Business filling scene data acquisition method and system, storage medium and electronic equipment

Abstract

The invention discloses a business filling scene data acquisition method and system, a storage medium and electronic equipment, and the method comprises the steps: firstly storing filling historical data and domain knowledge into a vector database, and according to the current input of a user, searching related information, and combining the current input to generate a filling suggestion through a generation type large model; multi-modal information of the multi-end filling system is obtained, semantic understanding and information extraction are completed through a large model, and the multi-modal information is converted into structured data; mapping the structured data to a specified field by means of a field mapping rule; a data verification rule is designed, data verification is corrected through a rule engine or large model dynamic logic verification, and a verification result and a prompt suggestion are output; and finally, storing the processed and verified data, and providing data support for the user. According to the invention, the data acquisition efficiency is improved; the intention of the user can be predicted, corresponding suggestions can be provided, and the legality and accuracy of data are ensured.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data collection, and particularly relates to a business reporting scene data collection method and system, a storage medium and an electronic device. BACKGROUND

[0002] At present, in the processes of government affairs processing in the government field, medical business processing in the medical field, and sales business processing in the sales field, data in the business processing process needs to be collected and reported. For example, government questionnaires need to collect and organize the questionnaire answers provided by citizens; administrative approval application processes need to extract key information and required data according to the application description of enterprises or individuals. For another example, in the medical field, medical record information needs to be extracted according to voice or text disease description. For another example, in the sales field, sales information needs to be extracted according to voice or text description.

[0003] In the traditional technology, manual data input collection is adopted, and manual input data is prone to error, and the technical requirements for users are high. Moreover, the workload of manual input and data organization is large, resulting in low efficiency of data collection. Moreover, many projects lack intelligent analysis technology in the reporting process, and cannot analyze the reporting intention of users, cannot effectively provide reporting suggestions to users, and have a lot of repetitive work. Therefore, it is of practical application significance to provide a high-efficiency data collection technical scheme in a business reporting scene. SUMMARY

[0004] Therefore, the present application provides a business reporting scene data collection method and system, a storage medium and an electronic device to improve the efficiency and accuracy of reporting scene data collection, and solve the problems of low efficiency, error-prone, and inability to provide reasonable reporting suggestions to users in the traditional manual input.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme: in the first aspect, a business reporting scene data collection method is provided, comprising:

[0006] Based on the retrieval enhancement generation model, the reporting history data and the context information of the user are stored in the vector database in combination with the reporting history data and the preset domain knowledge; according to the current input of the user, relevant information is retrieved from the vector database as the context; the relevant information and the current input of the user are provided to a generative large language model to generate reporting suggestions of a multi-terminal reporting system;

[0007] Multi-modal information input by the user through a multi-terminal reporting system interface is obtained, the multi-modal information including text, voice, image or document; a large language model is used for end-to-end semantic understanding and information extraction of the multi-modal information to convert into structured data;

[0008] Defining a data model and a field mapping rule, and mapping the structured data converted from the multi-modal information to specified data fields through the field mapping rule;

[0009] According to business requirements and data quality requirements, a data verification rule is designed, which at least includes rule engine-based verification or dynamically logical verification based on a generative large language model, and the data filled by the user is verified and corrected to obtain a verification result and an interpretable error prompt or correction suggestion;

[0010] The processed and verified data is stored for the user to perform data analysis, report generation, and decision support.

[0011] As an optimal solution for the business filling scene data collection method, the filling behavior model of the user is established by analyzing the historical filling data of the user, extracting the category, time, and content key features of the filling, and combining the geographic location and device information of the user.

[0012] As an optimal solution for the business filling scene data collection method, the generative large language model is fine-tuned by setting instructions of the business scene, so that the generative large language model generates filling content or suggestions in accordance with the retrieved context information and user input, which conforms to the business logic.

[0013] Processing the multi-modal information specifically includes: when the input is an image or a document, optical character recognition technology and visual language models are used to extract text and layout information therefrom, and the generative large language model is used for processing.

[0014] As an optimal solution for the business filling scene data collection method, it is used for sales performance filling, specifically including:

[0015] The sales information filled by the user is obtained, the historical filling data is analyzed, and specified features are extracted, including the category, time, and sales amount of the filling, and the filling intention of the user is predicted by combining the geographic location and device information of the user.

[0016] According to the filling intention and the analysis result of the historical data, a sales filling suggestion is generated, including automatically filling the known last month sales amount, customer group option information, and providing a sales amount range according to the historical data statistics, and providing a target area option according to the target market filled by the user.

[0017] As an optimal solution for the business filling scene data collection method, the morphological analysis performs word segmentation and part-of-speech tagging operations on the multi-modal information input by the user to obtain basic information of each word in the multi-modal information.

[0018] The syntax analysis processes the structure and syntax relationship of the multi-modal information sentence input by the user, and obtains semantic information of the multi-modal information.

[0019] The semantic understanding of the multi-modal information input by the user includes:

[0020] The generative large language model realizes entity recognition, relationship extraction and semantic role labeling in the input information through context learning or fine-tuning, and generates structured data according to a preset output format.

[0021] The semantic role labeling technology is used to label the predicate and argument in the multi-modal information to describe the semantic role of different components in the sentence in the multi-modal information.

