Data-driven text automatic generation method and device

Abstract

The data-driven text automatic generation method and device provided by the present application relate to the field of data processing technology, configure the search engine matrix and construct the automatic generation module, determine the pre-search terms and search the document set, extract the document information system, analyze the matching degree and perform extended search to determine the effective information system. Match the target template, perform information homology fitting and hierarchical layout, obtain the generated text, solve the technical problems that the information coverage of the source file in the prior art cannot be guaranteed, and the layout typesetting adaptability is not high, resulting in insufficient fit between the text generation result and the demand, and perform unified standardized retrieval through auxiliary word bag library conversion, introduce the search engine matrix and construct the automatic generation module, use multi-layer extended indexing and system extraction to improve the integrity and processing efficiency of the source document, perform template hierarchical layout and layout information conversion to ensure the quality of text generation.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data-driven automatic text generation method and device. Background Art

[0002] With the introduction of artificial intelligence, the combination of artificial intelligence and data-driven can assist in document text generation to improve efficiency. However, due to the influence of factors such as the system's own intelligence, the existing technology requires pre-defined macro commands and combined with AI tools, which has certain limitations. At present, there is a lack of a systematic automatic manuscript writing method that can take into account both content requirements and typesetting layout requirements to improve the quality of automatic text generation.

[0003] The existing technology has technical problems such as the source file's information coverage cannot be guaranteed and the layout and typesetting adaptability is not high, resulting in the text generation results not being sufficiently consistent with the requirements. Summary of the invention

[0004] The present application provides a data-driven text automatic generation method and device, which is used to solve the technical problems existing in the prior art that the information coverage of the source file cannot be guaranteed and the layout and typesetting adaptability is not high, resulting in insufficient fit between the text generation result and the demand.

[0005] In view of the above problems, the present application provides a data-driven text automatic generation method and device.

[0006] In a first aspect, the present application provides a data-driven text automatic generation method, the method comprising:

[0007] Configure a search engine matrix, and train an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block;

[0008] Read the text generation requirements, determine the pre-search terms in combination with the word bag library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set;

[0009] Receive self-uploaded documents, combine the retrieved document set, extract information entities and information relationships and reconstruct content based on the graph extraction model, and determine the document information system, wherein the graph extraction model establishes a communication connection with the automatic generation module;

[0010] Traversing the document information system, performing matching analysis in combination with the search management block, locating the expanded search target, and performing secondary search and information system reconstruction to determine the effective information system;

[0011] Traverse the template library, balance the subjective standards based on the text generation requirements and the fit of the information system, and match the target template;

[0012] The effective information system and the target template are transferred to the reconstruction management block, and information homology fitting and hierarchical layout are performed to obtain the generated text.

[0013] In a second aspect, the present application provides a data-driven text automatic generation device, the device comprising:

[0014] A construction unit, the construction unit is used to configure a search engine matrix, and train an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block;

[0015] A retrieval unit, the retrieval unit is used to read the text generation requirements, determine the pre-search terms in combination with the word bag library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set;

[0016] A system extraction unit, the system extraction unit is used to receive self-uploaded documents, extract information entities and information relationships and reconstruct content based on the search document set, and determine the document information system based on the graph extraction model. The graph extraction model establishes a communication connection with the automatic generation module;

[0017] An extended search unit, the extended search unit is used to traverse the document information system, perform matching analysis in combination with the search management block, locate the extended search target and perform secondary search and information system reconstruction to determine the effective information system;

[0018] A template matching unit, the template matching unit is used to traverse the template library, balance the subjective standard based on the text generation requirement and the information system fit, and match the target template;

[0019] A text generation unit is used to transfer the effective information system and the target template to the reconstruction management block, perform information homology fitting and hierarchical layout, and obtain generated text.

