Key data extraction method and system
Through the automated extraction method based on the ERNIE3.0 model, hotline work orders are preprocessed and key elements are extracted. Combined with the ChatGLM3-6B and MGeo models, the traditional manual extraction is solved, and the traditional artificial extraction is inefficient and subjective, achieving high efficiency and accuracy work order processing reduces costs.
Patent Information
- Application Number
- CN202411529014.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2025-02-18
AI Technical Summary
The traditional method of manually extracting key elements of hotline work orders is inefficient, highly subjective, and error-prone, difficult to adapt to diversified expressions, and is expensive, and cannot meet the requirements of modern business management for efficiency, accuracy and low cost.
The automated extraction method based on the ERNIE3.0 model is adopted to pre-process the work order data, including word segmentation, encoding, truncation and filling. Then, the text features are extracted using the ERNIE3.0 model to intelligently predict case categories, and combined with ChatGLM3-6B and The MGeo model extracts the responsible subject and address information to realize the automated extraction and integration of key elements.
It realizes a comprehensive, accurate and efficient analysis of work order content, extracts key elements, solves the problems of inefficiency, strong subjectivity and error-prone traditional manual processing, improves the efficiency of information collection and work order processing, and reduces operations cost.
Smart Images
Figure CN119476341A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing, and in particular, to a key data extraction method and system thereof. Background Art
[0002] With the improvement of social informatization, hotline work orders have become an important way for enterprises and government departments to handle user complaints, problem feedback, etc. However, due to the complexity of work order content and the large number of work orders, manual work order processing is not only time-consuming and laborious, but also prone to information omission or improper processing. Traditional work order processing methods often rely on keyword matching or simple rule engines. This method has obvious shortcomings when facing diverse and complex natural language expressions and cannot effectively extract the key elements in the work order, thus affecting the efficiency and accuracy of work order processing. At present, the extraction of key elements of work orders is mainly manual extraction, and information is entered based on the subjective judgment of information entry personnel. In traditional hotline work order processing, the extraction of key elements mainly relies on manual operation. Although manual extraction has certain flexibility and accuracy, this method has significant defects when facing a large number of work orders: (1) Low efficiency: Manual processing requires a lot of manpower and time, especially when the number of work orders is huge, the processing efficiency is difficult to meet actual needs. This not only increases the burden on operators, but also may lead to backlogs and delays in work orders. (2) High subjectivity: When manually extracting work order elements, it often relies on the operator's experience and judgment, which is easily affected by personal understanding and emotions, resulting in inconsistent or inaccurate extraction results. This subjectivity may affect the fairness and accuracy of subsequent work order processing. (3) Error-prone: When faced with a large number of complex and diverse work order contents, manual operations are bound to make mistakes such as omissions, misjudgments, or incomplete information extraction. Such errors not only affect the processing results of the work order, but may also lead to serious customer complaints and trust crises. (4) Difficult to adapt to diverse expressions: The natural language expressions used in work orders are diverse, and may involve different professional terms, colloquial expressions, and even spelling errors. When humans identify and understand these complex expressions, they are prone to misunderstandings and find it difficult to fully and accurately extract key information. (5) High cost: In order to ensure the timeliness and accuracy of work order processing, a large amount of manpower is usually required to process it, which not only increases the operating costs of enterprises or institutions, but also makes it difficult to achieve large-scale processing when the business expands. In summary, the manual extraction of key elements of hotline work orders has many defects in terms of efficiency, accuracy, consistency, etc., and it is difficult to meet the requirements of modern business management for efficient, accurate and low-cost work order processing.
[0003] Therefore, how to provide a more intelligent and automated key factor extraction method to replace or assist manual extraction and improve the overall efficiency of hotline work order processing has become an urgent problem to be solved in this field. Summary of the invention
[0004] The present application proposes a key data extraction method, comprising the following steps: obtaining work order data and preprocessing the work order data; extracting key element information data from the preprocessed work order data; and integrating the extracted key element information data.
[0005] As above, the work order data includes case type, dispatch department, work order topic, responsible party, incident address and incident time data.