[0022] As an optimal solution of the business reporting scene data collection method, converting the multi-modal information into structured data includes:

[0023] Through the API interface of natural language processing, the natural language description of the user is called and processed, and the processing result is returned to the calling party.

[0024] The structured data obtained after natural language processing is converted into a specified storage form to be stored in the database.

[0025] As an optimal solution of the business reporting scene data collection method, it is used for medical record collection and reporting, and specifically includes:

[0026] Obtain the patient's identity information and chief complaint information, the patient's identity information includes the patient's name and age, and the chief complaint information includes the disease information and the duration of the disease;

[0027] Preprocess the patient's identity information and the chief complaint information, preprocessing includes removing punctuation and stop words;

[0028] Perform entity recognition on the preprocessed patient identity information and chief complaint information, and perform patient name and disease information relationship extraction on the entity recognition result.

[0029] Convert the extracted patient name and disease information relationship into structured data and store it in the table in the database, and the table form includes a patient information table and a chief complaint information table.

[0030] As an optimal solution of the business reporting scene data collection method, the rule engine verifies and checks the data reported by the user according to the preset rules and conditions, and generates corresponding error prompts or correction suggestions; the rule engine uses a large language model LLM and a RAG architecture.

[0031] As the preferred data collection method for business reporting scenarios, it is used by employees to submit overtime applications, specifically including:

[0032] Obtaining the overtime hours and overtime reason information input by the user; performing data integrity verification on the overtime hours and overtime reason information, and determining whether required fields are included in the data integrity verification process. If required fields are missing, prompting the user to add additional fields;

[0033] Verify the data format and type of the overtime hours, and if the data format and type of the overtime hours do not conform to the preset rules, prompt the user to modify the overtime hours;

[0034] Perform business logic verification on the overtime hours to determine whether the overtime hours are within a preset range and whether the reason for overtime meets the predetermined emergency task conditions; if the overtime hours exceed the preset range or the reason for overtime does not meet the predetermined emergency task conditions, prompt the user to modify the overtime application;

[0035] The database of employee schedules for overtime hours is integrated to check whether the overtime hours reported by employees are consistent with the schedule information. If there is a conflict between the overtime hours and the schedule information, the user is prompted with an error or needs to re-fill the information.

[0036] A second aspect of the present invention provides a business reporting scenario data collection system, which adopts the business reporting scenario data collection method of the first aspect, including:

[0037] The reporting suggestion generation module is used to combine the user's reporting history data and context information based on the retrieval enhancement generation model, store the reporting history data and preset domain knowledge in a vector database; retrieve relevant information from the vector database as context based on the user's current input; and provide the relevant information and the user's current input to a generative large-scale language model to generate reporting suggestions for the multi-terminal reporting system;

[0038] A natural language description module is used to obtain multimodal information input by users through the multi-terminal reporting system interface, wherein the multimodal information includes text, voice, image or document;

[0039] A natural language analysis module, configured to use a large language model to perform end-to-end semantic understanding and information extraction on the multimodal information to convert it into structured data;

[0040] A structured data mapping module, configured to define a data model and field mapping rules, and map the structured data converted from the multimodal information to a specified data field according to the field mapping rules;

[0041] A data verification processing module is configured to design data verification rules according to business requirements and data quality requirements, wherein the data verification rules at least include rule engine-based verification or dynamically logical verification based on a generated large language model, and the data verification rules are used to verify and correct user-reported data to obtain verification results and explainable error prompts or correction suggestions.

[0042] A verified data storage module is configured to store processed and verified data for user data analysis, report generation and decision support.

[0043] As an optimal solution of the business reporting scene data collection system, in the natural language analysis module, the morphological analysis performs word segmentation and part-of-speech tagging on the multi-modal information input by the user to obtain basic information of each word in the multi-modal information.

[0044] The syntactic analysis processes the sentence structure and grammatical relationship of the multi-modal information input by the user to obtain semantic information of the multi-modal information.

[0045] The semantic understanding of the multi-modal information input by the user includes:

[0046] Entity recognition: using entity recognition technology to identify specific entities in the multi-modal information.

[0047] Relationship extraction: extracting the relationship between entities from the multi-modal information through relationship extraction technology.

[0048] Semantic role labeling: using semantic role labeling technology to label predicates and arguments in the multi-modal information to describe the semantic roles of different components in the sentence in the multi-modal information.

[0049] As an optimal solution of the business reporting scene data collection system, in the natural language analysis module, converting the multi-modal information into structured data includes:

[0050] Through the API interface of natural language processing, the natural language description of the user is called and processed, and the processing result is returned to the calling party.

[0051] The structured data obtained after natural language processing is converted into a specified storage form for storage in a database.

[0052] As an optimal solution of the business reporting scene data collection system, in the structured data mapping module, by analyzing user reporting history data, the reporting category, time, content key features are extracted, and combined with the user's geographic location, device information, a user reporting behavior model is established.

[0053] As the preferred solution of the business report filling scene data collection system, in the report suggestion generation module, the filling suggestion of the multi-terminal filling system is generated in combination with the filling history data and the context information of the user, an intention prediction model is constructed, and the current filling intention of the user is predicted according to the filling history and the context information of the user through the intention prediction model.

[0054] In the report suggestion generation module, the intention prediction model is trained using a machine learning algorithm according to historical data and filling intention labels.

