Multimodal zero-code form modeling intelligent question-answering method and related equipment

Abstract

The present application provides a multimodal zero-code form modeling intelligent question and answer method and related equipment. The present application can support the processing of intelligent questions of multiple users on multiple different business platforms. According to the intentions of different users, the business data of different users on different business platforms are analyzed and processed to generate multiple hypothetical answers for users to choose from, and multiple hypothetical answers are evaluated, thereby improving the accuracy and reliability of intelligent question and answer. The method provided in the embodiment of the present application can support multiple data modalities, including data processing in different forms such as text, voice, and image, and can obtain user intentions from different types of input data, which can effectively improve user satisfaction and experience, provide users with more accurate and comprehensive question and answer results, and improve the intelligence level of the intelligent question and answer system. At the same time, it can improve the efficiency of users using intelligent question and answer tools and achieve the effect of reducing costs and increasing efficiency.

Description

Technical Field

[0001] The present application relates to the field of intelligent question-answering technology, and in particular to a multimodal zero-code form modeling intelligent question-answering method, apparatus, device and readable storage medium. Background Art

[0002] With the development of science and technology, artificial intelligence technology has developed better and better. In some industries, some intelligent zero-code platforms have gradually emerged. Zero-code platforms can build some personalized business applications for enterprises based on the business form of the enterprise. AI technology can make the question and answer services of these different business applications more intelligent.

[0003] The question-and-answer system uses the company's knowledge base documents and empowers AI assistants to build the company's own intelligent customer service. In actual application, many business platforms have also launched AI question-and-answer assistants. When users solve business problems, they can use AI question-and-answer assistants to quickly search and analyze. However, for each corporate customer, there will be multiple business applications running within the company. Because the business platforms are independent, it is impossible to connect all business platforms through one AI question-and-answer assistant. When customers need to use the AI ​​question-and-answer assistant, they need to use the corresponding AI question-and-answer assistant in the corresponding business application to achieve fast search and analysis. This method of use not only fails to improve the customer's work efficiency, but also increases the user's cost of use. In addition, this method of use is cumbersome and is not conducive to improving the user's work efficiency, and cannot achieve the effect of reducing costs and increasing efficiency. Summary of the invention

[0004] The present application aims to solve at least one of the above-mentioned technical defects. In view of this, the present application provides a multimodal zero-code form modeling intelligent question and answer method and related equipment, which are used to solve the technical defect of low efficiency of existing intelligent question and answer solutions in the prior art.

[0005] A multimodal zero-code form modeling intelligent question-answering method, comprising:

[0006] Collecting first target business data corresponding to the user on different business platforms, wherein the first target business data includes different task items corresponding to the user on different business platforms and original documents corresponding to each task item;

[0007] Performing structured processing on the first target business data to obtain second target business data;

[0008] According to the first target question raised by the user, the second target business data is processed to determine the user's intention;

[0009] According to the user's intention, the second target service data is processed to determine at least one hypothetical answer corresponding to the user's intention;

[0010] Evaluate and rank each generated hypothetical answer, and determine a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers;

[0011] Based on the first target hypothetical answer, the original document corresponding to the user's intention is processed to determine a second target hypothetical answer corresponding to the user's intention as the final question-and-answer result corresponding to the first target question.

[0012] Preferably, the performing structured processing on the first target service data to obtain second target service data includes:

[0013] Determine a processing strategy for the first target business data based on the user's customized configuration information, wherein the user's customized configuration information includes configuration information for abnormal data processing, configuration information for recollection strategies for various types of data collected with abnormalities, configuration information for log records of errors in data collection and processing, configuration information for a notification mechanism when errors occur, configuration information for outputting processing results, configuration information for format requirements for output data, and configuration information for setting naming rules and storage paths for output files;

[0014] Locating the positions of each unstructured data of the first target business data according to the processing strategy for the first target business data;

[0015] Determining, according to the positions of the respective unstructured data of the first target business data, respective target unstructured data and their data types that need to be processed in the first target business data;

[0016] According to the processing strategy of each target unstructured data, after cleaning and formatting the target unstructured data, according to the data type of each target unstructured data, determine the data processing tool corresponding to the data structure type of each target unstructured data;

[0017] Each non-target structured data is structured by using a data processing tool of the data structure type of each target non-structured data to obtain structured data corresponding to each target non-structured data as the second target business data.

[0018] Preferably, the performing structured processing on the first target service data to obtain second target service data includes:

[0019] Determine each unstructured data of the first target business data and its data type;

[0020] Determine, according to the data type of each unstructured data, a data processing tool corresponding to the data structure type of each unstructured data;

[0021] Each unstructured data is structured by using a data processing tool of the data structure type of each unstructured data to obtain structured data corresponding to each unstructured data;

[0022] The structured data corresponding to all the unstructured data corresponding to the first target business data is used as the second target business data.

[0023] Preferably, the processing of the second target business data based on the first target question raised by the user to determine the user's intention includes:

[0024] According to each task item in the second target business data, setting parameters of a semantic recognition model corresponding to the target task item corresponding to the first target problem;

[0025] Identifying historical access data of the user on different business platforms based on the second target business data;

[0026] Analyze the context of the user's historical access data, improve the first target question according to the target task item and its corresponding original document, and determine the second target question raised by the user, wherein the second target question is related to the first target question;

[0027] Extracting keywords from the second target question and converting the extracted results into structured data of the second target question;

[0028] Based on the parameters of the semantic recognition model corresponding to the target task item and the structured data of the second target problem, the semantic recognition model corresponding to the target task item is used to perform semantic analysis on the business data corresponding to the target task item to determine the user's intention.

[0029] Preferably, the processing of the second target service data according to the user's intention to determine at least one hypothetical answer corresponding to the user's intention includes:

[0030] Determine, based on the user's intention and the second target business data, a task description of the task item corresponding to the user's intention, a category of the second target question raised by the user, and a scope of a hypothetical answer to the second target question;

[0031] Extracting first target feature data of the task item corresponding to the user's intention according to the task description of the task item corresponding to the user's intention, the category of the second target question, and the scope involved in the hypothetical answer to the second target question;

[0032] Converting the characteristic data of the first target into characteristic data of a second target in a preset format;

[0033] The second target feature data is analyzed to generate at least one hypothetical answer corresponding to the task item corresponding to the user's intention.

[0034] Preferably, the step of structuring each non-target structured data by using a data processing tool of the data structure type of each target non-structured data to obtain structured data corresponding to each target non-structured data as the second target business data includes:

[0035] When the target unstructured data is unstructured data in a portable document format, first text content in the unstructured data in the portable document format is obtained, and data content corresponding to a table in the unstructured data in the portable document format is parsed into second text content, and content corresponding to an image in the unstructured data in the portable document format is identified as third text content, and the first text content, the second text content, and the third text content are stored according to a preset data storage vectorization storage strategy as structured data corresponding to the unstructured data in the portable document format;

[0036] When the target unstructured data is data in a spreadsheet document format, each row of data and header information of the data in the spreadsheet document format is parsed into data stored in a database storage form as structured data corresponding to the unstructured data in the spreadsheet document format;

[0037] When the target unstructured data is user-defined unstructured data, a data processing component that matches the form data content selected by the user is determined, and the form data content selected by the user and the user business data content corresponding to the corresponding data processing component are converted into database format data as structured data corresponding to the user-defined unstructured data through the determined data processing component.

