Intelligent question answering system based on AI

Abstract

The invention discloses an intelligent question-answering system based on AI, and the system comprises a document recognition and extraction unit which is used for recognizing the source type and format of a document, and carrying out the auxiliary splitting of a Document document processing model according to different types and formats; the knowledge base data splitting and archiving unit is used for constructing a hierarchical distributed storage framework and carrying out data splitting and archiving according to different overall framework, classification system and authority configuration of the stored data; the searcher word segmentation optimization unit adopts AI Search, and endows enterprise vocabularies with priorities and weights through setting of indexes and an indexer, so that the enterprise vocabularies are preferentially segmented in the word segmentation process; the AI model training unit is used for selecting an adaptive AI basic model architecture according to an enterprise business target and question and answer requirements; and the intelligent document management unit is used for setting intra-department, cross-department and project knowledge permission groups on the basis of enterprise organization architecture and project team cooperation requirements, and ensuring question-answer interaction and document viewing and modifying operations of employees through an identity authentication token and digital signature encryption transmission technology.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of intelligent question answering systems, and particularly relates to an intelligent question answering system based on AI. BACKGROUND

[0002] The AI intelligent question answering system is suitable for a business environment with rapid iteration of business, and internal knowledge accumulation and efficient circulation of enterprises are crucial. After years of operation, enterprises have accumulated a large amount of knowledge, covering product research and development details (from raw material characteristics, design innovation points to manufacturing process optimization), market sales strategies (different regional customer preferences, successful marketing case review, competitor analysis insight), human resource specifications (recruitment process skills, employee training system essence, performance evaluation details) and administrative management processes (office approval process evolution) and many other categories.

[0003] The current common AI intelligent document system cannot perform knowledge retrieval and semantic analysis to quickly answer the content of the questions and queries input by employees or users. SUMMARY

[0004] In view of the problem that the existing AI intelligent document system cannot perform knowledge retrieval and semantic analysis to quickly answer the content of the questions and queries input by employees or users, the application provides an intelligent question answering system based on AI.

[0005] To achieve the above technical purpose, the technical solution adopted by the application is as follows:

[0006] An intelligent question answering system based on AI, comprising a document recognition and extraction unit, a knowledge base data splitting and archiving unit, a retriever word segmentation optimization unit, an AI model training unit and an intelligent document management unit.

[0007] The document recognition and extraction unit is used to recognize the source type and format of the document, and to split the document according to different types and formats with the aid of the Document intelligence document processing model; through keyword extraction, key sentence recognition, paragraph semantic clustering and key-value pair recognition means, the knowledge unit is defined to ensure that the split document data is independent and complete and is related to each other, meeting the subsequent retrieval and question answering logic requirements.

[0008] The knowledge base data splitting and archiving unit constructs a hierarchical distributed storage architecture, and the bottom layer adopts a high-performance SQLServer relational database to store the split raw data, and the data is split and archived according to the overall architecture, classification system and permission configuration of the stored data.

[0009] The retriever word segmentation optimization unit adopts AI Search, and through the setting of index and indexer, the enterprise vocabulary priority and weight are given to make it preferentially segmented in the word segmentation process.

[0010] The AI model training unit selects an AI base model architecture according to enterprise business objectives and question and answer demand characteristics, filters high-value enterprise data according to business importance, consultation frequency and knowledge timeliness indicators from the split and archived knowledge base for model training fine-tuning, and divides the training set, validation set and test set in proportion; ensures data quality and distribution rationality, and adapts the model iteration optimization rhythm.

[0011] The intelligent document management unit sets up departmental, cross-departmental and project knowledge permission groups based on enterprise organizational structure and project team collaboration needs, and ensures employee operations in question and answer interaction and document viewing and modification through identity authentication token and digital signature encryption transmission technology.

[0012] Further, the document recognition and extraction unit performs preliminary document integrity verification or marks abnormal data during the uploading process; so as to facilitate subsequent verification and repair.

[0013] For different business department documents, they are classified and registered according to preset rules, and part and classification directory lists are constructed. This facilitates tracking, tracing and management.

[0014] Further, the document integrity verification uses a hash algorithm to verify the integrity of the file by calculating the unique identifier (hash value) of the file; common hash algorithms include MD5, SHA-1, SHA-256, etc.

[0015] After copying the file, calculate the hash value of the target file and compare it with the hash value of the original file;

[0016] If the hash values are the same, the file contents are the same;

[0017] If they are different, the file contents may have been tampered with and are marked as abnormal data.

[0018] Further, the knowledge base data splitting and archiving unit splits data according to business domain knowledge architecture and stores it into the knowledge base as fine-grained knowledge units according to business classification. For example, in the human resources knowledge category, around the recruitment, training, performance, employee relations, etc. Plates, the corresponding documents are disassembled into specific job recruitment requirements details, training course chapter knowledge points, performance evaluation index interpretation, etc.