[0055] As the preferred solution of the business report filling scene data collection system, in the report suggestion generation module, the filling intention and the historical data analysis result of the user are combined to design a report suggestion generation algorithm, the report suggestion generation algorithm is based on rules, statistical models or deep learning models, and the report suggestion is generated in combination with the context information and the historical data of the user.

[0056] As the preferred solution of the business report filling scene data collection system, in the data verification processing module, the data filled by the user is verified and checked according to the preset rules and conditions through the rule engine, and corresponding error prompts or correction suggestions are generated; the rule engine uses a large language model (LLM) and a RAG architecture.

[0057] The third aspect of the present application provides a non-transitory computer readable storage medium, the computer readable storage medium stores a program code of a business report filling scene data collection method, and the program code includes instructions for executing the business report filling scene data collection method of the first aspect or any possible implementation manner thereof.

[0058] The fourth aspect of the present application provides an electronic device, which comprises a memory and a processor; the processor and the memory complete mutual communication through a bus; the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the business report filling scene data collection method of the first aspect or any possible implementation manner thereof.

[0059] The beneficial effects of the present application are as follows: based on retrieval enhancement generation model, the filling history data and context information of the user are stored in a vector database in combination with the filling history data and preset domain knowledge; relevant information is retrieved from the vector database as context according to the current input of the user; the relevant information and the current input of the user are provided to a generative large language model to generate filling suggestions of a multi-terminal filling system; multi-modal information input by the user through a multi-terminal filling system interface is obtained, the multi-modal information including text, voice, image or document; a large language model is used for end-to-end semantic understanding and information extraction of the multi-modal information to convert into structured data; a data model and a field mapping rule are defined, and the structured data converted from the multi-modal information is mapped to a specified data field through the field mapping rule; according to business requirements and data quality requirements, data verification rule design is performed, the data verification rule uses a rule engine or a self-defined logic checking algorithm to verify and correct the data filled by the user to obtain verification results and error prompt results; the processed and verified data is stored for the user to perform data analysis, report generation and decision support. Through natural language processing technology, the user uses natural language description to fill in the data, avoiding the tedious form filling process, converting the natural language description into structured data to improve the efficiency of data collection; the filling history and related data of the user can be analyzed to predict the intention of the user and provide corresponding suggestions; the data can be verified and consistency checked to ensure the legality and accuracy of the data; different user requirements and use scenarios can be adapted. BRIEF DESCRIPTION OF DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the following embodiment or prior art description will be briefly introduced. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can also obtain other implementation drawings from the provided drawings without creative labor.

[0061] The structures, proportions, sizes, etc. shown in the specification are only used to cooperate with the content disclosed in the specification for understanding and reading by those skilled in the art, and do not define the limiting conditions for implementing the present application, so they do not have technical significance. Any modification of structure, change of proportion relationship or adjustment of size, which does not affect the effects and purposes that can be achieved by the present application, should still fall within the scope of the technical content disclosed by the present application.

[0062] Figure 1 A business filling scene data acquisition method process schematic diagram is provided for the embodiments of the present application;

[0063] Figure 2A technical route schematic diagram of a service filling scene data collection method provided for an embodiment of the present application is shown.

[0064] Figure 3 An application interface of the service filling scene data collection method provided for an embodiment of the present application is shown.

[0065] Figure 4 A system architecture schematic diagram of the service filling scene data collection system provided for an embodiment of the present application is shown. DETAILED DESCRIPTION

[0066] The embodiments of the present application will be described in detail by specific embodiments, and those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0067] Embodiment 1

[0068] Referring to Figure 1 and Figure 2 , the embodiment 1 of the present application provides a service filling scene data collection method, comprising the following steps:

[0069] S1, based on a retrieval enhancement generation model, combining the filling history data and context information of the user, storing the filling history data and the preset domain knowledge into a vector database; according to the current input of the user, retrieving relevant information from the vector database as context; providing the relevant information and the current input of the user to a generative large language model to generate filling suggestions of a multi-terminal filling system;

[0070] S2, obtaining multi-modal information input by the user through the interface of the multi-terminal filling system, the multi-modal information including text, voice, image or document; using a large language model to perform end-to-end semantic understanding and information extraction on the multi-modal information to convert into structured data;

[0071] S3, defining a data model and a field mapping rule, mapping the structured data converted from the multi-modal information to a specified data field through the field mapping rule;

[0072] S4, according to the business requirements and data quality requirements, designing data verification rules, the data verification rules at least including verification based on a rule engine or dynamic logical verification based on a generative large language model, verifying and correcting the data filled by the user to obtain verification results and explainable error prompts or correction suggestions;

[0073] S5, store the processed and verified data for user data analysis, report generation and decision support.

[0074] In this embodiment, in step S1, during the historical data analysis process, the key features such as the category, time, content of the report, etc. are extracted by analyzing the user's past report history, and the user's report behavior model is established in combination with the user's context information such as geographic location, device information, etc. Using machine learning and data mining techniques, an intent prediction model is constructed. The intent prediction model can automatically predict the user's current report intent, such as filling in a specific field, selecting an option, etc. according to the user's report history and context information. Then, according to the user's report intent and historical data analysis results, intelligent report suggestions are generated, which can include automatic filling of known information, providing options, data verification, etc. to improve the accuracy and efficiency of the report.

