is the predicted value of the dirty data, and the repair thought chain records in detail the reasoning steps, considerations and contextual information in the repair process. This application uses a large language model to generate repair results and repair thought chains for erroneous data through retrieval enhancement, thought chain prompts and few sample prompts, which reduces the "hallucination" problem in the repair process and improves the accuracy of the repair results.

[0044] The error table data of the present invention can explain the repair method, which is implemented on an Ubuntu 20.04 system running on an Intel core 2.30GHz server with 512GB memory. The accuracy and efficiency of the system are tested under data sets containing data from different fields, different error rates, different error types, and different sizes. The data set introduction is shown in Table 1. For the measurement of system accuracy, we use precision (P), recall (R) and F1 value (F1). The calculation formula of precision is , the calculation formula of recall is ,in The number of errors correctly fixed for the system, is the number of errors that the system incorrectly repaired, is the number of errors that the system missed repairing. The larger the values ​​of these two indicators, the higher the repair accuracy of the system; for the measurement of system efficiency, the running time indicator is used, and the smaller the value of this indicator, the faster the running speed.

[0045] The performance of the fair prediction method for missing table data (i.e., ZeroEC) proposed in the present invention is analyzed through experimental tests, and the results are shown in Tables 2 and 3. It can be seen that ZeroEC has improved the accuracy by more than 42% compared with the existing error table data repair systems (Scare, Holoclean, Baran, FMs), because ZeroEC samples representative error data, learns the repair strategies of these representative errors, and generalizes these repair strategies to other data to be repaired in a few-sample learning manner, overcoming the "hallucination" problem of large language models when repairing data in specific fields. In addition, the average operating efficiency of ZeroEC is more than 2.8 times that of the existing optimal system, because ZeroEC can complete data repair without training, and automatically generates data repair rules for fast repair, thereby avoiding calling large language models one by one for repair of each error data.

[0046] Table 1: Dataset introduction Table 2: Comparison of the accuracy of repair systems on various datasets Table 3: Comparison of the efficiency of repair systems on various datasets (time unit: seconds) Corresponding to the aforementioned embodiment of the fair prediction method for missing table data, the present application also provides an embodiment of the fair prediction method for missing table data.

[0047] Figure 2 is a block diagram of the missing table data fair prediction method device of the present application. Figure 2 , the device comprises: An acquisition module, used for acquiring an error table data information matrix, a table data error mask matrix and an error table data semantic embedding matrix; A retrieval module, used to retrieve a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; A sampling and repairing module, used for sampling a number of representative error tuples in the error table data information matrix for repair; A repair thought chain generation module, used to generate a repair thought chain for each representative error data using a large language model according to the repair result and retrieval result of the representative error data; A repair rule generation module, used to generate multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; The rule repair module is used to apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; The large language model repair module is used to obtain the repair results and repair thought chains using the large language model based on the retrieval results of the erroneous data that has not been repaired by the rules.

[0048] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The device embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application scheme. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0049] Accordingly, the present application also provides an electronic device, comprising: one or more sensors; one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect. Accordingly, the present application also provides a computer-readable storage medium on which computer instructions are stored, and when the instructions are executed by a processor, the above-mentioned missing table data fairness prediction method is implemented.

[0050] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0051] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for interpretable repair of erroneous table data, characterized in that: include: S1: Obtain the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S2: Retrieving a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; S3: In the error table data information matrix, a number of representative error tuples are sampled for repair; S4: according to the repair and retrieval results of the representative error data, a repair thought chain for each representative error data is generated using a large language model; S5: generating multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; S6: Apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; S7: For erroneous data that are not repaired by the rules, a large language model is used to obtain the repair results and the repair thinking chain based on its retrieval results.

2. The method according to claim 1, characterized in that S1 specifically includes: S11: Acquire table data information and table data error information collected by the sensor; S12: constructing an error table data information matrix and a table data error mask matrix according to the table data information and the table data error information; S13: semantically embedding each element in the error table data information matrix through a semantic embedding method, thereby generating an error table data semantic embedding matrix.

3. The method according to claim 1, characterized in that S2 specifically includes: S21: for each column to be repaired, calculate the normalized mutual information value between it and the remaining columns, and regard the columns whose normalized mutual information values ​​are greater than or equal to the specified threshold as relevant columns; S22: Based on the weighted cosine similarity in the relevant column, measure the similarity between the remaining tuples and the tuple to be repaired, and preliminarily sort the tuples according to the similarity, and retrieve the top several tuples; S23: The preliminarily sorted tuples are further sorted according to data quality to ultimately determine the tuples with the highest ranking, that is, the tuples with the strongest correlation with the erroneous data.

4. The method according to claim 1, characterized in that: S3 specifically includes: S31: using clustering algorithm to group tuples, the user specifies the number of clusters, and divides all tuples into several clusters; S32: adopt a greedy algorithm to select a tuple from each cluster as a representative error tuple, so that the selected representative error tuple can cover as many error columns as possible and contain more error data; S33: The selected representative error tuple is handed over to the user for manual repair to obtain the corresponding correct result.

5. The method according to claim 1, characterized in that S4 specifically includes: S41: Based on the representative error tuples repaired by the user, a repair thought chain is constructed for each error column to generate prompt words; S42: Generate prompt words using the above-mentioned repair thought chain, and call the large language model to obtain the repair thought chain for each error data.

6. The method according to claim 1, characterized in that S5 specifically includes: S51: for each designated repair rule, using a prompt word paraphrase tool, based on some representative tuples and their repair thought chains, generate multiple different versions of repair rule generation prompt words; S52: Generate prompt words using the above repair rules, and call the large language model to generate specific instances of the repair rules; S53: Verify each rule on all representative tuples and all correct tuples, and only retain the rule instances that pass the verification.

7. The method according to claim 1, characterized in that S6 specifically includes: S61: For each column to be repaired, firstly execute all verified rule instances in parallel to generate multiple interpretable repair results for each erroneous data; S62: Among the multiple interpretable repair results, select a repair result supported by most rules as the final result of each erroneous data.

8. The method according to claim 1, characterized in that S7 specifically includes: S71: Construct an interpretable data repair prompt word for each error data that is not repaired by the rule based on the representative error tuple and its repair thinking chain; S72: Use the above-mentioned explainable data repair prompt words and call the large language model to generate a repair result and a repair thought chain for each erroneous data.

9. An interpretable repair device for erroneous table data, characterized in that: include: An acquisition module, used for acquiring an error table data information matrix, a table data error mask matrix and an error table data semantic embedding matrix; A retrieval module, used to retrieve a number of tuples with the strongest association with each error data according to the error table data information matrix, the table data error mask matrix and the error table data semantic embedding matrix; A sampling and repairing module, used for sampling a number of representative error tuples in the error table data information matrix for repair; A repair thought chain generation module, used to generate a repair thought chain for each representative error data using a large language model according to the repair result and retrieval result of the representative error data; A repair rule generation module, used to generate multiple rule instances for each repair rule using a large language model according to the repair thought chain of the representative error data; The rule repair module is used to apply all rule instances, generate multiple repair candidate results for each erroneous data, and vote for each erroneous data to select the final result; The large language model repair module is used to obtain the repair results and repair thought chains using the large language model based on the retrieval results of the erroneous data that has not been repaired by the rules.

10. An electronic device, characterized in that: include: one or more sensors; one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

Images