[0006] As mentioned above, the preprocessing of the work order data includes the following sub-steps: segmenting the work order data to generate a segmentation list; encoding the segmentation list; and truncating and filling the encoded segmentation list.
[0007] As above, truncating and padding the encoded word segmentation list includes determining the maximum input length that the model can accept. If the length of the encoded word segmentation list is less than the maximum input length that the model can accept, padding the length of the word segmentation list; if the length of the encoded word segmentation list exceeds the maximum input length that the model can accept, truncating the excess part.
[0008] As mentioned above, the extraction of key element information data includes extracting text features in the work order data based on the ERNIE3.0 model, intelligently predicting the case category to which the work order content belongs, and providing a confidence score.
[0009] A key data extraction system specifically includes: a preprocessing unit, a key data extraction unit and an integration unit; the preprocessing unit is used to obtain work order data and preprocess the work order data; the key data extraction unit is used to extract key element information data from the preprocessed work order data; the integration unit is used to integrate the extracted key element information data.
[0010] As mentioned above, the work order data obtained in the pre-processing unit includes case type, dispatch department, work order topic, responsible party, incident address and incident time data.
[0011] As mentioned above, the preprocessing unit performs preprocessing of the work order data, including the following sub-steps: performing word segmentation on the work order data to generate a word segmentation list; encoding the word segmentation list; and truncating and filling the encoded word segmentation list.
[0012] As above, the preprocessing unit truncates and fills the encoded word segmentation list, including determining the maximum input length that the model can accept. If the length of the encoded word segmentation list is less than the maximum input length that the model can accept, the length of the word segmentation list is padded. If the length of the encoded word segmentation list exceeds the maximum input length that the model can accept, the excess part is truncated.
[0013] As mentioned above, the key data extraction unit extracts key element information data, including extracting text features in the work order data based on the ERNIE3.0 model, intelligently predicting the case category to which the work order content belongs, and giving a confidence score.
[0014] This application has the following beneficial effects:
[0015] (1) This application uses artificial intelligence technology to conduct a comprehensive, accurate, and efficient analysis of the work order content and extract key elements, solving the problems of low efficiency, strong subjectivity, and proneness to errors in traditional manual processing.
[0016] (2) This application unifies, normalizes and standardizes the extraction of hotline work order content, solving the problems of confusion and non-standardization in the process of extracting work order elements.
[0017] (3) The key data extraction method proposed in this application can be deployed in an accelerated manner on multiple device terminals, greatly improving the efficiency of information collection and work order processing, and solving the problems of low efficiency and lack of convenience of traditional manual processing. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.
[0019] Figure 1 is a flowchart of a key data extraction method provided according to an embodiment of the present application;
[0020] Figure 2 It is a structural diagram of the ERNIE 3.0 model provided according to an embodiment of the present application;
[0021] Figure 3 It is a structural diagram of the ChatGLM model provided according to the embodiment of this application;
[0022] Figure 4 is a schematic diagram of the structure of the MGeo model provided according to an embodiment of the present application;
[0023] Figure 5 It is a schematic diagram of the internal structure of a key data extraction system provided according to an embodiment of the present application. DETAILED DESCRIPTION
[0024] The following is a clear and complete description of the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.
[0025] This application introduces an automated extraction method based on natural language processing (NLP) and large model technology, which can be deployed on multi-device terminals, thereby solving the problems of low efficiency, strong subjectivity, proneness to errors, and difficulty in adapting to diverse expressions in the traditional manual extraction of key elements of hotline work orders.
[0026] Embodiment 1
[0027] like Figure 1 As shown, this embodiment provides a key data extraction method, which specifically includes the following steps:
[0028] Step S1: Obtain work order data and pre-process the work order data.
[0029] The work order data includes case type, dispatch department, work order topic, responsible party, incident address and incident time, etc. The preprocessing of work order data includes word segmentation, encoding, truncation and padding of the work order data text. The preprocessing of work order data includes the following sub-steps:
[0030] Step S11: Perform word segmentation on the work order data and generate a word segmentation list.
[0031] Input the work order data text into the bert-base-chinese tokenizer, split the text into a series of words or subwords, and generate a corresponding token list. Process special characters, punctuation marks, and numbers in the work order to ensure that they can be correctly tokenized.