[0075] In one possible embodiment, the user's report history data is preprocessed, cleaned, filtered for outliers, etc. and key features are extracted. For example, time data is decomposed, the category of the report is extracted, the report frequency is calculated, etc. According to the historical data and the report intent label, a machine learning algorithm (such as a classification algorithm, a sequence model, etc.) is used to train the intent prediction model. Feature selection and conversion can be performed using feature engineering methods to improve the accuracy and generalization ability of the model. According to the user's report intent and historical data analysis results, a report suggestion generation algorithm is designed, which can be based on rules, statistical models or deep learning models, combined with user context information and historical data, to generate appropriate report suggestions.

[0076] Among them, according to the historical data and the report intent label, the trained intent prediction model can be a classification model, such as logistic regression or support vector machine, or a sequence model, such as hidden Markov model or recurrent neural network. The goal of the model is to predict the applicant's intent when filling out the application, such as the applicant's desire to obtain approval as soon as possible, or the applicant's desire to obtain a certain type of permit.

[0077] In one possible embodiment, the application first constructs a knowledge base containing the user's historical reporting data (after desensitization processing), relevant business regulations, operation manuals, etc. The knowledge base content is processed into high-dimensional vectors using a text embedding model (such as M3E, BGE, etc.) and stored in a Weaviate or Milvus vector database. When the user starts reporting, their input is also vectorized, and the system performs a fast similarity search in the vector database to recall the most relevant Top-K information. These information as "enhanced context", together with the user's original input, through a Prompt template, is submitted to a generative large language model (such as DeepSeek, Qwen after fine-tuning or calling commercial API). The LLM finally generates accurate reporting suggestions.

[0078] In one possible embodiment, the reporting suggestion generation algorithm is a rule-based, statistical model or deep learning model that combines user context information and historical data to generate appropriate reporting suggestions. For example, if the intent prediction model predicts that the applicant wants to obtain approval as soon as possible, the reporting suggestion generation algorithm suggests that the applicant provide more complete application information to speed up the approval process.

[0079] Taking the development of a multi-end reporting-based sales performance reporting system as an example, users can describe sales through voice or text, automatically perform natural language processing and provide reporting suggestions.

[0080] User input: "This month's sales exceeded 1 million, the main customer group is small and medium-sized enterprises, and the target market is the East China region."

[0081] Historical data analysis: Analyze the user's past reporting history data, extract key features such as reporting categories, time, sales, etc., and combine user context information such as geographic location, device information, etc.

[0082] Intent prediction: Based on the results of historical data analysis and context information, predict the user's reporting intent as "filling in this month's sales, customer groups and target market".

[0083] Reporting suggestion generation: According to the reporting intent and historical data analysis results, generate reporting suggestions. For example, reporting suggestions can include automatically filling in known information such as last month's sales, common options for customer groups, etc. According to historical data statistics, similar sales ranges can be provided for the user to choose. In addition, according to the user's reported target market, relevant regional options can be automatically provided, such as a list of provinces in the East China region.

[0084] In this embodiment, in step S2, the morphological analysis performs word segmentation and part-of-speech tagging on the user inputted multi-modal information to obtain the basic information of each word in the multi-modal information; the syntactic analysis processes the sentence structure and grammatical relationship of the user inputted multi-modal information to obtain the semantic information of the multi-modal information; the semantic understanding of the user inputted multi-modal information includes: the generative large language model realizes entity recognition, relationship extraction and semantic role labeling in the input information through context learning or fine-tuning, and generates structured data according to a preset output format; the predicate and argument in the multi-modal information are labeled using semantic role labeling technology to describe the semantic roles of different components in the sentence in the multi-modal information.

[0085] Specifically, in step S2, first, the multi-modal information is preprocessed, such as text cleaning, word segmentation, and removal of stop words, to reduce noise and standardize the text. Then, the text is segmented and tagged by using morphological analysis techniques to obtain the basic information of each word; the structure and grammatical relationship of the sentence are analyzed by using syntactic analysis techniques, such as subject-predicate-object relationship, modification relationship, etc., to obtain deeper semantic information; specific entities in the text are identified by using entity recognition techniques, such as names, places, organizations, etc., for subsequent entity relationship extraction and data integration; the relationship between entities is extracted from the text by using relationship extraction techniques, such as possession relationship, belonging relationship, etc., to construct the semantic connection between entities; the predicate and argument in the sentence are labeled by using semantic role labeling techniques to describe the semantic roles of different components in the sentence, such as agent, patient, time, etc.

[0086] In this embodiment, in step S2, converting the multi-modal information into structured data includes: calling and processing the user natural language description through the API interface of natural language processing, and returning the processing result to the calling party; converting the structured data obtained after natural language processing into a specified storage form to store in the database.

[0087] Specifically, a complete data collection process is designed, including receiving, preprocessing, NLP processing, data conversion, etc. of the user inputted natural language description. According to the specific task requirements, an NLP model is selected, such as a model based on a pre-trained model (such as BERT, GPT, etc.) or a self-defined model. A large-scale data set is used for model training to improve the semantic understanding and analysis ability of the model. By developing an API interface of NLP processing, the calling and processing of the user natural language description are realized, and the interface can receive user input, perform preprocessing and NLP processing, and return the processing result to the calling party.

[0088] The structured data obtained after NLP processing is converted into a form suitable for storage and subsequent processing, such as storing the data in a database or storing it in a specific data format (such as JSON, CSV, etc.). Quality checks and verifications are performed on the processed data to ensure its accuracy and completeness. Data quality control can be performed using rules, statistical methods, or manual review.