[0038] Preferably, the step of evaluating and ranking each generated hypothetical answer and determining a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers comprises:

[0039] Determining a weight coefficient of each evaluation dimension of each generated hypothetical answer according to the user's intention, wherein the evaluation dimension includes accuracy, completeness, semantic coherence, and information reliability of each hypothetical answer generated for the second target question;

[0040] According to the preset evaluation dimension and the preset scoring method, each generated hypothetical answer is evaluated to obtain an evaluation result of each generated hypothetical answer;

[0041] According to the evaluation result of each hypothetical answer, the matching degree of all the hypothetical answers with the second target question is sorted to obtain the matching degree sorting result of each hypothetical answer;

[0042] According to the matching ranking result of each hypothetical answer, a first target hypothetical answer that best matches the user's intention is determined from the generated hypothetical answers.

[0043] A multimodal zero-code form modeling intelligent question-answering system is applied to any of the multimodal zero-code form modeling intelligent question-answering methods described above, and the multimodal zero-code form modeling intelligent question-answering system includes: a data acquisition module, a first processing module, a model management module, a second processing module, and a third processing module;

[0044] The data collection module is responsible for collecting the first target business data corresponding to the user on different business platforms, and performing structured processing on the first target business data to obtain the second target business data and transmit it to the first processing module, wherein the first target business data includes different task items corresponding to each user on each business platform and the original document corresponding to each task item;

[0045] The first processing module receives and processes the second target service data, determines the user's intention, and transmits it to the second processing module;

[0046] The second processing module receives and, based on the user's intention, calls the data processing model tool corresponding to the user's intention in the model management module to process the second target business data, generates at least one hypothetical answer corresponding to the task item corresponding to the user's intention, evaluates each generated hypothetical answer, and determines a first target hypothetical answer that best matches the user's intention from all generated hypothetical answers, and sends the first target hypothetical answer and its data processing model tool to the third processing module, wherein the model management module includes at least one data processing model tool;

[0047] The third processing module processes the original document corresponding to the user's intention based on the first target hypothetical answer and its data processing model tool to determine the final question and answer result corresponding to the user's intention.

[0048] A multimodal zero-code form modeling intelligent question-answering device, comprising: one or more processors, and a memory;

[0049] The memory stores computer-readable instructions, which, when executed by the one or more processors, implement the steps of the multimodal zero-code form modeling intelligent question-answering method as described in any of the above introductions.

[0050] A readable storage medium having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors implement the steps of the multimodal zero-code form modeling intelligent question-answering method as described in any one of the aforementioned introductions.

[0051] It can be seen from the technical solutions introduced above that when it is necessary to provide users with intelligent question-and-answer services, the method provided in the embodiment of the present application can collect the first target business data corresponding to users on different business platforms, wherein the first target business data can include different task items corresponding to users on different business platforms and the original documents corresponding to each task item, and the original documents corresponding to each task item can feedback the details of each task item; therefore, by analyzing the first target business data, the user's needs can be better understood to provide question-and-answer results that better meet the user's needs. Since different business platforms store data in different formats, in order to improve the processing efficiency of the first target business data, after collecting the first target business data, the first target business data can be structured so that the second target business data can be obtained. After obtaining the second target business data, the second target business data is processed according to the first target question raised by the user to determine the user's intention; then the second target business data is processed according to the user's intention to determine at least one hypothetical answer corresponding to the user's intention; among the various hypothetical answers generated, each generated hypothetical answer can be further evaluated and ranked, and a first target hypothetical answer that best matches the user's intention can be determined from the generated hypothetical answers; in order to improve the matching of the determined hypothetical answer with the question raised by the user, the original document corresponding to the user's intention can be further processed according to the first target hypothetical answer, and the second target hypothetical answer corresponding to the user's intention is determined as the final question and answer result corresponding to the first target question.

[0052] It can be seen that the method provided in the embodiment of the present application can support the processing of intelligent questions of multiple users on multiple different business platforms, and can analyze and process the business data of different users on different business platforms according to the intentions of different users, so as to generate multiple hypothetical answers corresponding to the task items corresponding to the users on different business platforms for users to choose from, and also effectively improve the accuracy and reliability of intelligent question and answer by evaluating and sorting multiple hypothetical answers. The method provided in the embodiment of the present application can support multiple data modalities, including data processing in different forms such as text, voice, and image, and can obtain user intentions from different types of input data, and generate diversified and targeted hypothetical answers based on the generated user intentions, which can effectively improve user satisfaction and experience, provide users with more accurate and comprehensive question and answer results, improve the intelligence level of the intelligent question and answer system, and at the same time improve the efficiency of users using intelligent question and answer tools, thereby achieving the effect of reducing costs and increasing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0054] Figure 1 A schematic diagram of the system architecture of a zero-code form modeling intelligent question-answering system for implementing multimodality provided in an embodiment of the present application;

[0055] Figure 2 A flowchart of a method for implementing multimodal zero-code form modeling intelligent question-answering provided in an embodiment of the present application;

[0056] Figure 3-Figure 4 They are schematic diagrams of the implementation process of a multimodal zero-code form modeling intelligent question-answering method provided in an embodiment of the present application;

[0057] Figure 5 A schematic diagram of the structure of a multi-modal zero-code form modeling intelligent question-answering device exemplified in an embodiment of the present application;

[0058] Figure 6 This is a hardware structure block diagram of a multimodal zero-code form modeling intelligent question-and-answer device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0060] At present, many existing intelligent question-and-answer systems are pre-set scenarios and difficult to customize for specific industries or scenarios. As the business develops and changes, the capabilities of the question-and-answer system need to be constantly updated, but the model library of each platform is uneven, making it difficult to maintain and update in a timely manner. Existing question-and-answer systems may be difficult to integrate with other applications or services, and are limited to fixed businesses.

[0061] In view of the fact that most of the current multimodal zero-code form modeling intelligent question and answer solutions are difficult to adapt to complex and changeable business needs, the applicant has studied a multimodal zero-code form modeling intelligent question and answer solution. The multimodal zero-code form modeling intelligent question and answer method has a clear data structure and can support the processing of intelligent questions from multiple users on multiple different business platforms. It can also analyze and process the business data of different users on different business platforms according to the intentions of different users, so as to generate multiple hypothetical answers corresponding to the tasks corresponding to the users on different business platforms for users to choose from, and also effectively improve the accuracy and reliability of intelligent question and answer by evaluating and sorting multiple hypothetical answers. The method provided in the embodiment of the present application can support multiple data modes, including data processing in different forms such as text, voice, and image, and can obtain user intentions from different types of input data, and generate diversified and targeted hypothetical answers based on the generated user intentions, which can effectively improve user satisfaction and experience, provide users with more accurate and comprehensive question and answer results, and improve the intelligence level of the intelligent question and answer system. At the same time, it can improve the efficiency of users using intelligent question and answer tools and achieve the effect of reducing costs and increasing efficiency.