[0020] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0021] The data-driven text automatic generation method and device provided in the embodiment of the present application configure a retrieval engine matrix, train an automatic generation module in combination with the retrieval engine matrix, read text generation requirements, determine pre-retrieval terms in combination with a bag-of-words library, match a target retrieval engine based on the retrieval engine matrix, perform a paper pool search to determine a retrieval document set, receive self-uploaded documents, extract information entities and information relationships and reconstruct content based on a graph extraction model, determine a document information system, and then perform a matching degree analysis in combination with the retrieval management block, locate and expand search targets, perform secondary retrieval and information system reconstruction, and determine a valid information system. The template library is traversed, the subjective standards based on the text generation requirements are balanced with the information system fit, the target template is matched, and the effective information system is transferred to the reconstruction management block to perform information homology fitting and hierarchical layout to obtain the generated text. This solves the technical problems in the prior art that the information coverage of the source file cannot be guaranteed and the layout and typesetting adaptability is not high, resulting in insufficient fit between the text generation results and the requirements. A unified standardized retrieval is performed through auxiliary word bag library conversion, a retrieval engine matrix is ​​introduced and an automatic generation module is constructed. Multi-layer extended indexing and system extraction are used to improve the integrity and processing efficiency of the source document, and template hierarchical layout and layout information conversion are performed to ensure the quality of text generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 A flow chart of a data-driven automatic text generation method is provided for this application;

[0023] Figure 2 A structural schematic diagram of a data-driven text automatic generation device is provided for this application.

[0024] Explanation of the accompanying drawings: construction unit 11, retrieval unit 12, system extraction unit 13, extended retrieval unit 14, template matching unit 15, text generation unit 16. DETAILED DESCRIPTION

[0025] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0026] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0027] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict, and the terms "first\second" involved are merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "including" and "having" and any variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by technicians in the technical field of this application. The terms used herein are for the purpose of describing the embodiments of the present application only.

[0028] This application provides a data-driven text automatic generation method and device, configures a search engine matrix and constructs an automatic generation module, determines pre-search terms and retrieves document sets, extracts document information systems, analyzes matching degrees and conducts extended searches to determine effective information systems. Matches target templates, performs information homology fitting and hierarchical layout, and obtains generated text, which is used to solve the technical problems in the prior art that the information coverage of source files cannot be guaranteed, and the layout and typesetting adaptability is not high, resulting in insufficient fit between text generation results and requirements.

[0029] Embodiment 1: Figure 1 As shown, the present application provides a data-driven text automatic generation method, the method comprising:

[0030] S1: configuring a search engine matrix, and training an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block.

[0031] The introduction of artificial intelligence has provided convenience for document text generation and can effectively improve efficiency. However, due to the influence of factors such as the system's own intelligence, the existing technology still has certain limitations. The information coverage of the source file cannot be guaranteed, and the layout and typesetting adaptability is not high, resulting in insufficient fit between the text generation results and the requirements. This application provides a data-driven text automatic generation method and device, assists in bag-of-words conversion for unified standardized retrieval, introduces a retrieval engine matrix and constructs an automatic generation module, uses multi-layer extended indexing and system extraction to improve the integrity and processing efficiency of the source document, and performs template layered layout and layout information conversion to ensure the quality of text generation.

[0032] The search engine matrix is ​​composed of multiple search engines, each of which has its own unique search function and domain characteristics, forming a complete search system. Based on the relative relationship of the engines, such as the superior-subordinate relationship, a matrix arrangement is performed. The most suitable search engine is selected according to different query requirements or data types to optimize the search effect and efficiency. The search management module is constructed based on the search engine matrix, and data-driven training is performed until convergence.

[0033] Specifically, obtain historical retrieval samples, and use data retrieval, incomplete retrieval point positioning, and hierarchical retrieval as logic. Based on the historical retrieval samples, divide the retrieval tracking part, divide the training set and the validation set, perform supervised training based on the training set and cross-validation based on the validation set, and obtain the retrieval management block that meets the convergence condition. Exemplarily, the convergence condition can be the training accuracy. Similarly, based on the information system extraction and reconstruction part of the historical retrieval samples, supervise the training of the reconstruction management block to perform entity extraction and data relationship extraction of the retrieved documents, and build a data system based on key valid information to reduce information redundancy and improve subsequent processing efficiency. Pre- and post-associate the retrieval management block and the reconstruction management block to generate the automatic generation module.

[0034] In the embodiment of the present application, by establishing an efficient and collaborative system module, it is possible to utilize diversified search tools and effectively analyze, manage and transform the retrieved data through the automatic generation module, thereby supporting the automatic generation of high-quality text. It is particularly suitable for processing large-scale data sets, such as academic research, technical document generation, etc.