[0032] Step S12: Encode the word segmentation list.
[0033] The vocabulary of the bert-base-chinese pre-trained model is used to map each word to a corresponding digital ID. If some words do not exist in the vocabulary, they are usually encoded as special unknown word numbers. For some common words or characters that already exist in the vocabulary, they are directly converted to corresponding numbers.
[0034] Step S13: truncate and fill the encoded word segmentation list.
[0035] First, the maximum input length accepted by each model is determined, and the models are ERNIE3.0 model, Chatglm3-6B large model and Mgeo model.
[0036] If the length of the encoded word list exceeds the maximum input length of the three models, the excess part is truncated and only the previous word is retained. The input length of the three models is consistent with the bert-base-chinese model. If the length of the encoded word list is insufficient, the rear of the insufficient word list is padded to the maximum length to make the length of all input sequences consistent. At the same time, the corresponding attention mask matrix is generated to identify the padded part and the actual content part in the model to avoid the padding value affecting the calculation of the model.
[0037] Step S2: extract key element information data from the pre-processed work order data.
[0038] After the work order data is preprocessed by word segmentation, encoding, truncation and filling, this embodiment uses the ERNIE3.0 model to classify the labeled case category, dispatch department, and work order topic model, and extracts the responsible party and case time in the work order data based on the Chatglm3-6B large model. The Mgeo model is used to extract the address information mentioned in the work order data, and finally the extraction of key element information data in the work order data is realized.
[0039] The ERNIE 3.0 model plays a role in text intelligent classification tasks such as case classification, department allocation and dispatch, and work order topic extraction. The model structure diagram is shown in the figure below. Figure 2 As shown, the key technologies include:
[0040] (1) Self-supervised contrastive learning pre-training technology for word mixtures: This technology constructs contrast samples based on word mixture granularity segmentation, effectively solving problems in traditional contrastive training. For example, constructing contrast samples by adding, deleting or modifying some words can easily change the semantics of the original text, or using dropout to construct contrast samples tends to believe that sentences of the same or similar length are more semantically similar.
[0041] (2) Word-mixed data-enhanced self-adversarial fine-tuning strategy: In the fine-tuning stage, this strategy introduces word-mixed granularity segmentation based on the adversarial samples obtained through dropout, enhances the randomness of the adversarial samples, and introduces word information, which is a new word information usage scheme.
[0042] (3) Multi-task learning and adversarial training: The ERNIE 3.0 model adopts a multi-task learning strategy to simultaneously learn multiple related tasks, such as text classification and text generation, to improve the generalization ability of the model. In addition, through the adversarial training strategy, an adversarial sample generator is introduced to conduct adversarial training with the model, which enhances the robustness of the model.
[0043] ChatGLM (General Language Model) is a large bilingual language model based on the Transformer architecture, designed for processing Chinese and other bilingual tasks. The model structure is as follows Figure 3 As shown in the figure. The model uses hybrid expert model technology to select the appropriate expert network through dynamic routing, which greatly improves the computing efficiency. At the same time, ChatGLM uses a bidirectional Transformer architecture, which enables the model to consider context information at the same time, enhancing language understanding and dialogue generation capabilities. In addition, the model has been pre-trained in both Chinese and English, has strong cross-language processing capabilities, and supports efficient fine-tuning mechanisms, allowing the model to be quickly adjusted on specific tasks.
[0044] ChatGLM3-6B performs well in natural language processing tasks, especially in bilingual environments. By introducing low-rank adaptation (LoRA) technology to achieve efficient fine-tuning and combining it with advanced knowledge distillation technology, the model is not only more efficient in computing resources, but also shows strong versatility in practical applications, and is suitable for a variety of scenarios such as conversational artificial intelligence, content generation, and intelligent customer service systems.
[0045] like Figure 4 As shown in the figure, the specific structure of the MGeo model (Multi-Modal Geographic Language Mode), where l represents the geographic coordinates longitude and latitude, t represents the text, i represents the i-th geographic object, represents the discretized position feature of oi, that is, the coordinates of the circumscribed rectangle of the geographic object, e represents the encoding result of the geographic object coordinates, and h represents the feature representation after the geocoder. In the multimodal interaction module, CLS and SEP are specific identifiers, indicating the text embedding start character and delimiter.