[0062] The method provided in the embodiments of the present application can be used in many general or special computing device environments or configurations, such as personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor devices, and distributed computing environments including any of the above devices or equipment.

[0063] An embodiment of the present application provides a multimodal zero-code form modeling intelligent question and answer method, which can be applied to an intelligent question and answer system, and can also be applied to various computer terminals or smart terminals. Its execution subject can be a processor or server of a computer terminal or a smart terminal.

[0064] Combine the following Figure 1 , introduces an optional system architecture that can realize intelligent question answering given in the embodiment of the present application, such as Figure 1 As shown, the system architecture may include: a data acquisition module, a first processing module, a model management module, a second processing module, and a third processing module.

[0065] Among them, the data collection module can be responsible for collecting the first target business data corresponding to the user on different business platforms.

[0066] Among them, the first target business data can include different tasks corresponding to users on different business platforms and the original documents corresponding to each task; the original documents corresponding to each task can retain all the detailed information of each task from the beginning to the completion, and can be used as a reference in the execution of the task to ensure that all work is carried out according to the established plan. For example, in tasks involving complex processes, checking the original requirements at any time can ensure that the direction is not deviated. When problems or disputes arise, the source of the problem can be found through the original documents. For example, in quality inspection tasks, if there are quality problems with the product, the original quality standard documents can be used to determine whether it meets the requirements. The original documents of the task can also precipitate the experience and knowledge in the task, provide reference for similar tasks in the future, and improve overall work efficiency.

[0067] For example, the first target business data may include different tasks of at least one user on at least one business platform and an original document corresponding to each task.

[0068] Generally speaking, the business data of different business platforms may have different storage formats and data types. Therefore, the collected first target business data may be some unstructured data. If the first target business data is directly processed and analyzed, it will be difficult to process. Therefore, in order to improve the consistency of the data, it is possible to consider structuring the collected first target business data.

[0069] For example, when data is not structured, there may be inconsistencies in the format. For example, when recording dates, some data may be in the format of "year / month / day", while others may be in the format of "month / day / year". This will cause trouble in the use and analysis of data. By structuring unstructured data, the format of the data can be unified to ensure that the same type of data has the same representation. For example, all dates are standardized in the format of "year-month-day", so that when performing data query, screening or comparison operations, they can be performed accurately, improving the accuracy and reliability of the data.

[0070] Secondly, structured data can be stored in a database or other data storage system in a more orderly manner. For example, a relational database is designed based on the principle of structured data. It organizes data in the form of tables, and each table has clear field definitions. This method makes data storage more compact and efficient, reducing the waste of storage space. At the same time, it is also convenient for data management, such as data backup, recovery, and maintenance.

[0071] In a structured data environment, database management systems can use indexing and other technologies to quickly locate the required data. For example, when you look for a specific customer's record in a customer information database, if the data is structured, the database can quickly retrieve relevant information based on key fields such as the customer's number or name. Structuring unstructured data can speed up data retrieval and query; in contrast, if the data is unstructured, it is necessary to perform a full-text search on a large amount of text or other forms of data, which is not only inefficient but may also produce inaccurate results.

[0072] In order to improve the efficiency and accuracy of data analysis in massive data, different data analysis tools are often needed to process it. Structuring unstructured data can effectively support different data analysis tools for data analysis and decision-making; for example, data analysis tools and algorithms usually require structured data as input. For example, when performing tasks such as data mining, statistical analysis, or machine learning, structured data can be more easily processed and modeled. Taking sales data analysis as an example, if the sales data records product information, sales time, sales location, sales quantity, and amount in a structured manner, valuable information such as sales trends, popular products, and regional sales differences can be found through data analysis, thereby providing support for corporate decision-making, such as formulating marketing strategies and optimizing inventory management.

[0073] In particular, when different systems or business platforms need to share data, structured data is easier to share and integrate. If each system or platform uses its own unique unstructured data format, data exchange and sharing will become very difficult. Structured data can achieve data transmission and integration between different systems by defining unified data standards and interfaces. For example, various business systems or business platforms within an enterprise can achieve information interconnection and improve the overall operational efficiency of the enterprise by sharing structured financial data, customer data, etc.

[0074] In summary, structuring data is of great significance to improving the quality, availability and value of data, and is the basis for effective management and utilization of data.

[0075] Therefore, after collecting the business data of different users on different business platforms, the first target business data may be further structured, thereby obtaining the second target business data.

[0076] For example, the data collection module can train various task items and original documents from unstructured data to structured data according to custom configuration; wherein, the data collection module can use NLP processing methods, information extraction methods or text analysis methods to train the collected business data during data training.

[0077] For example, the process of training the task items and original documents corresponding to the questions raised by the user from unstructured data to structured data by the data collection module according to the custom configuration can be as follows:

[0078] (1) Perform data cleaning and preprocessing on the tasks and original documents corresponding to the questions raised by the user, including removing useless information, processing missing values ​​and outliers, etc.;

[0079] (2) Determine appropriate processing methods based on user needs and the characteristics of user business data on different business platforms; for example, one or more technologies such as NLP processing methods, information extraction methods, or text parsing methods may be combined to determine appropriate processing methods for user business data on different business platforms.

[0080] (3) Based on the configuration information corresponding to the determined processing method, the data is trained using relevant algorithms and models to convert unstructured data into structured data.

[0081] (4) After training is completed, verify and evaluate the effectiveness of the model, and tune and optimize the model as needed to ensure that the generated structured data meets the expected format and requirements.

[0082] Among them, preferably, in the actual application process, when choosing to use the NLP processing method to process the business data of users on different business platforms, the word frequency statistics, part-of-speech tagging, entity recognition and other technologies can be used to perform word segmentation, vectorization and other processing on the text data, so as to convert it into a structured data form.

[0083] When using information extraction methods to process users' business data on different business platforms, rules or models can be designed to identify key information and entities in the text corresponding to the business data of users on different business platforms, and extract them to form structured data. For example, technologies such as entity recognition and relationship extraction can be included.

[0084] In particular, you can also choose text parsing methods to process users' business data on different business platforms. For example, you can use technologies such as grammatical analysis, syntactic analysis, and semantic analysis to understand the grammatical structure and meaning of the text, and convert the business data of users on different business platforms into structured data forms in the future.

[0085] The second target business data may be structured data, and a large amount of data may be efficiently processed by analyzing the second target business data. Therefore, after the second target business data is determined, it may be transmitted to the first processing module for processing.

[0086] The second target business data can reflect the relevant information of the user; the second target business data is the data generated by the user on different business platforms, for example, the second target business data can be the type of goods purchased by the user, the purchase frequency, the purchase amount and other data, which can directly reflect their consumption preferences and demand intentions. For example, users who frequently purchase sports equipment may have a strong intention to exercise and keep fit.