[0089] For example, a medical record collection system based on multi-terminal reporting can be developed. Users can describe their symptoms through voice or text, and the system will automatically perform NLP processing to extract medical record information.

[0090] User input: "Patient's name is Zhang San, age 42, chief complaint of fever, cough and sore throat, lasting for 3 days."

[0091] Preprocessing: Clean the text, remove punctuation and stop words, and get "Patient's name Zhang San, age 42, chief complaint of fever, cough and sore throat, lasting for 3 days."

[0092] Entity recognition: Identify "Zhang San" as the patient's name and "42" as the age.

[0093] Relationship extraction: Extract the relationship between "patient's name - Zhang San" and "chief complaint - fever, cough, sore throat."

[0094] Data conversion: Convert the extracted entities and relationships into structured data, such as a table stored in a database:

[0095] Patient information table:

[0096] Patient name Age Zhang San 42

[0097] Chief complaint information table:

[0098] Patient name Complaints Zhang San Fever Zhang San Cough Zhang San Sore throat

[0099] In one possible embodiment, taking medical record collection as an example, the doctor can not only speak "patient Zhang San, 42 years old..." but also directly upload a picture of a handwritten medical record draft. The system of the present invention first identifies the handwritten text and table structure in the picture through the VLM or OCR module, then sends the recognized text "Name: Zhang San, Chief Complaint: Fever for 3 days..." to the LLM, and the LLM automatically extracts structured information according to the preset medical record template and fills in the corresponding "patient information table" and "chief complaint information table".

[0100] In this embodiment, at step S3, the data model and field mapping rules are defined to map the structured data output by the natural language processing model to specific data fields, ensuring data accuracy and consistency, and facilitating subsequent data storage and management. Intelligent filling suggestions are implemented based on historical data and contextual information using machine learning and data mining techniques. By analyzing the user's filling history and related data, the system automatically predicts the user's intent and provides corresponding suggestions, including automatic filling and data validation functions.

[0101] In this embodiment, at step S4, the rule engine is used to verify and check the user's filled data based on pre-defined rules and conditions, generating corresponding error prompts or correction suggestions. The rule engine uses a large language model (LLM) and RAG architecture.

[0102] Specifically, in step S4, the user's filled data is verified and checked for consistency to ensure its legality and accuracy. This is achieved through rule engines, logical checks, and other technical means, avoiding the tediousness and errors of manual checks.

[0103] The data integrity verification process can verify whether the user's filled data is complete, i.e., whether it lacks necessary fields or information. By defining data models and rules, the system can check whether the user's filled data contains all required fields and provide corresponding error prompts or automatic completion functions.

[0104] The data format and type verification process can verify whether the user's filled data meets the pre-defined format and type requirements. For example, for date fields, the system can check whether the date format is correct; for numerical fields, the system can verify whether the data is of the numerical type and perform data format conversion or correction if necessary.

[0105] The business logic verification process can verify the user's filled data based on business rules and logical requirements. For example, for a sales performance reporting system, the system can check whether the sales amount is greater than zero and whether the target market is within the pre-defined optional range. If the filled data does not meet the business logic, the system will provide corresponding error prompts or correction suggestions.

[0106] The data consistency check process can check the consistency of the filled data with other related data to ensure consistency between the filled data and existing data. For example, in a multi-terminal filling system, if the user fills in the contact information of a client, the system can check whether the contact information is consistent with the information in the client database to avoid conflicts or contradictions between the filled data and existing data.

[0107] A rule engine is designed to define the rules for data validation and consistency checks. The rule engine can automatically validate and check the data filled by the user based on pre-set rules and conditions, and generate corresponding error prompts or correction suggestions. Open-source rule engine frameworks such as Large Language Model (LLM) and RAG architecture can be used, or a rule engine can be developed independently.

[0108] In one possible embodiment, a natural language processing-based rule engine is used, which can automatically validate and check the data filled by the user based on pre-set rules and conditions, and generate corresponding error prompts or correction suggestions. The natural language processing model uses natural language processing models, word embedding models or recurrent neural network models to model and train pre-set rules and conditions. The rule engine can understand and interpret the natural language text filled by the user, and perform data validation and consistency checks based on pre-set rules and conditions. For example, in the scenario of building permit application in the city planning department, a rule engine is designed to verify the legality of the building permit application, whether the scale meets the requirements, whether the applicant has the corresponding qualifications, etc.

[0109] A multi-terminal reporting system is developed for employees to fill out overtime applications. The system needs to verify and check the consistency of the reported overtime.

[0110] User filling: "Overtime time is August 14, 2023, 6:00 pm to 8:00 pm, overtime reason is to complete the urgent task of project X."

[0111] Data completeness verification: The system checks whether the overtime application form contains mandatory fields such as overtime time and overtime reason. If any field is missing, the system will provide an error prompt and require the user to complete it.

[0112] Data format and type verification: The system verifies whether the format of the overtime time field is correct, such as whether the date format is YYYY-MM-DD HH:MM. If the format is incorrect, the system will provide an error prompt and require the user to fill it out according to the specified format.

[0113] Business logic verification: The system verifies according to the pre-set business rules, such as checking whether the overtime time is within the legal range and whether the overtime reason meets the specified urgent task conditions. If the overtime time exceeds the specified working hours range, the system will provide an error prompt and require the user to modify the overtime time.