[0019] Further, the indexer in the retriever word segmentation optimization unit is used to collect and organize high-frequency professional terms, industry abbreviations, department-specific appellations, and common expressions used in various business scenarios within the enterprise, forming an enterprise-specific glossary. The meanings of the words, usage contexts and synonym relationships are audited and calibrated to ensure the accuracy of the word library.

[0020] Further, the retriever word segmentation optimization unit uses semantic error correction function while word segmentation, automatically identifies and corrects the question text errors, improper word use problems based on predefined grammar rules, common error combination patterns, and restores the correct semantic expression before word segmentation retrieval.

[0021] Further, the AI model training unit adopts incremental fine-tuning and regular training method, continuously integrates new high-quality data into the training process as the enterprise knowledge updates and business expands, and retraces the full data for retraining regularly.

[0022] Further, the intelligent document management unit divides ordinary employees, department supervisors, senior managers, system administrators and other roles by building a multi-level user role system, corresponding to different operation permissions (ordinary employees can only read common knowledge, department supervisors can edit department knowledge, and administrators can control globally) and knowledge access levels (basic business processes, deep analysis reports, and sensitive core data are limited in turn).

[0023] Compared with the prior art, the present application has the following beneficial effects:

[0024] It brings beneficial effects in knowledge management and utilization, semantic understanding and interactive experience, office collaboration and cost benefit, system operation and security, user experience and employee empowerment, etc. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The present application provides an overall structure block diagram of an AI-based intelligent question answering system. DETAILED DESCRIPTION

[0026] In order to facilitate the understanding of those skilled in the art, the present application will be further described below in conjunction with the embodiments and drawings.

[0027] As shown in the drawings, Figure 1 The present application provides an AI-based intelligent question answering system, which includes a document recognition and extraction unit, a knowledge base data splitting and archiving unit, a retriever word segmentation optimization unit, an AI model training unit and an intelligent document management unit.

[0028] The document recognition and extraction unit is used to identify the document source type and format, and to assist in splitting according to different types and formats using the Document intelligence document processing model; by means of keyword extraction, key sentence identification, paragraph semantic clustering and key-value pair identification means, the knowledge unit is defined to ensure that the split document data is independent and complete and is mutually related, and meets the subsequent retrieval and question and answer logic requirements.

[0029] The knowledge base data splitting and archiving unit constructs a hierarchical distributed storage architecture, the bottom layer uses a high-performance SQLServer relational database to store the split raw data, and the data is split and archived according to the overall architecture, classification system and permission configuration of the stored data;

[0030] The retriever word segmentation optimization unit uses AI Search to set up indexes and indexers, and gives priority and weight to enterprise vocabulary to make it preferentially segmented in the word segmentation process.

[0031] The AI model training unit selects an appropriate AI base model architecture according to the business objectives and question and answer requirements of the enterprise, filters high-value enterprise data for model training and fine-tuning from the split and archived knowledge base according to business importance, consultation frequency and knowledge timeliness indicators, and divides the training set, validation set and test set in proportion; ensures data quality and reasonable distribution, and adapts to the model iteration and optimization rhythm.

[0032] The intelligent document management unit sets up intra-department, cross-department and project knowledge permission groups based on the enterprise organizational structure and project team collaboration requirements, and ensures that employees can interact and modify documents through identity authentication tokens and digital signature encryption transmission technology.

[0033] The document recognition and extraction unit performs preliminary document integrity checking or marks abnormal data during the uploading process, so as to facilitate subsequent verification and repair.

[0034] According to different business departments, the documents are classified and registered according to preset rules to construct part and classification directory lists, which are convenient for tracking, tracing and management.

[0035] The document integrity check uses a hash algorithm to verify the integrity of the file by calculating its unique identifier (hash value); common hash algorithms include MD5, SHA-1, SHA-256, etc.

[0036] After copying the file, calculate the hash value of the target file and compare it with the hash value of the original file.

[0037] If the hash values are the same, the file contents are the same.

[0038] If they are different, the file contents may have been tampered with and are marked as abnormal data.

[0039] The knowledge base data splitting and archiving unit splits data into fine-grained knowledge units according to business classification based on business domain knowledge architecture and stores the data into the knowledge base. For example, in the human resource knowledge category, around the recruitment, training, performance, employee relationship, etc. board, the corresponding documents are disassembled into specific post recruitment requirements details, training course chapter knowledge points, performance evaluation index interpretation, etc.

[0040] The indexer in the retriever word segmentation optimization unit is used to collect and arrange the high-frequency professional terms, industry abbreviations, department-specific appellations, and common expressions used in various business scenarios within the enterprise to form an enterprise-specific glossary. The meanings of the glossary, usage context, and synonym relationships are audited and calibrated to ensure the accuracy of the glossary.

[0041] The retriever word segmentation optimization unit uses semantic error correction function while segmenting words, based on predefined grammar rules and common error combination patterns, to automatically identify and correct the problems of question text errors, improper word usage, and restore the correct semantic expression before re-segmenting and retrieving.