[0087] The second target business data may be the browsing behavior data of users on different business platforms, such as the pages browsed by users on websites or applications, the duration of stay, etc., which may indicate the areas of interest. For example, if users browse travel guide pages for a long time, the browsing behavior data of these users may indicate that they have the intention to travel.

[0088] The second target business data can be the search keyword data of users on different business platforms. The search keywords entered by users are the embodiment of their active expression of needs and accurately reflect their current intentions. For example, if a user searches for "English learning courses", it is obvious that they have the intention to learn English.

[0089] The second target business data can be the interactive behavior data of users on different business platforms. The interactive behaviors of users such as likes, comments, and shares on social media can reflect the topics they are concerned about and their emotional tendencies, and then infer their potential intentions. For example, users who often interact with environmental protection related content may pay more attention to environmental protection and have corresponding participation intentions.

[0090] It can be seen that the second target service data can reflect the user's intention. Therefore, the first processing module can receive the second target service data and process it, for example, it can determine the user's intention and transmit it to the second processing module.

[0091] Among them, before the first processing module processes the second target business data to extract content, in order to improve the processing efficiency of the second target business data, the model parameters can be set according to the task items corresponding to the questions raised by the user; and then the semantic recognition model can be automatically matched according to the parameter settings to perform semantic analysis on the second target business data.

[0092] The setting of model parameters may include the setting of model type, feature selection, algorithm selection, hyperparameter adjustment and other parameters to ensure that the data processing model can better adapt to different types of tasks and original documents.

[0093] Among them, before semantic analysis, the first processing module can also supplement and improve the questions raised by the user based on the context of the user's historical access records, and the task items and original documents corresponding to the questions raised by the user, and then extract the core features based on the questions raised by the user after the supplement and improvement, and convert the extracted features into structured data.

[0094] After extracting the features of the question, the first processing module automatically matches the semantic recognition model according to the user-defined parameters to perform semantic analysis on the task items and original documents corresponding to the question raised by the user, thereby obtaining the user's intention semantic recognition model, which includes at least one or more of the bag-of-words model, Word2Vec model, GloVe model, attention model, semantic role labeling model and Transformer model.

[0095] In actual application, the semantic recognition model can also include traditional machine learning models (such as support vector machines, decision trees, etc.) or deep learning models (such as BERT, GPT, etc.) to understand the semantic information and user intent in the text in the business data, and perform semantic analysis on the task items and original documents through the matching semantic recognition model. Semantic analysis can include identifying keywords, entities, understanding sentence structure and semantic relationships, etc. to extract deep semantic information of the text. Automating the model parameter setting and semantic recognition model matching process makes the entire processing flow more efficient and intelligent. According to the changes and needs of the task items, the appropriate model parameters and semantic recognition models can be quickly adapted to improve the adaptability and flexibility of the system.

[0096] For example, the process in which the first processing module receives the second target service data and processes it to determine the user's intention may include the following:

[0097] Step S101, receiving second target service data;

[0098] Step S102, setting parameters of a semantic recognition model corresponding to each task item in the second target business data;

[0099] Step S103, identifying the user's historical access data based on the second target business data;

[0100] Step S104, analyzing the context of the user's historical access data, improving the question raised by the user according to the current task item and its corresponding original document, and determining the first target question raised by the user;

[0101] Step S105, extracting keywords from the first target question raised by the user, and converting the extracted results into structured data of the target question;

[0102] Step S106, based on the parameters of the semantic recognition model corresponding to each task item and the structured data of the target question, the semantic recognition model corresponding to each task item is used to perform semantic analysis on the business data corresponding to each task item to determine the user's intention.

[0103] Determining the user's intention can better provide personalized intelligent services to users. For example, according to the intention, more intelligent answers can be provided to the questions raised by users. For example, when the user's intention is to buy outdoor sports equipment and needs the business platform to recommend some products, the e-commerce platform can recommend suitable sports brands and professional equipment for the user after learning that the user has the intention to buy outdoor sports equipment, and provide relevant sports suggestions and matching guides.

[0104] After determining the user's intention, we can improve product functions and user experience based on the user's intention. For example, a music player software can analyze the user's search and playback intentions and find that the user has a high preference for a specific music style. In this case, we can increase the recommendation weight of this type of music in the recommendation algorithm and optimize the interface layout to make the classification and search of this style of music more convenient.

[0105] Therefore, the second processing module can receive and, based on the user's intention, call the data processing model tool corresponding to the user's intention in the model management module to process the second target business data, generate at least one hypothetical answer corresponding to the task item corresponding to the user's intention, evaluate each hypothetical answer generated, and determine a first target hypothetical answer that best matches the user's intention from all the hypothetical answers generated, and send the first target hypothetical answer and its data processing model tool to the third processing module.

[0106] Among them, the model management module may include at least one data processing model tool; for example, the data processing model tool may at least include: chart tools with different functions, statistical models with multiple functions, natural language processing models, computer vision models, knowledge graph models, time series prediction models, and multiple integrated learning models.

[0107] For example, the second processing module, based on the user's intention, calls the data processing model tool corresponding to the user's intention in the model management module to process the second target business data, and generates at least one hypothetical answer corresponding to the task item corresponding to the user's intention, which may include the following:

[0108] Step S201, determining, based on the user's intention, a task description of the task item corresponding to the user's intention, a category of the second target question raised by the user, and a scope of a hypothetical answer to the first target question raised by the user;

[0109] Step S202, extracting first target feature data of the task item corresponding to the user's intention based on the task description of the task item corresponding to the user's intention, the category of the first target question raised by the user, and the scope involved in the hypothetical answer to the first target question raised by the user;

[0110] Step S203, converting the extracted first target feature data into second target feature data in a preset format;

[0111] Step S204, calling the data processing model tool corresponding to the user's intention in the model management module, analyzing the second target feature data, and generating at least one hypothetical answer corresponding to the task item corresponding to the user's intention.

[0112] In order to further improve the matching degree between the generated hypothetical answers and the user's intentions and the questions raised by the user, the third processing module can process the original document corresponding to the user's intentions based on the first target hypothetical answers and their data processing model tools to determine the final question and answer results corresponding to the user's intentions.

[0113] From the above introduction, it can be seen that the intelligent question-answering system of this application makes full use of multimodal data, fully utilizes the advantages of multimodal data, and presents more accurate, comprehensive and personalized question-answering results to users through intelligent data processing and evaluation. The entire system architecture design is designed to improve the accuracy, reliability and user experience of the intelligent question-answering system, so that users can more easily obtain the information and answers they need.