[0114] Data consistency check: The system can be integrated with the employee scheduling database to check whether the overtime time filled by the employee is consistent with the scheduling information. If the reported overtime time conflicts with the scheduling information, the system will provide an error prompt and require the user to re-fill or coordinate with the relevant departments.

[0115] In a possible embodiment, the data collection method of the present application is applied to multi-terminal reporting system to support different terminals such as web applications, mobile applications, etc. to adapt to different user needs and usage scenarios. The user interface of the system facilitates users to input natural language descriptions, view reporting suggestions and verification results.

[0116] In a possible embodiment, the data collection method of the present application is applied to government reporting: government departments often need to collect various types of reporting data, such as enterprise tax reporting, statistical reporting, etc. Through the multi-terminal reporting system, users use natural language descriptions to fill in data, and the system automatically extracts key information and provides reporting suggestions to improve reporting efficiency and accuracy.

[0117] In a possible embodiment, the data collection method of the present application is applied to government consultation and complaints: citizens submit consultation and complaints to government departments through the multi-terminal reporting system, describe the problem and demand through natural language, and the system automatically converts it into structured data and provides corresponding processing and reply.

[0118] In a possible embodiment, the data collection method of the present application is applied to policy research and public opinion monitoring: the government uses the multi-terminal reporting system to collect opinions and feedback on policy implementation, and uses natural language processing and data analysis technology to monitor and analyze public opinion, providing support for policy making and decision making.

[0119] In summary, the embodiment of the application is based on a retrieval-enhanced generation model, combines the user's historical data and context information, stores the historical data and pre-set domain knowledge into a vector database; according to the user's current input, retrieves relevant information from the vector database as context; provides the relevant information and the user's current input to a generative large language model to generate a report suggestion for a multi-terminal reporting system; obtains multi-modal information input by the user through the multi-terminal reporting system interface, the multi-modal information including text, voice, image or document; uses a large language model to perform end-to-end semantic understanding and information extraction on the multi-modal information to convert it into structured data; defines a data model and a field mapping rule, maps the structured data converted from the multi-modal information to a specified data field through the field mapping rule; according to business requirements and data quality requirements, designs a data verification rule, which uses a rule engine or a custom logic check algorithm to verify and correct the user's reported data, obtains a verification result and an error prompt result; stores the processed and verified data for the user to perform data analysis, report generation and decision support. The application improves data collection efficiency: through natural language processing technology, the user uses natural language description to report data, avoiding the tedious form filling process and improving the efficiency of data collection; provides intelligent reporting suggestions: based on historical data and context information, intelligent reporting suggestions can be provided. Through machine learning and data mining technology, the user's reporting history and related data are analyzed, the user's intention is predicted and corresponding suggestions are provided; data verification and consistency check are realized: the user's reported data is verified and checked for consistency to ensure the legality and accuracy of the data; the data can be verified according to the pre-set rules and the corresponding verification results and error prompts can be provided; different terminals such as web applications and mobile applications can be supported to adapt to different user needs and usage scenarios.

[0120] Reference Figure 3 An application interface design for the embodiment of the application, which includes:

[0121] Report suggestion interface: the report suggestion interface displays the report suggestions generated based on the user's historical data and context information for the user's reference.

[0122] Lexical analysis result item: the lexical analysis result item displays the processing results of the multi-modal information input by the user, including the structured data after lexical analysis, syntactic analysis and semantic understanding.

[0123] Relation extraction result item: the relation extraction result item displays the process of how the structured data converted from the multi-modal information is mapped to the specified data field through the field mapping rule.

[0124] The filling result display project: display the data storage after processing and verification, including the operations such as query, modification, deletion, etc.

[0125] It should be noted that the method of the embodiments of the present disclosure can be executed by a single device, such as a computer or a server, etc. The method of the embodiments can also be applied to a distributed scenario, and be completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only execute one or more steps in the method of the embodiments of the present disclosure, and the multiple devices can interact with each other to complete the method.

[0126] It should be noted that some embodiments of the present disclosure have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than the order described above and still achieve desirable results. Additionally, the processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0127] Embodiment 2

[0128] Referring to Figure 4 Embodiment 2 of the present disclosure provides a business filling scene data collection system, which adopts the business filling scene data collection method of Embodiment 1, and comprises:

[0129] The filling suggestion generation module 1 is configured to store the filling history data and the preset domain knowledge into a vector database based on the retrieval enhancement generation model and in combination with the filling history data and the context information of the user; retrieve relevant information from the vector database as the context according to the current input of the user; and provide the relevant information and the current input of the user to a generative large language model to generate filling suggestions of the multi-terminal filling system.

[0130] The natural language description module 2 is configured to acquire multi-modal information input by the user through a multi-terminal filling system interface, wherein the multi-modal information comprises text, voice, image or document.

[0131] The natural language analysis module 3 is configured to perform end-to-end semantic understanding and information extraction on the multi-modal information by using a large language model to convert the multi-modal information into structured data.

[0132] The structured data mapping module 4 is configured to define a data model and a field mapping rule, and map the structured data converted from the multi-modal information to a specified data field through the field mapping rule.

[0133] The data verification processing module 5 is used for designing data verification rules according to business requirements and data quality requirements, wherein the data verification rules at least include rule engine-based verification or dynamic logic verification based on a generated large language model, and the data filled by the user is verified and corrected to obtain a verification result and an interpretable error prompt or correction suggestion.