[0035] Among them, the configuration of the search engine matrix, step S1 also includes: setting the engine search characteristics, wherein the engine search characteristics are determined based on at least the search method and the search domain; traversing the engine search characteristics, performing search engine configuration and integration association, and determining the search engine matrix, wherein the search engine matrix has relative independence and coordination, and the coordination includes upper and lower level association coordination and same-level concurrent coordination.

[0036] In an embodiment of the present application, the engine retrieval characteristics must be adapted to the retrieval requirements to ensure the retrieval effect. Among them, the engine can perform queries according to different retrieval methods, such as keyword retrieval, full-text retrieval, semantic retrieval, etc. Each method is suitable for different query requirements and data types. Determine the retrieval scope or field that each engine can cover. For example, some engines may focus on specific academic fields, while others cover broader content areas, in order to determine the retrieval method and the retrieval domain for the configuration of the retrieval engine.

[0037] Furthermore, the engine retrieval characteristics are traversed and analyzed in order to configure and integrate these engines. Specifically, according to the characteristics of each engine and system requirements, corresponding configurations are performed, including functions, underlying logic, execution rules, etc., to ensure that each engine can play the best role in the system. At the same time, the engines need to be able to work together, and through appropriate integration and association, ensure that they can effectively exchange information and jointly process query tasks.

[0038] In this application, the retrieval needs are taken as the guide, including small-scale independent retrieval, large-scale collaborative retrieval, or multi-functional dimension collaborative retrieval, etc. Therefore, each engine has a certain degree of independence in performing its functions, and can independently handle specific query tasks even if they run in the same system. At the same time, engines at different levels can work together through hierarchical relationships. For example, the upper-level engine can guide the lower-level engine to perform specific retrieval tasks. Multiple engines at the same level can work in parallel to improve the overall retrieval efficiency and response speed.

[0039] By building a flexible and efficient retrieval engine system, we can select the most appropriate engine according to different query requirements and data types, and ensure the performance and availability of the overall system through collaborative work.

[0040] S2: Read the text generation requirements, determine the pre-search terms in combination with the bag-of-words library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set.

[0041] First, the text generation requirements entered by the user are read and understood. This may include information such as the type of text, subject, scope, and required level of detail. The bag-of-words library is a tool that converts subjective semantics into standardized objective terms commonly used by the engine, which can effectively avoid understanding bias caused by subjective language and avoid retrieval bias. It is a collection of pre-defined keywords or phrases that are associated with specific topics or fields.

[0042] Furthermore, the text generation requirement is identified, the pre-search terms are determined in combination with the word bag library, and one or more search engines that best suit the current requirement are matched for search based on the characteristics of the pre-search terms and the search engine. Preferably, the matching process needs to consider the search method, domain and efficiency of the search engine, that is, for the search engine matrix, the optimal combination is selected from multiple configured engines to achieve the most efficient search.

[0043] Then, use the pre-search terms to conduct an in-depth literature search in the paper pool. Determine all relevant documents for subsequent information extraction and text generation steps. Preferably, the search volume can be constrained by setting a search ratio to avoid too many or too few total documents. Obtain the most relevant and highest quality information resources according to user needs, laying a solid foundation for generating high-quality customized texts.

[0044] Among them, the pre-search terms are determined in combination with the word bag library, and step S2 also includes: identifying the text generation requirements, performing semantic recognition and key search point reading and grading, and determining the original search terms, wherein the grading standard is the key degree of the search point, configuring the search ratio based on the grading result, and the search ratio is positively correlated with the search point level; traversing the word bag library, matching the original search terms, performing search term conversion, and determining the pre-search terms.

[0045] First, identify and understand the text generation requirements proposed by users, such as core topics, specific information points, and related contexts. Specifically, through advanced natural language processing technology, the semantic structure of the required text is analyzed to accurately capture key concepts and topics. Further, key retrieval points are extracted from the semantic analysis results, that is, those entries or phrases that are most critical to generating specific texts. These key retrieval points are graded based on their criticality, that is, the importance of each retrieval point in meeting the text generation requirements. The higher the criticality, the higher the corresponding retrieval point level.

[0046] Furthermore, according to the level of the key retrieval point, the corresponding retrieval ratio is configured, that is, the weight of each retrieval point in the overall retrieval strategy is determined. The retrieval ratio is positively correlated with the level of the retrieval point, that is, the retrieval point with a higher level will have a greater influence or higher priority in the retrieval process.