[0114] Combine the following Figure 2 , introduces the process of the multimodal zero-code form modeling intelligent question-answering method given in the embodiment of the present application, which can be applied to the multimodal zero-code form modeling intelligent question-answering system introduced above, such as Figure 1 As shown, the process can include the following steps:

[0115] Step S301, collecting first target business data corresponding to the user on different business platforms, wherein the first target business data includes different task items corresponding to the user on different business platforms and the original document corresponding to each task item;

[0116] Step S302, performing structured processing on the first target business data to obtain second target business data;

[0117] The process of performing structured processing on the first target service data to obtain the second target service data may include the following implementation methods:

[0118] The first one,

[0119] (1) determining a processing strategy for the first target business data based on the user's customized configuration information, wherein the user's customized configuration information may include configuration information for abnormal data processing, configuration information for recollection strategies for various types of data collected with abnormalities, configuration information for log records of errors in data collection and processing, configuration information for notification mechanisms when errors occur, configuration information for outputting processing results, configuration information for format requirements for output data, and configuration information for setting naming rules and storage paths for output files;

[0120] (2) determining the location of each unstructured data of the first target business data according to the processing strategy for the first target business data;

[0121] (3) determining each target unstructured data and its data type that need to be processed in the first target business data according to the location of each unstructured data of the first target business data;

[0122] (4) After cleaning and formatting each target unstructured data according to the processing strategy of each target unstructured data, determine the data processing tool corresponding to the data structure type of each target unstructured data according to the data type of each target unstructured data;

[0123] (5) Using a data processing tool of the data structure type of each target unstructured data, each non-target structured data is structured to obtain structured data corresponding to each target unstructured data as the second target business data.

[0124] The second type,

[0125] (1) determining each unstructured data of the first target business data and its data type;

[0126] (2) determining a data processing tool corresponding to the data structure type of each unstructured data according to the data type of each unstructured data;

[0127] (3) using a data processing tool of the data structure type of each unstructured data, each unstructured data is structured to obtain structured data corresponding to each unstructured data;

[0128] (4) The structured data corresponding to all the unstructured data corresponding to the first target business data is used as the second target business data.

[0129] The third type,

[0130] (1) Determine a processing strategy for the first target business data based on the user's customized configuration information, locate the position of each unstructured data of the first target business data based on the processing strategy for the first target business data; determine each target unstructured data and its data type that need to be processed in the first target business data based on the position of each unstructured data of the first target business data; clean and format each target unstructured data based on the processing strategy of each target unstructured data, and then determine a data processing tool corresponding to the data structure type of each target unstructured data based on the data type of each target unstructured data; perform structured processing on each non-target structured data through the data processing tool of the data structure type of each target unstructured data, and obtain structured data corresponding to each target unstructured data as the second target business data.

[0131] (2) Determine each unstructured data and its data type; determine the data processing tool corresponding to the data structure type of each unstructured data according to the data type of each unstructured data; perform structured processing on each unstructured data through the data processing tool of the data structure type of each unstructured data to obtain the structured data corresponding to each unstructured data; use the structured data corresponding to all the unstructured data corresponding to the first target business data as the second target business data.

[0132] In particular, the process of structuring each non-target structured data by using a data processing tool of the data structure type of each target non-structured data to obtain structured data corresponding to each target non-structured data as the second target business data may include the following:

[0133] When the target unstructured data is unstructured data in a portable document format, a first text content in the unstructured data in the portable document format can be obtained, and the data content corresponding to the table in the unstructured data in the portable document format can be parsed into a second text content, and the content corresponding to the picture in the unstructured data in the portable document format can be identified as a third text content, and the first text content, the second text content and the third text content can be stored according to a preset data storage vectorization storage strategy as structured data corresponding to the unstructured data in the portable document format;

[0134] When the target unstructured data is data in a spreadsheet document format, each row of data and header information of the data in the spreadsheet document format may be parsed into data stored in a database storage format as structured data corresponding to the unstructured data in the spreadsheet document format;

[0135] When the target unstructured data is user-defined unstructured data, the data processing component that matches the form data content selected by the user can be determined, and the form data content selected by the user and the user business data content corresponding to the corresponding data processing component can be converted into database format data as structured data corresponding to the user-defined unstructured data through the determined data processing component.

[0136] Step S303, processing the second target business data according to the first target question raised by the user to determine the user's intention;

[0137] For example, according to the first target question raised by the user, the process of processing the second target business data and determining the user's intention may include the following:

[0138] (1) according to each task item in the second target business data, setting parameters of a semantic recognition model corresponding to the target task item corresponding to the first target problem;

[0139] (2) identifying the user's historical access data on different business platforms based on the second target business data;

[0140] (3) analyzing the context of the user's historical access data, improving the first target question based on the target task items and their corresponding original documents, and determining the second target question raised by the user, wherein the second target question is related to the first target question;

[0141] In the actual application process, the context of the user's historical access data can also be set to periodic denoising processing, where the periodic denoising processing can include removing duplicate and erroneous information. Processing abnormal data can ensure the quality and accuracy of historical records. Denoising processing helps to improve the subsequent understanding and answer quality of user questions.

[0142] By regularly denoising the context of historical access data, we can combine the steps of supplementing and improving the questions raised by users, extracting the core of the questions, converting structured data and semantic analysis to better understand user needs and provide more accurate and personalized answers, thereby improving user experience and the intelligence level of the system.

[0143] (4) extracting keywords for the second target question and converting the extracted results into structured data for the second target question;

[0144] (5) Based on the parameters of the semantic recognition model corresponding to the target task item and the structured data of the second target question, the semantic recognition model corresponding to the target task item is used to perform semantic analysis on the business data corresponding to the target task item to determine the user's intention.

[0145] Step S304, processing the second target business data according to the user's intention, and determining at least one hypothetical answer corresponding to the user's intention;

[0146] For example, according to the user's intention, the process of processing the second target service data and determining at least one hypothetical answer corresponding to the user's intention may include the following:

[0147] (1) determining, based on the user's intention and the second target business data, a task description of the task item corresponding to the user's intention, a category of the second target question raised by the user, and a scope of a hypothetical answer to the second target question;

[0148] Determining the task description of the task item corresponding to the user's intention, the category of the second target question raised by the user, and the scope involved in the hypothetical answer to the second target question can help the data processing model determine the field to which the question raised by the user belongs, the relevant knowledge base, and the types of answers that may be involved.

[0149] (2) extracting first target feature data of the task item corresponding to the user's intention based on the task description of the task item corresponding to the user's intention, the category of the second target question, and the scope involved in the hypothetical answer to the second target question;

[0150] (3) converting the characteristic data of the first target into characteristic data of the second target in a preset format;

[0151] The preset format may be a format acceptable to the data processing model tool;

[0152] (4) Analyze the second target feature data and generate at least one hypothetical answer corresponding to the task item corresponding to the user's intention.

[0153] Step S305, evaluating and ranking each generated hypothetical answer, and determining a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers;

[0154] For example, the process of evaluating and ranking each generated hypothetical answer and determining a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers may include the following:

[0155] (1) determining a weight coefficient of each evaluation dimension of each generated hypothetical answer according to the user's intention, wherein the evaluation dimension includes accuracy, completeness, semantic coherence, and information reliability of each hypothetical answer generated for the second target question;

[0156] (2) evaluating each generated hypothetical answer according to a preset evaluation dimension and a preset scoring method to obtain an evaluation result of each generated hypothetical answer. For example, the preset scoring method may include: a weight-based scoring method, a similarity matching-based scoring method, a machine learning-based scoring method, a deep learning-based scoring method, or a rule-based scoring method;

[0157] (3) ranking the matching degree of all the hypothetical answers with the second target question according to the evaluation result of each hypothetical answer, and obtaining the matching degree ranking result of each hypothetical answer;

[0158] (4) Based on the matching ranking result of each hypothetical answer, a first target hypothetical answer that best matches the user's intention is determined from the generated hypothetical answers.