[0134] The verified data storage module 6 is used for storing the processed and verified data for the user to perform data analysis, report generation and decision support.

[0135] In a possible embodiment, in the filling suggestion generation module 1, the filling behavior model of the user is established by analyzing the filling history data of the user, extracting the category, time and content key features of the filling, and combining the geographic location and device information of the user.

[0136] In a possible embodiment, in the filling suggestion generation module 1, the generated large language model is fine-tuned by an instruction of a business scenario, so that the generated large language model generates filling content or suggestions conforming to a business logic according to the retrieved context information and the user input.

[0137] In the natural language description module 2, processing the multi-modal information specifically includes: when the input is an image or a document, adopting an optical character recognition technology and a visual language model to extract text and layout information therein, and then processing by the generated large language model.

[0138] In a possible embodiment, in the natural language analysis module 3, the morphological analysis performs word segmentation and part-of-speech tagging operations on the multi-modal information input by the user to obtain basic information of each word in the multi-modal information.

[0139] The syntactic analysis processes the sentence structure and grammatical relationship of the multi-modal information input by the user to obtain semantic information of the multi-modal information.

[0140] The semantic understanding of the multi-modal information input by the user includes:

[0141] The generated large language model realizes entity recognition, relationship extraction and semantic role labeling in the input information through context learning or fine-tuning, and generates structured data according to a preset output format; the semantic role labeling technology is used to label the predicate and argument in the multi-modal information to describe the semantic roles of different components in the sentence in the multi-modal information.

[0142] In a possible embodiment, in the natural language analysis module 3, converting the multi-modal information into structured data includes:

[0143] The API interface of the natural language processing is called and processed to the natural language description of the user, and the processing result is returned to the calling party.

[0144] The structured data obtained after the natural language processing is converted into a specified storage form to be stored in a database.

[0145] In a possible embodiment, in the data verification processing module 5, the data filled by the user is verified and checked according to preset rules and conditions by the rule engine, and corresponding error prompts or correction suggestions are generated; the rule engine uses a large language model LLM and a RAG architecture.

[0146] It should be noted that the information interaction and execution process between the modules of the system described above are based on the same concept as the method embodiments in Embodiment 1 of the present application, and the technical effects brought by them are the same as those of the method embodiments of the present application. For specific content, please refer to the description in the method embodiments described above.

[0147] Embodiment 3

[0148] Embodiment 3 of the present application provides a non-transitory computer-readable storage medium, the computer-readable storage medium stores a program code of a business filling scene data acquisition method, and the program code includes instructions for executing the business filling scene data acquisition method of Embodiment 1 or any possible implementation manner thereof.

[0149] The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (Solid State Disk, SSD)) and the like.

[0150] Embodiment 4

[0151] Embodiment 4 of the present application provides an electronic device, comprising a memory and a processor.

[0152] The processor and the memory complete mutual communication through a bus; the memory stores program instructions executable by the processor, and the processor calling the program instructions can execute the business filling scene data acquisition method of Embodiment 1 or any possible implementation manner thereof.

[0153] Specifically, the processor can be implemented by hardware or software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented by software, the processor can be a general-purpose processor, which reads software codes stored in a memory to implement the processor. The memory can be integrated in the processor or located outside the processor.

[0154] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means.

[0155] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present application can be implemented by general computing devices, which can be concentrated on a single computing device or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by program codes executable by computing devices, so that they can be stored in storage devices and executed by computing devices, and in some cases, the steps shown or described can be executed in different order, or they can be manufactured into individual integrated circuit modules, or multiple modules or steps can be manufactured into a single integrated circuit module. Thus, the present application is not limited to any specific combination of hardware and software.

[0156] Although the present application has been described in detail above with general description and specific embodiments, some modifications or improvements can be made on the basis of the present application, which is obvious to those skilled in the art. Therefore, these modifications or improvements made on the basis of not deviating from the spirit of the present application are within the scope of the present application.

Claims

1. The data collection method for business reporting scenarios is characterized by: include: Based on the retrieval enhancement generation model, the user's reporting history data and context information are combined to store the reporting history data and preset domain knowledge into a vector database; Retrieving relevant information from the vector database as context based on the user's current input; providing the relevant information and the user's current input to a generative large-scale language model to generate reporting suggestions for the multi-terminal reporting system; Acquire multimodal information input by users through a multi-terminal reporting system interface, the multimodal information including text, voice, image, or document; use a large language model to perform end-to-end semantic understanding and information extraction on the multimodal information to convert it into structured data; Define data models and field mapping rules, and map structured data converted from multimodal information to specified data fields using the field mapping rules; Design data validation rules based on business needs and data quality requirements. The data validation rules include at least rule engine-based validation or dynamic logic validation based on a generative large language model. Verify and correct the data submitted by users, and obtain verification results and explainable error prompts or correction suggestions. The processed and verified data is stored for users to perform data analysis, report generation and decision support.

2. The method for collecting business reporting scenario data according to claim 1, characterized in that: By analyzing historical user reporting data, we extract key features such as reporting category, time, and content, and build a user reporting behavior model based on the user's geographic location and device information. The generative large-scale language model is fine-tuned by instructions of a set business scenario, so that the generative large-scale language model generates reporting content or suggestions that conform to business logic based on the retrieved context information and user input; Processing the multimodal information specifically includes: when the input is an image or document, using optical character recognition technology and a visual language model to extract the text and layout information therein, and then processing it by the generative large language model.

