Text structured extraction method and system of office automation industry depending on large model

Abstract

The invention relates to the technical field of natural language processing, in particular to an office automation industry text structured extraction method and system depending on a large model. Identifying a text type; key fields are extracted; a continuous learning mechanism; updating the system by the dynamic template; the text structured extraction method and system for the office automation industry depending on the large model have the beneficial effects that the time for manual reading and data entry is remarkably shortened through automatic type identification and structured extraction, so that the processing efficiency of text data in an enterprise office automation system is improved; the large model is used for deep learning and feature extraction, the system can more accurately identify text types and extract key fields, human errors are reduced, and the accuracy of text analysis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a text structured extraction method and system for the office automation industry that relies on a large model. Background Art

[0002] Traditional text processing methods usually rely on manual reading and understanding, which is not only inefficient but also easily affected by subjective judgment, resulting in inaccurate and inconsistent information extraction. In existing technologies, the automated processing of text content mainly faces the following two difficulties:

[0003] 1. Diversity of text types: There are many types of texts generated in enterprises, such as leave applications, expense reports, meeting minutes, etc. Each type of text has its own specific structure and key information fields.

[0004] 2. Difficulty in extracting key fields: Traditional text processing technology has difficulty in accurately identifying key information fields in text, especially when the text format is irregular or the information is not clearly expressed. Summary of the invention

[0005] The purpose of the present invention is to provide a text structured extraction method and system for the office automation industry that relies on a large model, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a text structured extraction method for the office automation industry that relies on a large model, the method comprising the following steps:

[0007] Model building module;

[0008] Text type identification;

[0009] Key field extraction;

[0010] Continuous learning mechanism;

[0011] Dynamic template update system.

[0012] Preferably, the specific operations of model building include:

[0013] A large model with strong semantic understanding ability is selected for training to learn the intrinsic features and patterns of different text types.

[0014] Preferably, the specific operations of text type recognition include:

[0015] The input text is preprocessed, including word segmentation, stop word removal, and part-of-speech tagging. The trained model is then used to extract high-dimensional feature representations of the text and predict the most likely type of the text.

[0016] Preferably, the specific operations of extracting key fields include:

[0017] Based on the results of text type recognition, a pre-designed prompt template is selected to extract key fields. The template defines the key fields required for each text type.

[0018] Preferably, the specific operations of the continuous learning mechanism include:

[0019] Allow the model to continuously learn and evolve through continuous exposure to new data, using incremental learning techniques to gradually absorb new knowledge, and collect new text data from multiple sources through the data acquisition module;

[0020] The specific operations of the dynamic template update system include:

[0021] By combining continuous learning, accurate recognition, and dynamic template updates, it provides a powerful, flexible, and adaptive text processing solution to meet ever-changing text analysis needs and improve processing efficiency and accuracy.

[0022] A text structured extraction system for the office automation industry that relies on a large model, the system consisting of a model building module, a text type recognition module, a key field extraction module, a continuous learning mechanism, and a dynamic template update system;

[0023] Model building module, used for model building module;

[0024] A text type recognition module, used for text type recognition;

[0025] Key field extraction module, used for key field extraction;

[0026] Continuous learning mechanism, used for continuous learning mechanism;

[0027] Dynamic template update system, used for dynamic template update system.

[0028] Preferably, the model building module selects a large model with strong semantic understanding ability for training to learn the intrinsic features and patterns of different text types.

[0029] Preferably, the text type identification module performs preprocessing on the input text, including word segmentation, stop word removal, and part-of-speech tagging, and then uses the trained model to extract high-dimensional feature representation of the text and predict the most likely type of the text.

[0030] Preferably, the key field extraction module selects a pre-designed prompt template to extract key fields based on the result of text type recognition, and the template defines the key fields required for each text type.

[0031] Preferably, the continuous learning mechanism allows the model to continuously learn and evolve by continuously being exposed to new data, using incremental learning techniques to gradually absorb new knowledge, and collect new text data from multiple sources through a data acquisition module;

[0032] The dynamic template update system, by combining continuous learning, precise recognition and dynamic template updating, provides a powerful, flexible and adaptive text processing solution to meet the ever-changing text analysis needs and improve processing efficiency and accuracy.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] The text structured extraction method and system for the office automation industry that relies on a large model proposed in the present invention significantly reduces the time for manual reading and data entry through automatic type recognition and structured extraction, thereby improving the processing efficiency of text data in the enterprise office automation system;

[0035] By using large models for deep learning and feature extraction, the system can more accurately identify text types and extract key fields, reducing human errors and improving the accuracy of text analysis;

[0036] Through the continuous learning mechanism, the system can continuously access and learn new data, so that the model can adapt to new text types and language variants, maintaining long-term prediction accuracy and adaptability;

[0037] The system can automatically process text, reducing the reliance on a large number of manpower, thereby reducing the operating costs of the enterprise, especially when the amount of text data is huge;

[0038] The dynamic template update system allows users to adjust the extraction templates based on feedback, making the system more personalized and user-friendly, improving end-user satisfaction and system usability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Example 1 Please refer to Figure 1The present invention provides a technical solution: a text structured extraction method for the office automation industry that relies on a large model, the method comprising the following steps:

[0042] Model construction: select a large model with strong semantic understanding capabilities for training to learn the intrinsic features and patterns of different text types.