[0159] For example, each hypothetical answer can be scored based on the target dimensions of the preset hypothetical answers; wherein the target dimensions may include the accuracy, completeness, fluency, credibility, etc. of the hypothetical answers, and each hypothetical answer is quantitatively scored based on these dimensions.

[0160] According to the importance of each dimension, different weights are assigned to different dimensions by weighted average. The weight ratio of each dimension is determined according to business needs and user expectations;

[0161] For each hypothetical answer, a comprehensive score is calculated based on the scores and weights of each dimension, which will reflect the overall quality of the answer. Then, all hypothetical answers are sorted according to the comprehensive score in order to select the optimal solution;

[0162] Finally, the hypothetical answer with the highest score is selected as the final output based on the sorted results. This process ensures that the selected answer reaches a relatively high level in each target dimension, improving the accuracy and reliability of the question-answering system.

[0163] Step S306, based on the first target hypothetical answer, the original document corresponding to the user's intention is processed to determine the second target hypothetical answer corresponding to the user's intention as the final question-and-answer result corresponding to the first target question.

[0164] It can be seen from the technical solutions introduced above that the method provided in the embodiment of the present application can support the processing of intelligent questions of multiple users on multiple different business platforms, and can analyze and process the business data of different users on different business platforms according to the intentions of different users, so as to generate multiple hypothetical answers corresponding to the task items corresponding to the users on different business platforms for users to choose from, and also effectively improve the accuracy and reliability of intelligent question and answer by evaluating and sorting multiple hypothetical answers, and the device provided in the embodiment of the present application can support multiple data modalities, including data processing in different forms such as text, voice, and image, and can obtain user intentions from different types of input data, and generate diversified and targeted hypothetical answers based on the generated user intentions, which can effectively improve user satisfaction and experience, provide users with more accurate and comprehensive question and answer results, improve the intelligence level of the intelligent question and answer system, and at the same time improve the efficiency of users using intelligent question and answer tools, so as to achieve the effect of reducing costs and increasing efficiency.

[0165] Next, combine Figure 3-Figure 4 , introduces the implementation process of the multimodal zero-code form modeling intelligent question-answering method provided in the embodiment of the present application, as follows:

[0166] As attached Figure 3-4 As shown, it is a specific application of the multi-modal zero-code form modeling intelligent question-answering system provided in the embodiment of the present application:

[0167] (1) The user's form data is obtained through the user's custom configuration to obtain the required components, and then converted into the storage format required by the structured database, for example, converted into the storage format required by the MySQL database.

[0168] Specifically, when a user selects a form that requires question and answer (QA), the metadata information of the user's QA form can be obtained. Through the metadata information of the user's QA form, the user can select the component field content information that is understood by the big model and asks questions. The components included are: text component, member information component, date component, amount component, geographic location component, single-select / multiple-select component, association relationship component, etc. After determining the required components, these components need to be adapted to the data types supported by the database, and adapted to the type and length description corresponding to the database according to the component information, and automatically extract keywords from the component names, and generate pinyin or translate into English to adapt to the field names of the database.

[0169] (2) Extracting the document content from the document information uploaded by the user, where the document content may include OCR form information. After extracting the document content, the extracted content needs to be segmented into blocks, and word segmentation is performed on each block to extract the part of speech of the document content.

[0170] For example, the large amount of extracted text or data can be divided into smaller, easier-to-process "chunks" according to certain rules. This allows for subsequent operations such as grammatical analysis and keyword extraction. For example, when processing a long document, the document can be divided into chunks by paragraphs, and then feature extraction and classification judgment can be performed on each paragraph. Furthermore, in data storage and transmission, dividing the extracted data into chunks of appropriate sizes can be stored and transmitted more efficiently.

[0171] (3) When a user asks a question based on unstructured data, the user's intention can be identified to confirm whether the user intends to ask and answer questions or to collect statistical form information.

[0172] (4) When users ask questions based on unstructured data, the questions raised by the users can be converted into a language that the database can understand.

[0173] (5) During the data retrieval process, multiple large models need to be coordinated so that hypothetical answers corresponding to the questions can be generated.

[0174] For example, it is necessary to use the big model to obtain the keywords in the question, and then obtain the corresponding documents through keyword and vector search. It is necessary to use the big model to compare the matching degree between the searched document and the hypothetical answer provided, based on the document content. Then the document is further split into real-time description language, and the split description is given to the big model instead of the original record, and finally the big model outputs the result. The output result takes into account the understanding of the information content of OCR information and images.

[0175] (6) When a user asks to generate statistical chart records based on unstructured data, the user's question can be converted into a function based on business implementation and adapted for output.

[0176] For example, based on the questions raised by users, a large model is used to analyze the users' questions to confirm whether the users' questions are to generate answers to questions or to generate business-based statistical charts.

[0177] If the answer is based on questions and answers, it is necessary to use a large model to obtain the keyword information of the user's questions and answers, and then generate SQL statements that the database can understand based on the form metadata information. Pre-set prompt words are used to ensure that SQL can execute related mathematics, date operations, multi-table operations, etc.

[0178] If it is a business-based statistical chart, you can extract the keyword information from the user's question, convert the question data according to the keyword information, and convert the question into fact dimension display, indicator description, data grouping, data condition filtering, etc., so that it is more in line with the requirements for generating statistical charts. Design a suitable visualization solution based on the user's questions and data characteristics. For example, it can include selecting the appropriate statistical chart type, layout and color matching. According to the questions raised by the user, customize the visualization solution to make it more in line with the user's aesthetic and business needs. After determining the visualization solution, it can be converted into a specific implementation function. These implementation functions can perform corresponding statistics and calculations on the data according to the user's needs and generate statistical charts. Customized code can also be written for the user's questions to ensure the accuracy and beauty of the statistical charts. After generating the statistical chart, it can also be output adapted. For example, it can include adjusting the chart size, format and clarity to ensure the display effect of the chart in different devices and scenarios. That is, according to the user's question and answer requirements, provide a variety of output formats and adaptation solutions to meet the user's usage needs in different scenarios, and the generated statistical charts can also be used for image download, business adaptation, etc. Finally, continue to pay attention to changes in user needs and provide users with continuous services and support.

[0179] (7) When users solve problems based on documents, they can retrieve the most similar data blocks based on the user's similarity matching, and reorder the retrieved data blocks. Finally, the large model understands and outputs the acquired data.

[0180] It can be seen that the multimodal zero-code form modeling intelligent question-answering method provided by this application can support multiple data modalities, including text, voice, image and other forms, and can obtain user intentions from different input data and generate diversified answers. According to the intentions and needs of different users, the system can call suitable data processing model tools to generate personalized and targeted hypothetical answers to improve user satisfaction and experience.