[0047] The word bag library covers a wide range of topics and professional fields, including predefined standardized word sets and word conversion rules, matching the original search terms with the terms in the word bag library to find the most relevant or similar terms. In the specific matching process, the original search terms may be adjusted or converted according to the data in the word bag library to form more accurate or standardized pre-search terms. The finalized pre-search terms will be used in the subsequent detailed search steps to maximize the search efficiency and the relevance of text generation.

[0048] This process not only ensures that the generated text is highly relevant to user needs, but also optimizes the retrieval process, making it more efficient and targeted.

[0049] Among them, the determination of the pre-search terms, step S2 also includes: based on the search engine matrix, determining a standardized objective term pattern; for the objective term pattern, performing conversion mapping between standardized terms and interpreted semantics, and constructing a word bag library; traversing the word bag library, matching the original search terms, determining the converted search terms and performing term diversification, and determining a multi-term group; establishing a mapping between the multi-term group and the original search terms, and determining the pre-search terms.

[0050] In the embodiment of the present application, a search engine matrix is ​​used to define a standardized objective term model, including the standardized forms of various terms, to ensure that the system can use consistent and standardized query terms when performing search tasks. The objective term model is intended to reduce ambiguity during search and improve search accuracy, and avoid search bias caused by non-uniformity and subjectivity.

[0051] The standardized terms based on the search engine matrix are mapped with their semantic meanings to construct a word bag library, which includes all standardized terms and their corresponding detailed explanations and variants to adapt to different query requirements and contexts.

[0052] Based on the constructed word bag library, the original search terms extracted from user needs are matched with the standardized terms in the library to identify the most appropriate standardized terms to ensure the consistency and efficiency of the retrieval process. After identifying the closest standardized terms, these terms are further diversified to generate a set of multi-terms. Exemplarily, the diversified processing can be different semantic angles and usages to enhance the flexibility of the system's response to user queries. The multi-term group provides a richer semantic network to support a wider range of retrieval and a deeper understanding of content.

[0053] Furthermore, a mapping relationship is established from the original search terms to the multi-term group, which ensures that the search activity can make full use of the enriched term set. The system will determine the final pre-search terms based on these mapping relationships, and these terms will be used in the actual document search process to ensure that the user's needs and intentions are accurately covered. This allows the document search process to show higher accuracy and flexibility when dealing with complex and diverse queries.

[0054] S3: Receive self-uploaded documents, combine them with the search document set, extract information entities and information relationships and reconstruct content based on the graph extraction model, and determine the document information system. The graph extraction model establishes a communication connection with the automatic generation module.

[0055] Specifically, receiving documents or files uploaded by users provides additional information sources that meet the needs, which can be used in combination with the retrieved documents to provide more comprehensive data support. The self-uploaded documents are merged with the retrieved document set to ensure the integrity and diversity of information in the text generation process.

[0056] Then, the graph extraction model is used to extract key information entities and the relationships between them from the merged document set. These entities may include names, places, dates, terms, concepts, etc., while relationships describe the interactions and connections between these entities. The graph extraction model can identify and parse the complex structure in the document and convert unstructured data into a structured knowledge graph for further processing and analysis.

[0057] The following is a feasible way to construct the graph extraction model. Specifically, the model can be constructed based on a sample training method, and a sample source document and a demand-oriented sample data system are obtained. A mapping association between the source document and the sample data system is established, and training samples and verification samples are divided. Based on the training samples, supervised training based on a neural network is performed to determine the initial construction model. The verification sample is then used to calibrate the output based on the initial construction model and the data system of the verification sample to determine whether the system difference meets the standard. Otherwise, the sample division and retraining are performed again until the verification meets the standard, and the constructed graph extraction model is obtained.

[0058] Furthermore, based on the graph extraction model, the entities and relationships of the self-uploaded documents and the retrieved document set are extracted and restructured to form a logical and hierarchical document information system. A communication connection is established between the graph extraction model and the automatic generation module to ensure the continuity and real-time nature of the information flow, and to achieve an efficient collaborative working loop between document retrieval, information extraction, and text generation. A large amount of document data is efficiently processed and converted to support the generation of high-quality, information-rich, and structured text content.

[0059] S4: traverse the document information system, perform matching analysis in combination with the search management block, locate the expanded search target, perform secondary search and information system reconstruction, and determine the effective information system.