[0046] It plays an important role in the work order address extraction process. This model is a new method that combines the geocoder and the multimodal interaction module. The geocoder represents the geographic context as a new modality, and the multimodal interaction module combines text and geographic features.
[0047] The geocoder generates a geographic context representation by extracting the relationship between the query / POI geographic location and its surrounding geographic objects. Specifically, the geocoder encodes the ID, shape, map location, relationship type, and relative position of the geographic object into an embedding vector and uses a multi-layer bidirectional transformer to learn the interactions between these features.
[0048] During multimodal pre-training, MGeo learns the interaction between text and geographic context through three tasks: unimodal masked language model (MLM), multimodal MLM, and multimodal masked geographic model (MGM). These tasks are trained sequentially to gradually align text and geographic context.
[0049] The MGeo model can use dual encoder and cross encoder architecture to measure the relevance between query and POI. The dual encoder is used for efficient indexing and is suitable for retrieval tasks; the cross encoder allows fine-grained word-level interactions and is suitable for ranking tasks.
[0050] The extraction of key element information data includes the following sub-steps:
[0051] Step S21: classify the work order data into cases.
[0052] The ERNIE 3.0 model specifically used for case classification is defined as the ERNIE3.0 case classification model. Based on the ERNIE3.0 case classification model, the text features in the work order data are extracted, the case category to which the work order content belongs is intelligently predicted, and a confidence score is given.
[0053] Step S22: Make disposal recommendations based on the work order data.
[0054] The ERNIE 3.0 model specifically used for handling recommendations is defined as the ERNIE3.0 department allocation and dispatch model. Based on the ERNIE3.0 department allocation and dispatch model, the work order data and the text features of the street to which it belongs are extracted, the hotline work order handling department is intelligently recommended, and a confidence score is given.
[0055] Step S23: extract topics from the work order data.
[0056] The ERNIE 3.0 model specifically used for topic extraction is defined as the ERNIE3.0 topic extraction model. Based on the ERNIE3.0 topic extraction model, text features in the work order data are extracted, the problem description topic is analyzed, and a confidence score is given.
[0057] If the topic result matches the pre-set topic library with a similarity higher than 0.9, it is recommended as a standard topic, otherwise it is a smart topic, and the smart topic is optimized to the standard topic library through manual review.
[0058] Step S24: Determine the responsible party based on the work order data.
[0059] Based on the chatglm3 subject extraction model, text features in the work order data are extracted and the relevant responsible parties are obtained through analysis. The ChatGLM3 model is one of the ChatGLM model family, and the chatglm3 subject extraction model refers to using chatglm3 for subject extraction.
[0060] If the similarity between the responsible subject result and the matching subject library is higher than 0.9, it is recommended as a standard subject, otherwise it is an intelligent subject, and the intelligent subject is optimized to the standard subject library through manual review.
[0061] Step S25: extracting the incident address according to the hotline work order.
[0062] The addresses mentioned in the work order data are extracted based on the Mgeo model, and are subdivided according to administrative division information (such as province, city, county, and township information), road network information (such as road names, road numbers, road facilities, etc.), detailed geographic information (such as POI (points of interest), building numbers, household numbers, etc.), and non-geographic information (such as supplementary instructions, erroneous input, etc.).
[0063] Step S26: extracting the time of the incident according to the hotline work order.
[0064] Based on the Chatglm3 model, key time is extracted according to the problem description, and time labels such as early morning, morning, morning, noon, afternoon, dusk, and night are returned to predict the time range for the occurrence of the case.
[0065] Through the above steps, the extraction of key information data in the hotline work order is completed, making the whole process fast, automated and intelligent.
[0066] Step S3: Integrate the extracted key element information data.
[0067] After the key elements of the work order data are extracted, the key element information data is integrated together, for example, in the form of a document or a table.
[0068] Embodiment 2
[0069] like Figure 5 As shown, this embodiment provides a key data extraction system, which specifically includes: a pre-processing unit 501, a key data extraction unit 502, and an integration unit 503.