[0181] In the actual application process, the implementation process of the multi-modal zero-code form modeling intelligent question-answering method provided in the embodiment of the present application can be summarized as follows:

[0182] (1) The collected unstructured data can be trained into structured data through user-defined configuration;

[0183] (2) Through user-defined configuration, text documents can be trained into a vector database, and the data can be divided, calculated, and summarized for storage;

[0184] (3) All unstructured data and text document data can be trained in real time and on a regular basis.

[0185] Among them, for unstructured documents such as PDF, a neural network-based PDF parser can be used. First, the regular text content of the PDF is obtained, and then the data content in the PDF, such as table OCR, is parsed into text content similar to markdown. Then the image content in the PDF is obtained, and the image is represented as a text description by performing text recognition and image analysis on the image through a large model. Finally, by setting a variety of data storage vectorization storage strategies, such as segmented summary storage, the text content of the PDF obtained by parsing is stored in a vector database.

[0186] 2) For structured documents like Excel, by representing each row of data and header information in Excel as a database storage format, when conducting intelligent question and answer, the data can be accessed through the function call of the large model.

[0187] 3) For the user's unstructured data, when the user selects the required form for Q&A, the components required for Q&A are selected in the form, and then the form metadata information and the user business data content corresponding to the component are represented as accessible database format data, and the data integrity is ensured through incremental synchronization and scheduled synchronization methods.

[0188] In particular, OCR and image information can be saved as text.

[0189] (4) Through user configuration, the parameters required for model operation can be specified to guide the behavior of the model;

[0190] (5) When the user performs natural language processing, the user's intention is identified to confirm whether to generate a chart or perform a natural language understanding answer;

[0191] (6) Process the context of the user's historical access data, improve and rewrite the questions asked by the user, and nest the questions asked by the user to achieve more accurate intent processing;

[0192] (7) Regularly organize the context of the user's historical access data to filter out noise data

[0193] (8) Extract the core features of the questions raised by users and convert them into the answers and answer requirements that need to be made;

[0194] (9) Use data processing model tools to analyze users’ business data to answer users’ questions, convert users’ questions into text2sql and text2API, and retrieve questions through vector databases

[0195] (10) Rearrange the retrieved files;

[0196] (11) The retrieved data is understood and output by the large model;

[0197] (12) During the data retrieval process, different large models can be used to perform hypothetical answers to questions raised by users and generate at least one answer result. Furthermore, the questions raised by users can be parsed, keywords can be extracted, and document fragments can be obtained from the vector database based on keyword search and vector search. The large model is used to compare the hypothetical answers to the questions and the retrieved document fragments, and the document fragments corresponding to the scores above the set threshold are retained through score comparison. The retrieved document fragments are then split into descriptions of atomic content facts. Finally, based on the split atomic fragments and the answers to the user's questions, the large model summarizes the output content. In particular, when searching, the information output corresponding to OCR and images can be considered.

[0198] (13) Different models can be used for cross-processing at different steps.

[0199] The multimodal zero-code form modeling intelligent question and answer device provided in an embodiment of the present application is described below. The multimodal zero-code form modeling intelligent question and answer device described below and the multimodal zero-code form modeling intelligent question and answer method described above can be referenced to each other.

[0200] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of a multimodal zero-code form modeling intelligent question-and-answer device disclosed in an embodiment of the present application.

[0201] like Figure 5 As shown, the multimodal zero-code form modeling intelligent question-answering device may include:

[0202] The data collection unit 101 is used to collect first target business data corresponding to the user on different business platforms, wherein the first target business data includes different task items corresponding to the user on different business platforms and the original document corresponding to each task item;

[0203] A structured processing unit 102, configured to perform structured processing on the first target service data to obtain second target service data;

[0204] The intention determination unit 103 is used to process the second target business data according to the first target question raised by the user to determine the user's intention;

[0205] An answer generating unit 104 is used to process the second target service data according to the user's intention and determine at least one hypothetical answer corresponding to the user's intention;

[0206] The answer evaluation unit 105 is used to evaluate and sort each generated hypothetical answer, and determine a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers;

[0207] The answer determination unit 106 is used to process the original document corresponding to the user's intention based on the first target hypothetical answer, and determine the second target hypothetical answer corresponding to the user's intention as the final question and answer result corresponding to the first target question.

[0208] It can be seen from the technical solutions introduced above that when it is necessary to provide users with intelligent question and answer services, the device provided in the embodiment of the present application can support the processing of intelligent questions of multiple users on multiple different business platforms, and can analyze and process the business data of different users on different business platforms according to the intentions of different users, so as to generate multiple hypothetical answers corresponding to the task items corresponding to the users on different business platforms for users to choose from, and also effectively improve the accuracy and reliability of intelligent question and answer by evaluating and sorting multiple hypothetical answers, and the device provided in the embodiment of the present application can support multiple data modalities, including data processing in different forms such as text, voice, and image, and can obtain user intentions from different types of input data, and generate diversified and targeted hypothetical answers based on the generated user intentions, which can effectively improve user satisfaction and experience, provide users with more accurate and comprehensive question and answer results, and improve the intelligence level of the intelligent question and answer system. At the same time, it can improve the efficiency of users using intelligent question and answer tools and achieve the effect of reducing costs and increasing efficiency.

[0209] Among them, the specific processing flow of each unit included in the above-mentioned multimodal zero-code form modeling intelligent question and answer device can refer to the relevant introduction of the multimodal zero-code form modeling intelligent question and answer method in the previous article, and will not be repeated here.

[0210] The multimodal zero-code form modeling intelligent question-answering device provided in the embodiment of the present application can be applied to multimodal zero-code form modeling intelligent question-answering devices, such as terminals: mobile phones, computers, etc. Optionally, Figure 6 The hardware structure diagram of the multi-modal zero-code form modeling intelligent question-answering device is shown. Figure 6 The hardware structure of the multimodal zero-code form modeling intelligent question-answering device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4.

[0211] In the embodiment of the present application, the number of the processor 1 , the communication interface 2 , the memory 3 , and the communication bus 4 is at least one, and the processor 1 , the communication interface 2 , and the memory 3 communicate with each other through the communication bus 4 .

[0212] The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application, etc.;

[0213] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory, such as at least one disk memory;

[0214] Among them, the memory stores a program, and the processor can call the program stored in the memory, and the program is used to: implement each processing flow in the aforementioned terminal multi-modal zero-code form modeling intelligent question and answer solution.

[0215] An embodiment of the present application also provides a readable storage medium, which can store a program suitable for execution by a processor, and the program is used to: implement various processing flows of the aforementioned terminal in a multi-modal zero-code form modeling intelligent question and answer solution.