[0060] Traverse the literature information system to review and analyze these information entities and relationships, and evaluate the consistency of the information system content with the user's text generation needs. If the consistency is low, further expansion of the search is required to fill in the gaps or deficiencies in the information system. Based on the assessment of the coverage and depth of the current information system, ensure that no key information is missed.

[0061] Combined with the search engine matrix, repeat the above search steps to find more relevant documents or data to supplement the existing information system to improve the accuracy and completeness of the information system. After the second search and reconstruction of the information system, it is evaluated again to ensure that it fully meets the user's text generation needs. Otherwise, continue to perform repeated search steps, which can effectively improve the quality and completeness of the information system and ensure that the final generated text can accurately reflect user needs and is highly customizable and practical.

[0062] Among them, the positioning of the expanded search target and the secondary search and information system reconstruction, step S4 also includes: reading the retrieval ratio of the key retrieval point, traversing the document information system, measuring the matching degree, and the matching degree includes measuring the retrieval point matching degree and the global matching degree; based on the matching degree, positioning the expanded search target, retrieving and determining the secondary search document; combining the graph extraction model, extracting the information system and performing matching degree analysis on the secondary search document, performing iterative retrieval analysis until the preset matching degree is met, and integrating the information system to determine the effective information system.

[0063] First, the retrieval ratio is pre-set, that is, the weight of each key retrieval point in the overall retrieval, reflecting the importance of different retrieval points in meeting the text generation requirements.

[0064] Matching analysis is used to evaluate the consistency between the content in the information system and user needs, specifically including the retrieval point matching and the global matching, that is, to evaluate the information coverage and accuracy of each retrieval point and the breadth and completeness of the entire information system, to determine whether there are information gaps or deficiencies, and whether further retrieval and optimization are needed.

[0065] Exemplarily, a single-point matching threshold and a global matching threshold are set as judgment criteria. If the matching degree does not meet the corresponding threshold standard, it is used as the extended search target, that is, a key retrieval point or subject area supported by more information, and a secondary search is required. The specific retrieval method is the same as above, and a secondary search is performed to find documents that supplement and enhance the existing information system, which helps to fill in previously missing information or improve the accuracy of information. Use the graph extraction model to conduct an in-depth analysis of the newly retrieved documents, extract relevant information entities and relationships, and reconstruct the content. The reconstructed information system is analyzed for matching again to evaluate its degree of satisfaction with user needs.

[0066] If the matching degree of the current information system still does not meet the preset standard, the iterative retrieval and analysis process will continue until the predetermined matching degree is met. After multiple rounds of iteration and optimization, the formed information system should be able to fully reflect the needs of users and have a high degree of relevance and accuracy.

[0067] S5: traverse the template library, balance the subjective standards based on the text generation requirements and the fit with the information system, and match the target template.

[0068] In an embodiment of the present application, the template library is integrated and constructed by predefining multiple types of text templates, wherein each template is designed with a specific structure, format, and purpose. Determine the user's subjective criteria, such as text style, tone, specific requirements for detail, etc. When selecting a template, the system needs to take into account the user's subjective criteria and evaluate the fit of each template with the current information system. To ensure that the selected template can effectively display and utilize the integrated information entities and relationships. Exemplarily, the matching degree can be measured by setting multiple evaluation indicators. Templates with high fit can better utilize the data in the information system, thereby generating more accurate and insightful texts.

[0069] Taking into account the subjective criteria and the degree of fit of the information system, the optimal template is matched for text generation. For example, matching decisions are made based on a weighted approach, and matching methods, algorithms, etc. are introduced to assist in determining the target template. For example, the target template includes text layout, chapter division, information presentation, etc., providing a basis for subsequent highly customized text generation.

[0070] S6: transferring the effective information system and the target template to the reconstruction management block, performing information homology fitting and hierarchical layout, and obtaining generated text.

[0071] First, the integrated effective information system and the target template are transferred to the reconstruction management block. The first thing to do is to fit the information homology, which specifically includes adjusting and optimizing the alignment between the information entity and the template to ensure that the information display and content flow are perfectly matched with the template design.

[0072] Fit multi-source information with the same target to remove information redundancy and ensure that the information content is consistent with the structure, language and style of the template, thereby improving the overall consistency and professionalism of the text.