3. The method for collecting business reporting scenario data according to claim 2, characterized in that: Used for sales performance reporting, including: Obtain sales information submitted by users, analyze historical reporting data, extract specified features, including reported category, time, and sales amount, and combine the user's geographic location and device information to predict the user's reporting intention; Generate sales reporting suggestions based on reporting intentions and historical data analysis results, including automatically filling in known sales figures from the previous month, customer group option information, sales ranges provided based on historical data statistics, and target area options based on the target market reported by the user.

4. The method for collecting business reporting scenario data according to claim 1, characterized in that: Using a natural language processing algorithm to perform lexical analysis, syntactic analysis, and semantic understanding on the multimodal information input by the user, and converting the multimodal information into structured data; The lexical analysis performs word segmentation and part-of-speech tagging operations on the multimodal information input by the user to obtain basic information of each word in the multimodal information; The syntactic analysis processes the sentence structure and grammatical relationship of the multimodal information input by the user to obtain semantic information of the multimodal information; Performing semantic understanding on the multimodal information input by the user includes: The generative large-scale language model realizes entity recognition, relationship extraction and semantic role labeling in the input information through context learning or fine-tuning, and generates structured data according to a preset output format; Annotating predicates and arguments in the multimodal information using a semantic role annotation technique to describe semantic roles of different components in sentences in the multimodal information; Converting the multimodal information into structured data includes: Through the natural language processing API interface, the user's natural language description is called and processed, and the processing results are returned to the caller; Convert the structured data obtained after natural language processing into a specified storage format for storage in the database.

5. The method for collecting business reporting scenario data according to claim 4, characterized in that: Used for collecting and reporting medical records, including: Obtaining the user's patient identification information and chief complaint information, wherein the patient identification information includes the patient's name and age, and the chief complaint information includes symptom information and symptom duration; Preprocessing the patient identification information and the chief complaint information, wherein the preprocessing includes removing punctuation marks and stop words; Performing entity recognition on the pre-processed patient identity information and the chief complaint information, and extracting the relationship between the patient name and symptom information from the entity recognition results; The relationship between the extracted patient names and symptom information is converted into structured data and stored in tables in the database. The tables include a patient information table and a chief complaint information table.

6. The method for collecting business reporting scenario data according to claim 1, characterized in that: The rule engine verifies and checks the data submitted by the user according to preset rules and conditions, and generates corresponding error prompts or correction suggestions; The rule engine uses Large Language Model (LLM) and RAG architecture.

7. The method for collecting business reporting scenario data according to claim 6, characterized in that: Used by employees to fill out overtime applications, including: Obtaining the overtime hours and overtime reason information input by the user; performing data integrity verification on the overtime hours and overtime reason information, and determining whether required fields are included in the data integrity verification process. If required fields are missing, prompting the user to add additional fields; Verify the data format and type of the overtime hours, and if the data format and type of the overtime hours do not conform to the preset rules, prompt the user to modify the overtime hours; Perform business logic verification on the overtime hours to determine whether the overtime hours are within a preset range and whether the reason for overtime meets the predetermined emergency task conditions; if the overtime hours exceed the preset range or the reason for overtime does not meet the predetermined emergency task conditions, prompt the user to modify the overtime application; The database of employee schedules for overtime hours is integrated to check whether the overtime hours reported by employees are consistent with the schedule information. If there is a conflict between the overtime hours and the schedule information, the user is prompted to indicate an error or to re-fill the information.

8. A business reporting scenario data collection system, using the business reporting scenario data collection method according to any one of claims 1 to 7, characterized in that: include: The reporting suggestion generation module is used to combine the user's reporting history data and context information based on the retrieval enhancement generation model, store the reporting history data and preset domain knowledge in a vector database; retrieve relevant information from the vector database as context based on the user's current input; and provide the relevant information and the user's current input to a generative large-scale language model to generate reporting suggestions for the multi-terminal reporting system; A natural language description module is used to obtain multimodal information input by users through the multi-terminal reporting system interface, wherein the multimodal information includes text, voice, image or document; A natural language analysis module, configured to use a large language model to perform end-to-end semantic understanding and information extraction on the multimodal information to convert it into structured data; A structured data mapping module, configured to define a data model and field mapping rules, and map the structured data converted from the multimodal information to a specified data field according to the field mapping rules; The data validation processing module designs data validation rules based on business needs and data quality requirements. The data validation rules include at least rule engine-based validation or dynamic logic validation based on a generative large language model. The module verifies and corrects the data submitted by the user, and obtains verification results and explainable error prompts or correction suggestions. The verification data storage module is used to store processed and verified data for users to perform data analysis, report generation and decision support.

9. A non-transitory computer-readable storage medium storing program code for a method for collecting data from a business reporting scenario, characterized in that: The program code includes instructions for executing the business reporting scenario data collection method described in any one of claims 1 to 7.

10. An electronic device comprising: memory and processor; The processor and the memory communicate with each other via a bus; The memory stores program instructions that can be executed by the processor, and is characterized in that the processor calls the program instructions to execute the business reporting scenario data collection method described in any one of claims 1 to 7.

Images