[0043] Text type identification: perform preprocessing on the input text, including word segmentation, stop word removal, and part-of-speech tagging, and then use the trained model to extract the high-dimensional feature representation of the text and predict the most likely type of the text.

[0044] Key field extraction: Based on the results of text type recognition, select a pre-designed prompt template to extract key fields. The template defines the key fields required for each text type.

[0045] Continuous learning mechanism; allows the model to continuously learn and evolve through continuous exposure to new data, adopts incremental learning techniques, gradually absorbs new knowledge, and collects new text data from multiple sources through the data acquisition module.

[0046] The dynamic template update system, by combining continuous learning, precise recognition and dynamic template updating, provides a powerful, flexible and adaptive text processing solution to meet the ever-changing text analysis needs and improve processing efficiency and accuracy.

[0047] Embodiment 2, based on embodiment 1, proposes a text structured extraction system for the office automation industry that relies on a large model, the system consists of a model building module, a text type recognition module, a key field extraction module, a continuous learning mechanism and a dynamic template update system;

[0048] Model building module: select a large model with strong semantic understanding capabilities for training and learn the intrinsic features and patterns of different text types.

[0049] Text type identification module: The system first performs preprocessing on the input text, including word segmentation, stop word removal, part-of-speech tagging, etc. Then, it uses the trained model to extract the high-dimensional feature representation of the text and predict the most likely type of the text.

[0050] Key field extraction module: Based on the results of text type recognition, the system will select a pre-designed prompt template to extract key fields. These templates define the key fields required for each text type, such as leave type, estimated duration, start time, etc.

[0051] The technical solution of the present invention mainly focuses on the process of using the existing open source large model to perform text type recognition and structured extraction. Therefore, the model building stage will not be introduced in detail. Although there are many cases of using large models to complete text structured extraction, there are still many challenges in applying it to actual production. These challenges mainly include but are not limited to the following aspects:

[0052] (1) Limited generalization ability: Although large models perform well on specific tasks, they may face challenges in generalizing to new or unseen text types. Large models are usually trained on large-scale datasets that may not cover all possible text structures and language variants. Therefore, when the model encounters text that differs greatly from the training data, its structured extraction effect may be greatly reduced. In addition, large models may be too focused on specific patterns in the training data and ignore the deep semantics of the text, which limits the applicability of the model to diverse text processing tasks.

[0053] (2) Strong data dependence: The performance of large models depends heavily on a large amount of high-quality annotated data. In the text structured extraction task, various types of text need to be thoroughly annotated, which is not only time-consuming and labor-intensive, but also costly. If the dataset is biased or mislabeled, the model may learn the wrong pattern, resulting in a decrease in the accuracy of the extraction results. In addition, it is particularly difficult to obtain sufficient training data for texts in certain specific fields or niche languages.

[0054] In response to the above two problems, the technical solutions proposed by this patent include the following points:

[0055] 1. Continuous learning mechanism

[0056] The continuous learning mechanism is a key component of the model optimization and adjustment phase. This mechanism allows the model to continuously learn and evolve through continuous exposure to new data, thereby maintaining its predictive accuracy. To achieve this, the system will use incremental learning techniques, which means that the model can gradually absorb new knowledge without retraining the entire dataset. In addition, the system will design a data acquisition module that can continuously collect new text data from multiple sources (such as user-uploaded documents, real-time news, social media updates, etc.) to ensure the diversity and timeliness of the data.

[0057] 2. Accurate text type recognition capability

[0058] Text type recognition technology is a key technology in the field of natural language processing, which can automatically classify the input text data into predefined categories. The system involved in this patent adopts a series of preprocessing steps and advanced machine learning models in the text type recognition stage to achieve efficient and accurate text classification. The system first receives the text data input by the user and then performs a series of preprocessing operations. These preprocessing steps include but are not limited to word segmentation, that is, splitting the continuous text string into meaningful units such as words or phrases; removing stop words, that is, deleting those words that frequently appear in the text but contribute little to the meaning of the text, such as "de", "shi", etc.; and part-of-speech tagging, that is, identifying the part of speech of each word in the text, such as nouns, verbs, etc. These preprocessing steps are crucial for subsequent feature extraction and model training because they can help the system better understand the text content.