[0073] Further, according to the order of information display, division of chapters, allocation of titles and subtitles, etc., hierarchical layout processing is carried out based on the framework from large to small to obtain the generated text to ensure the high quality, high relevance and high applicability of the text.

[0074] Among them, the information homology fitting and hierarchical layout are performed to obtain the generated text, and step S6 also includes: traversing the effective information system, performing information homology fitting, and determining the fitting information system, wherein the fitting criteria include dimension fusion and redundancy removal; determining a multi-level layout structure based on the target template; traversing the fitting information system, executing the information hierarchical layout based on the multi-level layout structure, and determining the generated text.

[0075] First, perform multi-source data fitting based on the same information target, that is, merge relevant multi-dimensional information in the information system to improve the integration and availability of information; perform redundant system information screening, that is, delete duplicate or unnecessary information to streamline the content and enhance the focus and clarity of the text, so as to adjust the information system to fully align it with the structure and requirements of the target template. Preferably, the fitting process ensures that the information content is compatible with the style and structure of the template, thereby improving the overall quality and consistency of the text.

[0076] Furthermore, a multi-level layout structure is defined, for example, the main chapters, sub-chapters, and the order and manner of information display of the text. The fitted information system is traversed, that is, information is hierarchically laid out in order from large to small based on the framework, preferably, the information is arranged in combination with the importance, relevance, and logical order to determine the generated text. The generated text can effectively ensure the accuracy and relevance of the content, and also improve the overall quality and professionalism of the text through meticulous structural design.

[0077] Among them, the execution of the information hierarchical layout based on the multi-level layout structure, step S6 also includes: determining the global framework, segmenting and mapping the fitting information system, and determining a layer of layout results; reading the first structural partition of a layer of layout results, performing sub-level segmentation and mapping on the mapping information system based on the partition framework, performing iterative hierarchical layout, and determining the text layout result; for the text layout result, performing adaptive conversion of layout information based on data structure and language pattern, and determining the generated text.

[0078] First, it is necessary to determine the global framework of the text, that is, the overall structure and main components of the text. For example, chapter layout. Based on the global framework, the fitting information system is segmented and mapped, and a large amount of information is reasonably distributed to different chapters and parts to determine a layer of layout results. The first structural partition is read from the first layer of layout results, that is, the framework partition under the global framework, such as a sub-chapter, and the information in this partition is further sub-divided and mapped, and the iterative hierarchical layout of each sub-level partition is continued. The process of gradual refinement ensures that the information of each part is displayed logically and effectively through continuous iterative adjustments. Each iteration may adjust the order of information arrangement or redistribute content to achieve the best expression effect.

[0079] Furthermore, for the text layout result, adaptive conversion of layout information is performed, specifically including adjusting the expression of the text according to the data structure and language mode, ensuring the coherence and logic of the language, and obtaining the generated text. It is ensured that the generated text not only meets the requirements in terms of content, but also has a high degree of professionalism and readability in terms of structure and expression.

[0080] The data-driven text automatic generation method and device provided by this application have the following technical effects:

[0081] 1. Configure the search engine matrix and build an automatic generation module, determine the pre-search terms and search the document set, extract the document information system, analyze the matching degree and conduct an extended search to determine the effective information system. Match the target template, perform information homology fitting and hierarchical layout, obtain the generated text, solve the technical problems in the prior art that the information coverage of the source file cannot be guaranteed, and the layout and typesetting adaptability is not high, resulting in insufficient fit between the text generation results and the requirements. Through the auxiliary word bag library conversion, unified standardized retrieval is performed, the search engine matrix is ​​introduced and the automatic generation module is constructed, with multi-layer extended indexing and system extraction to improve the integrity and processing efficiency of the source document, and perform template hierarchical layout and layout information conversion to ensure the quality of text generation.

[0082] 2. By converting the word bag library, we can avoid the understanding bias caused by subjective language and avoid retrieval bias; based on the actual retrieval situation, we can make external retrieval judgments to ensure the information coverage and demand fit of the pre-text generated files; by making a layered layout, we can ensure the orderliness of the layout, and by converting the heterogeneous information and the expression method, we can improve the quality of text generation. Introducing the search engine matrix, automatic generation module and graph extraction model, and assisting the algorithm model, we can improve the processing efficiency and automation level.