[0070] The preprocessing unit 501 is used to obtain work order data and preprocess the work order data.
[0071] The work order data includes case type, dispatch department, work order topic, responsible party, incident address and incident time, etc. The preprocessing of the work order data includes word segmentation, encoding, truncation and padding of the work order data text. The preprocessing unit 501 includes the following submodules:
[0072] The list generation module is used to segment the work order data and generate a segmentation list.
[0073] Input the work order data text into the bert-base-chinese tokenizer, split the text into a series of words or subwords, and generate a corresponding token list. Process special characters, punctuation marks, and numbers in the work order to ensure that they can be correctly tokenized.
[0074] The encoding module is used to encode the word list.
[0075] The vocabulary of the bert-base-chinese pre-trained model is used to map each word to a corresponding digital ID. If some words do not exist in the vocabulary, they are usually encoded as special unknown word IDs. For some common words or characters that already exist in the vocabulary, they are directly converted to the corresponding IDs.
[0076] The truncation and padding module is used to truncate and pad the encoded word list.
[0077] First, the maximum input length accepted by each model is determined, and the models are ERNIE3.0 model, Chatglm3-6B large model and Mgeo model.
[0078] If the length of the encoded word list exceeds the maximum input length of the three models, the excess part is truncated and only the previous word is retained. The input length of the three models is consistent with the bert-base-chinese model. If the length of the encoded word list is insufficient, the rear of the insufficient word list is padded to the maximum length to make the length of all input sequences consistent. At the same time, the corresponding attention mask matrix word segmentation is generated to identify the padded part and the actual content part in the model to avoid the padding value affecting the calculation of the model.
[0079] The key data extraction unit 502 is used to extract key element information data from the pre-processed work order data.
[0080] This embodiment uses the ERNIE3.0 model to classify labeled case categories, dispatch departments, and work order topic models, extracts the responsible parties and case times in the work order data based on the Chatglm3-6B large model, and uses the Mgeo model to extract the address information mentioned in the work order data, ultimately realizing the extraction of key element information data in the work order data.
[0081] The key data extraction unit 502 includes the following submodules:
[0082] The case classification module is used to classify work order data. It extracts text features from work order data based on the ERNIE3.0 case classification model, intelligently predicts the case category to which the work order content belongs, and gives a confidence score.
[0083] The disposal recommendation module is used to make disposal recommendations based on work order data. Based on the ERNIE3.0 department allocation and dispatch model, it extracts the work order data and the text features of the street to which it belongs, intelligently recommends the hotline work order disposal department, and gives a confidence score.
[0084] The topic extraction module is used to extract topics from work order data. The text features in the work order data are extracted based on the ERNIE3.0 topic extraction model, and the problem description topics are analyzed and given confidence scores. If the topic result matches the pre-set topic library with a similarity higher than 0.9, it is recommended as a standard topic, otherwise it is an intelligent topic, and the intelligent topic is optimized to the standard topic library through manual review.
[0085] The responsible party determination module determines the responsible party based on the work order data.
[0086] Based on the chatglm3 subject extraction model, the text features in the work order data are extracted and the relevant responsible subjects are obtained through analysis. If the responsible subject result has a matching similarity with the subject library higher than 0.9, it is recommended as a standard subject, otherwise it is an intelligent subject, and the intelligent subject is optimized to the standard subject library through manual review.
[0087] The incident address extraction module is used to extract the incident address based on the hotline work order.
[0088] The addresses mentioned in the work order data are extracted based on the Mgeo model, and are subdivided according to administrative division information (such as province, city, county, and township information), road network information (such as road names, road numbers, road facilities, etc.), detailed geographic information (such as POI (points of interest), building numbers, household numbers, etc.), and non-geographic information (such as supplementary instructions, erroneous input, etc.).
[0089] The incident time extraction module is used to extract the incident time based on the hotline work order.
[0090] Based on the Chatglm3 model, key time is extracted according to the problem description, and time labels such as early morning, morning, morning, noon, afternoon, dusk, and night are returned to predict the time range for the occurrence of the case.
[0091] The integration unit 503 is used to integrate the extracted key element information data.
[0092] Integrate key information data together, for example, into a document or table.