[0059] After preprocessing, the system uses a trained model to extract the high-dimensional feature representation of the text. This model is trained on a large amount of labeled data and can capture the semantic information and structural features in the text. Through deep learning techniques such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), or Transformer models, the system can convert the original text into a high-dimensional vector that contains the key information of the text. Finally, the system uses this high-dimensional feature representation to predict the most likely type of the text.

[0060] 3. Dynamic Template Update System

[0061] To adapt to the diversity and complexity of text content, the present invention proposes a dynamic template update system. This system can automatically identify the key fields and structures in the text without manual intervention. When the system encounters a new text type, it will prompt the user that a new text type has appeared, request the customer to specify the keywords to be extracted, and give several similar texts for the user to manually mark the key fields. According to the user's feedback, the system can automatically generate or adjust the prompt template to meet the requirements of key field extraction. For example, when the system recognizes the new type of text content "Query the remaining number of annual leave days of Teacher Zhang San", the system prompts the user [Whether to add "Query holiday days"] and gives the recommended extraction field format.

[0062] {

[0063] "method":"getBackList",

[0064] "parameter":{

[0065] "roleName":"Zhang San"

[0066] "type":"Annual leave"

[0067] }

[0068] }

[0069] Users can make targeted adjustments based on the results returned by the large model, and after confirmation, a prompt keyword extraction template for this type of text will be automatically generated.

[0070] In general, the system of this patent provides a powerful, flexible and adaptive text processing solution by combining continuous learning, precise recognition and dynamic template updates, which can meet the ever-changing text analysis needs and improve processing efficiency and accuracy.

[0071] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A text structured extraction method for the office automation industry that relies on a large model, characterized by: The method comprises the following steps: Model building; Text type identification; Key field extraction; Continuous learning mechanism; Dynamic template update system.

2. The text structure extraction method for office automation industry relying on a large model according to claim 1 is characterized by: The specific operations of model construction include: A large model with strong semantic understanding ability is selected for training to learn the intrinsic features and patterns of different text types.

3. The text structure extraction method for office automation industry relying on a large model according to claim 1 is characterized by: The specific operations of text type recognition include: The input text is preprocessed, including word segmentation, stop word removal, and part-of-speech tagging. The trained model is then used to extract high-dimensional feature representations of the text and predict the most likely type of the text.

4. The text structure extraction method for office automation industry relying on a large model according to claim 1 is characterized by: The specific operations of key field extraction include: Based on the results of text type recognition, a pre-designed prompt template is selected to extract key fields. The template defines the key fields required for each text type.

5. The text structure extraction method for office automation industry relying on a large model according to claim 1 is characterized by: The specific operations of the continuous learning mechanism include: Allow the model to continuously learn and evolve through continuous exposure to new data, using incremental learning techniques to gradually absorb new knowledge, and collect new text data from multiple sources through the data acquisition module; The specific operations of the dynamic template update system include: By combining continuous learning, accurate recognition, and dynamic template updates, it provides a powerful, flexible, and adaptive text processing solution to meet ever-changing text analysis needs and improve processing efficiency and accuracy.

6. A text structure extraction system for office automation industry relying on a large model according to the text structure extraction method for office automation industry relying on a large model according to any one of claims 1 to 5, characterized in that: The system consists of a model building module, a text type recognition module, a key field extraction module, a continuous learning mechanism and a dynamic template update system; Model building module, used for model building module; A text type recognition module, used for text type recognition; Key field extraction module, used for key field extraction; Continuous learning mechanism, used for continuous learning mechanism; Dynamic template update system, used for dynamic template update system.

7. A text structured extraction system for office automation industry that relies on a large model according to claim 6, characterized in that: The model building module selects a large model with strong semantic understanding ability for training and learns the intrinsic features and patterns of different text types.

8. The text structure extraction system for office automation industry relying on a large model according to claim 6, characterized in that: The text type recognition module performs preprocessing on the input text, including word segmentation, stop word removal, and part-of-speech tagging, and then uses the trained model to extract the high-dimensional feature representation of the text and predict the most likely type of the text.

9. The text structure extraction system for office automation industry relying on a large model according to claim 6, characterized in that: The key field extraction module selects a pre-designed prompt template to extract key fields based on the result of text type recognition, and the template defines the key fields required for each text type.

10. The text structure extraction system for office automation industry relying on a large model according to claim 6, characterized in that: The continuous learning mechanism allows the model to continuously learn and evolve by continuously being exposed to new data, using incremental learning techniques to gradually absorb new knowledge, and collect new text data from multiple sources through the data acquisition module; The dynamic template update system, by combining continuous learning, precise recognition and dynamic template updating, provides a powerful, flexible and adaptive text processing solution to meet the ever-changing text analysis needs and improve processing efficiency and accuracy.

Images