[0083] Embodiment 2: Based on the same inventive concept as the data-driven text automatic generation method in the above embodiment, Figure 2 As shown, the present application provides a data-driven text automatic generation device, the device comprising:

[0084] A construction unit 11, the construction unit 11 is used to configure a search engine matrix, and train an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block;

[0085] A retrieval unit 12 is used to read the text generation requirements, determine the pre-search terms in combination with the word bag library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set;

[0086] A system extraction unit 13, the system extraction unit 13 is used to receive self-uploaded documents, extract information entities and information relationships and reconstruct content based on the search document set, and determine the document information system based on the graph extraction model. The graph extraction model establishes a communication connection with the automatic generation module;

[0087] An extended search unit 14, which is used to traverse the document information system, perform matching analysis in combination with the search management block, locate the extended search target, perform secondary search and information system reconstruction, and determine the effective information system;

[0088] A template matching unit 15, the template matching unit 15 is used to traverse the template library, balance the subjective standard based on the text generation requirement and the information system fit, and match the target template;

[0089] The text generation unit 16 is used to transfer the effective information system and the target template to the reconstruction management block, perform information homology fitting and hierarchical layout, and obtain generated text.

[0090] Furthermore, the construction unit 11 also includes the following steps: setting engine search characteristics, wherein the engine search characteristics are determined based on at least a search method and a search domain; traversing the engine search characteristics, performing search engine configuration and integration association, and determining a search engine matrix, wherein the search engine matrix has relative independence and synergy, and the synergy includes superior-subordinate association collaboration and same-level concurrent collaboration.

[0091] Furthermore, the retrieval unit 12 also includes the following steps: identifying the text generation requirements, performing semantic recognition and key retrieval point reading and grading, and determining the original retrieval terms, wherein the grading standard is the retrieval point keyness, configuring the retrieval ratio based on the grading result, and the retrieval ratio is positively correlated with the retrieval point level; traversing the word bag library, matching the original retrieval terms, performing retrieval term conversion, and determining the pre-retrieval terms.

[0092] Furthermore, the retrieval unit 12 also includes the following steps: based on the retrieval engine matrix, determining a standardized objective term pattern; for the objective term pattern, performing conversion mapping between standardized terms and interpreted semantics to construct a bag-of-words library; traversing the bag-of-words library, matching the original search terms, determining converted search terms and performing term diversification to determine a multi-term group; establishing a mapping between the multi-term group and the original search terms to determine the pre-search terms.

[0093] Furthermore, the extended retrieval unit 14 also includes the following steps: reading the retrieval ratio of the key retrieval point, traversing the document information system, measuring the matching degree, and the matching degree includes measuring the retrieval point matching degree and the global matching degree; based on the matching degree, locating the extended search target, retrieving and determining the secondary retrieval document; combining the graph extraction model, extracting the information system and performing matching degree analysis on the secondary retrieval document, performing iterative retrieval analysis until the preset matching degree is met, and integrating the information system to determine the effective information system.

[0094] Furthermore, the text generation unit 16 also includes the following steps: traversing the effective information system, performing information homology fitting, and determining the fitting information system, wherein the fitting criteria include dimension fusion and redundancy removal; determining a multi-level layout structure based on the target template; traversing the fitting information system, executing information hierarchical layout based on the multi-level layout structure, and determining the generated text.

[0095] Furthermore, the text generation unit 16 also includes the following steps: determining a global framework, segmenting and mapping the fitting information system, and determining a layer of layout results; reading a first structural partition of a layer of layout results, performing sub-level segmentation and mapping of the mapping information system based on a partition framework, performing iterative hierarchical layout, and determining a text layout result; for the text layout result, performing adaptive conversion of layout information based on data structure and language pattern to determine the generated text.

[0096] Through the above detailed description of the data-driven automatic text generation method in this specification, those skilled in the art can clearly understand the data-driven automatic text generation method and device in this embodiment. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0097] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. 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 data-driven text automatic generation method, characterized in that: The method comprises: Configure a search engine matrix, and train an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block; Read the text generation requirements, determine the pre-search terms in combination with the word bag library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set; Receive self-uploaded documents, combine the retrieved document set, extract information entities and information relationships and reconstruct content based on the graph extraction model, and determine the document information system, wherein the graph extraction model establishes a communication connection with the automatic generation module; Traversing the document information system, performing matching analysis in combination with the search management block, locating the expanded search target, and performing secondary search and information system reconstruction to determine the effective information system; Traverse the template library, balance the subjective standards based on the text generation requirements and the fit of the information system, and match the target template; The effective information system and the target template are transferred to the reconstruction management block, and information homology fitting and hierarchical layout are performed to obtain the generated text.