[0093] The present application also provides a computer storage medium, wherein the computer storage medium stores computer instructions, and when the computer instructions are called, they are used to execute the key data extraction method.
[0094] The embodiments disclosed in the present application provide a computer-readable storage medium, in which computer program instructions are stored. When the computer program instructions are executed on a computer, the computer executes the above-mentioned key data extraction method.
[0095] An embodiment of the present application provides a processor for processing the above-mentioned key data extraction method.
[0096] In the embodiment of the present application, the processor may be an integrated circuit chip having the ability to process signals. The processor may be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0097] The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and completes the steps of the above method in combination with its hardware.
[0098] The storage medium may be a memory, which may be, for example, a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memory.
[0099] Among them, the non-volatile memory can 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 can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct RAM bus random access memory (DRRAM).
[0100] This application has the following beneficial effects:
[0101] (1) This application uses artificial intelligence technology to conduct a comprehensive, accurate, and efficient analysis of the work order content and extract key elements, solving the problems of low efficiency, strong subjectivity, and proneness to errors in traditional manual processing.
[0102] (2) This application unifies, normalizes and standardizes the extraction of hotline work order content, solving the problems of confusion and non-standardization in the process of extracting work order elements.
[0103] (3) The key data extraction method proposed in this application can be deployed in an accelerated manner on multiple device terminals, greatly improving the efficiency of information collection and work order processing, and solving the problems of low efficiency and lack of convenience of traditional manual processing.
[0104] Although the present application is described with reference to examples, this is for illustrative purposes only and is not intended to limit the present application, and changes, additions and / or deletions to the embodiments may be made without departing from the scope of the present application.
[0105] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A key data extraction method, characterized in that: The following steps are involved: Obtain work order data and pre-process the work order data; Extract key element information data from pre-processed work order data; Integrate the extracted key element information data.
2. The key data extraction method according to claim 1, characterized in that: Work order data includes case type, dispatch department, work order topic, responsible party, incident address and incident time data.
3. The key data extraction method according to claim 1, characterized in that: The preprocessing of work order data includes the following sub-steps: Perform word segmentation on the work order data and generate a word segmentation list; Encode the word list; Truncate and pad the encoded word list.
4. The key data extraction method as claimed in claim 3, characterized in that: Truncating and padding the encoded word segmentation list includes determining the maximum input length that the model can accept. If the length of the encoded word segmentation list is less than the maximum input length that the model can accept, padding the length of the word segmentation list; if the length of the encoded word segmentation list exceeds the maximum input length that the model can accept, truncating the excess part.
5. The key data extraction method according to claim 1, characterized in that: The extraction of key element information data includes extracting text features from work order data based on the ERNIE3.0 model, intelligently predicting the case category to which the work order content belongs, and providing a confidence score.
6. A key data extraction system, characterized in that: Specifically include: Pre-processing unit, key data extraction unit and integration unit; A preprocessing unit, used to obtain work order data and preprocess the work order data; A key data extraction unit is used to extract key element information data from the pre-processed work order data; The integration unit is used to integrate the extracted key element information data.
7. The key data extraction system as claimed in claim 6, characterized in that: The work order data obtained in the pre-processing unit includes case type, dispatch department, work order topic, responsible party, incident address and incident time data.
8. The key data extraction system as claimed in claim 6, characterized in that: The preprocessing unit performs preprocessing of work order data, including the following sub-steps: Perform word segmentation on the work order data and generate a word segmentation list; Encode the word list; Truncate and pad the encoded word list.
9. The key data extraction system as claimed in claim 8, characterized in that: The preprocessing unit truncates and fills the encoded word segmentation list, including determining the maximum input length that the model can accept. If the length of the encoded word segmentation list is less than the maximum input length that the model can accept, the length of the word segmentation list is filled. If the length of the encoded word segmentation list exceeds the maximum input length that the model can accept, the excess part is truncated.
10. The key data extraction system as claimed in claim 6, characterized in that: The key data extraction unit extracts key element information data, including extracting text features in the work order data based on the ERNIE3.0 model, intelligently predicting the case category to which the work order content belongs, and giving a confidence score.