[0042] The AI model training unit adopts incremental fine-tuning and regular training method, continuously integrates new high-quality data into the training process as the enterprise knowledge updates and business expands, and regularly retraces the full data for retraining. On the basis of consolidating the general knowledge understanding of the model, it strengthens the mastery of dynamic changing knowledge of the enterprise, and ensures the long-term performance stability and knowledge adaptability of the model.

[0043] The intelligent document management unit divides ordinary employees, department supervisors, senior managers, system administrators, etc. through the construction of a multi-level user role system, corresponding to different operation permissions (ordinary employees can only read common knowledge, department supervisors can edit department knowledge, and administrators can globally control) and knowledge access levels (basic business processes, deep analysis reports, and sensitive core data are limited in turn).

[0044] Compared with the prior art, the present application has the following beneficial effects:

[0045] It brings beneficial effects in knowledge management and utilization, semantic understanding and interactive experience, office collaboration and cost benefit, system operation and security, user experience and employee empowerment, etc. For example, in the knowledge management aspect, it emphasizes breaking down knowledge silos and realizing dynamic updating and inheritance of knowledge; in the semantic understanding aspect, it highlights accurate semantic analysis and smooth multi-round dialogue interaction; in the office collaboration aspect, it focuses on integration with office software and reduces costs to improve efficiency, etc.

[0046] The AI-based intelligent question and answer system provided in the application is described in detail. The description of the specific embodiments is only used to help understand the method and its core idea of the application. It should be pointed out that for ordinary skilled persons in the art, without departing from the principles of the application, the application can be improved and modified, and these improvements and modifications also fall within the protection scope of the claims of the application.

Claims

1. An AI-based intelligent question-answering system, characterized in that: It includes document recognition and extraction unit, knowledge base data splitting and archiving unit, search engine word segmentation optimization unit, AI model training unit and intelligent document management unit; Document recognition and extraction unit, used to identify the source type and format of documents, and use the Document Intelligence document processing model to assist in splitting according to different types and formats; The knowledge base data is split and archived into units, with a layered distributed storage architecture. The underlying layer uses a high-performance SQL Server relational database to store the split raw data. Data is split and archived according to the overall storage data architecture, classification system, and permission configuration. The search engine word segmentation optimization unit uses AI Search to assign priority and weight to enterprise words through indexing and indexer settings, giving them priority in word segmentation. The AI ​​model training unit selects an appropriate AI basic model architecture based on the company's business objectives and question-answering needs. It then selects high-value enterprise data from the archived knowledge base for model training and fine-tuning based on business importance, consultation frequency, and knowledge timeliness, and divides the data into training, validation, and test sets in proportion. The intelligent document management unit sets up departmental, cross-departmental, and project knowledge permission groups based on the enterprise organizational structure and project team collaboration needs. It ensures employees' ability to interact with others in questions and answers and view and modify documents through identity authentication tokens and digital signature encryption transmission technology.

2. The AI-based intelligent question-answering system according to claim 1, characterized in that: The document recognition and extraction unit performs preliminary document integrity checks or marks abnormal data during the upload process; For documents from different business departments, classify and register them according to preset rules, and build a list of partial and classified directories.

3. The AI-based intelligent question-answering system according to claim 2, characterized in that: Document integrity verification uses a hash algorithm to verify the integrity of the file by calculating the hash value of the file; After copying the file, calculate the hash value of the target file and compare it with the hash value of the original file; If the hash values ​​are the same, the file contents are the same; If they are different, the file content may have been tampered with and marked as abnormal data.

4. The AI-based intelligent question-answering system according to claim 3, characterized in that: The knowledge base data is split into archiving units according to the business domain knowledge architecture, and the data is split into fine-grained knowledge units according to business classification and stored in the knowledge base.

5. The AI-based intelligent question-answering system according to claim 4, characterized in that: The indexer in the search engine's word segmentation optimization unit is used to collect and organize frequently used professional terms, industry abbreviations, department-specific titles, and common expressions in various business scenarios within the enterprise to form an enterprise-specific vocabulary.

6. The AI-based intelligent question-answering system according to claim 5, characterized in that: While performing word segmentation, the search engine's word segmentation optimization unit uses a semantic error correction function to automatically identify and correct grammatical errors and inappropriate word usage in the question text based on predefined grammatical rules and common error collocation patterns, and then performs word segmentation retrieval after restoring the correct semantic expression.

7. The AI-based intelligent question-answering system according to claim 6, characterized in that: The AI ​​model training unit adopts incremental fine-tuning and regular training methods. As the company's knowledge is updated and its business expands, it continuously incorporates new high-quality data into the training process and regularly retraces the full data for retraining.

8. The AI-based intelligent question-answering system according to claim 7, characterized in that: The intelligent document management unit builds a multi-level user role system, dividing ordinary employees, department heads, senior managers and system administrators, corresponding to different operating permissions and knowledge access levels.

Images