2. The data-driven automatic text generation method according to claim 1, characterized in that: The configuration retrieval engine matrix includes: Setting an engine search characteristic, wherein the engine search characteristic is determined based on at least a search method and a search domain; The engine search characteristics are traversed, search engine configuration and integration association are performed, and a search engine matrix is ​​determined, wherein the search engine matrix has relative independence and coordination, and the coordination includes upper and lower level association coordination and same level concurrent coordination.

3. The data-driven automatic text generation method according to claim 1, characterized in that: The combining of the word bag library to determine the pre-search terms includes: Identify the text generation requirements, perform semantic recognition and key search point reading and grading, and determine the original search terms, wherein the grading standard is the key degree of the search point, and configure the search ratio based on the grading result, and the search ratio is positively correlated with the search point level; The bag-of-words database is traversed to match the original search terms, perform search term conversion, and determine the pre-search terms.

4. The data-driven automatic text generation method according to claim 3, characterized in that: The step of determining the pre-search terms includes: Determining a standardized objective term pattern based on the search engine matrix; According to the objective entry mode, the conversion mapping between the standardized entry and the interpreted semantics is performed to construct a word bag library; Traversing the bag-of-words database, matching the original search terms, determining the converted search terms and performing term diversification to determine a multi-term group; A mapping between the multi-term group and the original search terms is established to determine the pre-search terms.

5. The data-driven automatic text generation method according to claim 3, characterized in that: The positioning and expansion of the search target and the secondary retrieval and information system reconstruction include: Read the search ratio of the key search point, traverse the document information system, and measure the matching degree, wherein the matching degree includes measuring the matching degree of the search point and the global matching degree; Based on the matching degree, the expanded search target is located, and the secondary search documents are retrieved and determined; In combination with the graph extraction model, the information system is extracted and the matching degree is analyzed for the secondary search documents, and iterative search analysis is performed until the preset matching degree is met, and the information system is integrated to determine the effective information system.

6. The data-driven automatic text generation method according to claim 1, characterized in that: The step of performing information homology fitting and hierarchical layout to obtain generated text includes: Traversing the effective information system, performing information homology fitting, and determining a fitting information system, wherein the fitting criteria include dimension fusion and redundancy removal; Based on the target template, determining a multi-level layout structure; The fitting information system is traversed, and the information hierarchical layout based on the multi-level layout structure is executed to determine the generated text.

7. The data-driven automatic text generation method according to claim 6, characterized in that: The performing of information hierarchical layout based on the multi-level layout structure includes: Determine a global framework, segment and map the fitting information system, and determine a layer layout result; Read the first structural partition of the one-layer layout result, perform sub-level segmentation mapping on the mapping information system based on the partition framework, perform iterative hierarchical layout, and determine the text layout result; For the text layout result, based on the data structure and language mode, adaptive conversion of layout information is performed to determine the generated text.

8. A data-driven text automatic generation device, characterized in that: The steps for implementing the data-driven automatic text generation method according to any one of claims 1 to 7, wherein the data-driven automatic text generation device comprises: A construction unit, the construction unit is used to configure a search engine matrix, and train an automatic generation module in combination with the search engine matrix, wherein the automatic generation module includes a search management block and a reconstruction management block; A retrieval unit, the retrieval unit is used to read the text generation requirements, determine the pre-search terms in combination with the word bag library, match the target search engine based on the search engine matrix, and perform a paper pool search to determine the search document set; A system extraction unit, the system extraction unit is used to receive self-uploaded documents, extract information entities and information relationships and reconstruct content based on the search document set, and determine the document information system based on the graph extraction model. The graph extraction model establishes a communication connection with the automatic generation module; An extended search unit, the extended search unit is used to traverse the document information system, perform matching analysis in combination with the search management block, locate the extended search target and perform secondary search and information system reconstruction to determine the effective information system; A template matching unit, the template matching unit is used to traverse the template library, balance the subjective standard based on the text generation requirement and the information system fit, and match the target template; A text generation unit is used to transfer the effective information system and the target template to the reconstruction management block, perform information homology fitting and hierarchical layout, and obtain generated text.

Images