[0216] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0217] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0218] The above description of the disclosed embodiments enables professionals and technicians in the field to implement or use the present application. Various modifications to these embodiments will be apparent to professionals and technicians in the field, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. The various embodiments can be combined with each other. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multimodal zero-code form modeling intelligent question-answering method, characterized in that: include: Collecting first target business data corresponding to the user on different business platforms, wherein the first target business data includes different task items corresponding to the user on different business platforms and original documents corresponding to each task item; Performing structured processing on the first target business data to obtain second target business data; According to the first target question raised by the user, the second target business data is processed to determine the user's intention; According to the user's intention, the second target service data is processed to determine at least one hypothetical answer corresponding to the user's intention; Evaluate and rank each generated hypothetical answer, and determine a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers; Based on the first target hypothetical answer, the original document corresponding to the user's intention is processed to determine a second target hypothetical answer corresponding to the user's intention as the final question-and-answer result corresponding to the first target question.

2. The method according to claim 1, characterized in that The performing structured processing on the first target service data to obtain second target service data includes: Determine a processing strategy for the first target business data based on the user's customized configuration information, wherein the user's customized configuration information includes configuration information for abnormal data processing, configuration information for recollection strategies for various types of data collected with abnormalities, configuration information for log records of errors in data collection and processing, configuration information for a notification mechanism when errors occur, configuration information for outputting processing results, configuration information for format requirements for output data, and configuration information for setting naming rules and storage paths for output files; Locating the positions of each unstructured data of the first target business data according to the processing strategy for the first target business data; Determining, according to the positions of the respective unstructured data of the first target business data, respective target unstructured data and their data types that need to be processed in the first target business data; According to the processing strategy of each target unstructured data, after cleaning and formatting the target unstructured data, according to the data type of each target unstructured data, determine the data processing tool corresponding to the data structure type of each target unstructured data; Each non-target structured data is structured by using a data processing tool of the data structure type of each target unstructured data to obtain structured data corresponding to each target unstructured data as the second target business data.

3. The method according to claim 1, characterized in that The performing structured processing on the first target service data to obtain second target service data includes: Determine each unstructured data of the first target business data and its data type; Determine, according to the data type of each unstructured data, a data processing tool corresponding to the data structure type of each unstructured data; Each unstructured data is structured by using a data processing tool of the data structure type of each unstructured data to obtain structured data corresponding to each unstructured data; The structured data corresponding to all the unstructured data corresponding to the first target business data is used as the second target business data.

4. The method according to claim 1, characterized in that: The processing of the second target business data according to the first target question raised by the user to determine the user's intention includes: According to each task item in the second target business data, setting parameters of a semantic recognition model corresponding to the target task item corresponding to the first target problem; Identifying historical access data of the user on different business platforms based on the second target business data; Analyze the context of the user's historical access data, improve the first target question according to the target task item and its corresponding original document, and determine the second target question raised by the user, wherein the second target question is related to the first target question; Extracting keywords from the second target question and converting the extracted results into structured data of the second target question; Based on the parameters of the semantic recognition model corresponding to the target task item and the structured data of the second target problem, the semantic recognition model corresponding to the target task item is used to perform semantic analysis on the business data corresponding to the target task item to determine the user's intention.

5. The method according to claim 4, characterized in that The processing of the second target service data according to the user's intention to determine at least one hypothetical answer corresponding to the user's intention includes: Determine, based on the user's intention and the second target business data, a task description of the task item corresponding to the user's intention, a category of the second target question raised by the user, and a scope of a hypothetical answer to the second target question; Extracting first target feature data of the task item corresponding to the user's intention according to the task description of the task item corresponding to the user's intention, the category of the second target question, and the scope involved in the hypothetical answer to the second target question; Converting the characteristic data of the first target into characteristic data of a second target in a preset format; The second target feature data is analyzed to generate at least one hypothetical answer corresponding to the task item corresponding to the user's intention.

6. The method according to claim 2, characterized in that The method of processing each non-target structured data into a structured data by using a data processing tool of the data structure type of each target non-structured data to obtain structured data corresponding to each target non-structured data as the second target business data includes: When the target unstructured data is unstructured data in a portable document format, first text content in the unstructured data in the portable document format is obtained, and data content corresponding to a table in the unstructured data in the portable document format is parsed into second text content, and content corresponding to an image in the unstructured data in the portable document format is identified as third text content, and the first text content, the second text content, and the third text content are stored according to a preset data storage vectorization storage strategy as structured data corresponding to the unstructured data in the portable document format; When the target unstructured data is data in a spreadsheet document format, each row of data and header information of the data in the spreadsheet document format is parsed into data stored in a database storage form as structured data corresponding to the unstructured data in the spreadsheet document format; When the target unstructured data is user-defined unstructured data, a data processing component that matches the form data content selected by the user is determined, and the form data content selected by the user and the user business data content corresponding to the corresponding data processing component are converted into database format data as structured data corresponding to the user-defined unstructured data through the determined data processing component.

7. The method according to claim 4, characterized in that The step of evaluating and ranking each generated hypothetical answer and determining a first target hypothetical answer that best matches the user's intention from the generated hypothetical answers includes: Determining a weight coefficient of each evaluation dimension of each generated hypothetical answer according to the user's intention, wherein the evaluation dimension includes accuracy, completeness, semantic coherence, and information reliability of each hypothetical answer generated for the second target question; According to the preset evaluation dimension and the preset scoring method, each generated hypothetical answer is evaluated to obtain an evaluation result of each generated hypothetical answer; According to the evaluation result of each hypothetical answer, the matching degree of all the hypothetical answers with the second target question is sorted to obtain the matching degree sorting result of each hypothetical answer; According to the matching ranking result of each hypothetical answer, a first target hypothetical answer that best matches the user's intention is determined from the generated hypothetical answers.

8. A multimodal zero-code form modeling intelligent question-answering system, characterized in that: The multimodal zero-code form modeling intelligent question-answering method applied to any one of claims 1-7, the multimodal zero-code form modeling intelligent question-answering system comprises: a data acquisition module, a first processing module, a model management module, a second processing module, and a third processing module; The data collection module is responsible for collecting the first target business data corresponding to the user on different business platforms, and performing structured processing on the first target business data to obtain the second target business data and transmit it to the first processing module, wherein the first target business data includes different task items corresponding to each user on each business platform and the original document corresponding to each task item; The first processing module receives and processes the second target service data, determines the user's intention, and transmits it to the second processing module; The second processing module receives and, based on the user's intention, calls the data processing model tool corresponding to the user's intention in the model management module to process the second target business data, generates at least one hypothetical answer corresponding to the task item corresponding to the user's intention, evaluates each generated hypothetical answer, and determines a first target hypothetical answer that best matches the user's intention from all generated hypothetical answers, and sends the first target hypothetical answer and its data processing model tool to the third processing module, wherein the model management module includes at least one data processing model tool; The third processing module processes the original document corresponding to the user's intention based on the first target hypothetical answer and its data processing model tool to determine the final question and answer result corresponding to the user's intention.

9. A multi-modal zero-code form modeling intelligent question-answering device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, implement the steps of the multimodal zero-code form modeling intelligent question-answering method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors implement the steps of the multimodal zero-code form modeling intelligent question-answering method as described in any one of claims 1 to 7.

Images