Two-chemical fusion analysis report generation method and device, electronic equipment and storage medium

Abstract

The invention provides a two-element fusion analysis report generation method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a report target text and to-be-analyzed data input by a user, carrying out the retrieval based on the report target text, obtaining a closest sample report, carrying out the layout analysis of the closest sample report, and obtaining a two-element fusion analysis report. Determining the content and form corresponding to each layout element, and determining the key point problem of the content corresponding to each layout element based on a large language model; determining a matching layout element of the report target text based on a matching result of each level of index in the report target text and the index in the key point problem, and replacing the matched index in the key point problem corresponding to the matching layout element based on each level of index in the report target text; and on the basis of a large language model, by utilizing the key point problem corresponding to the replaced matched layout element, the form corresponding to the matched layout element and the to-be-analyzed data, a dual-fusion analysis report is generated, and the personalized requirements of different application scenes on report contents and report forms are met.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to a method, device, electronic device and storage medium for generating an analysis report integrating information and industrialization. Background Art

[0002] The integration of information technology and industrialization is the abbreviation of the deep integration of information technology and industrialization, which aims to promote enterprises to improve production efficiency, optimize business processes and innovate management models through the application of information technology in the process of digital transformation. In this context, enterprises need to process complex information from multiple data sources such as production equipment, business systems, market data and customer feedback to support scientific decision-making and efficient operation.

[0003] In typical scenarios of the integration of industrialization and information technology, in order to effectively meet the needs of scientific evaluation in the current integrated development field, accurately grasp the key directions of future integrated development, and accurately grasp the overall trend of integrated development, the demand for generating development analysis reports corresponding to indicators at all levels under the enterprise's integration of industrialization and information technology development level evaluation system is particularly significant. Enterprises need to generate development analysis reports of different dimensions based on data corresponding to indicators at different levels from different data sources, such as production equipment data, product data, service data, etc., to evaluate the development trend of enterprises under the corresponding indicator system.

[0004] Due to the large demand for analysis reports, the efficiency of the previous manual writing of analysis reports can no longer meet the current needs, so an automatic report generation solution is needed to improve efficiency. However, different application scenarios have different emphasis on report content and different requirements for report format. The current report generation solution is difficult to flexibly respond to such diverse needs. Therefore, a report generation method that can be compatible with the personalized requirements of different application scenarios for report content and report format is needed to meet the needs of generating development analysis reports corresponding to indicators at all levels under the background of the integration of industrialization and informationization. Summary of the invention

[0005] The present invention provides a method, device, electronic device and storage medium for generating an analysis report of the integration of information and information technology, so as to solve the defect that the prior art is difficult to flexibly respond to the diversified needs of scenarios.

[0006] The present invention provides a method for generating an analysis report of the integration of information technology and industrialization, comprising: Determine the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; Retrieving the closest sample report based on the report target text, and performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element; Determine the key issues of the content corresponding to each layout element based on the large language model, and mark the indicators at all levels involved in the key issues based on the knowledge base; Based on the matching results of the indicators at all levels in the report target text and the indicators in the key questions of the contents corresponding to the various layout elements, the matching layout elements of the report target text are determined, and the matching indicators in the key questions corresponding to the matching layout elements are replaced based on the indicators at all levels in the report target text; Based on the large language model, an analysis report on the integration of information and industrialization is generated by using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements, and the data to be analyzed.

[0007] According to a method for generating an analysis report for the integration of information and industrialization provided by the present invention, the layout analysis of the closest sample report is performed to determine the content and form corresponding to each layout element, including: An image feature extraction module based on a text layout segmentation model extracts primary image features of an image corresponding to the closest sample report; Extracting the position coding features of the image corresponding to the closest sample report based on the position coding module of the text layout segmentation model; Based on the fusion coding module of the text layout segmentation model, the primary image features and the position coding features of the image corresponding to the closest sample report are fused to obtain a position fusion feature map of the image corresponding to the closest sample report; Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report; Based on the bounding boxes corresponding to the respective layout elements of the closest sample report, the content and form corresponding to the respective layout elements of the closest sample report are determined.

[0008] According to a method for generating an analysis report of the integration of information and information technology provided by the present invention, the target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report, including: Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box and its category corresponding to each minimum layout unit of the closest sample report; the minimum layout unit is a natural paragraph, a title, a picture or a table; Determine a candidate paragraph bounding box corresponding to any image bounding box or table bounding box; Determine a paragraph bounding box associated with any one of the picture bounding boxes or the table bounding boxes based on a result of a text match between any one of the picture bounding boxes or the table bounding boxes and the text in the candidate paragraph bounding boxes; Each picture bounding box or table bounding box is merged with the associated paragraph bounding box to obtain the bounding box corresponding to each layout element.

[0009] According to a method for generating an analysis report of the integration of information and information technology provided by the present invention, the target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each minimum layout unit of the closest sample report, including: Based on the deep perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain deep perception features; Based on the shallow perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain shallow perception features; Based on the object detection module of the text layout segmentation model, performing object detection on the deep perception feature and the shallow perception feature respectively to obtain a first bounding box detection result and a second bounding box detection result; The first bounding box detection result and the second bounding box detection result are matched to obtain a bounding box corresponding to each minimum layout unit.

[0010] According to a method for generating an analysis report for integration of information and information technology provided by the present invention, matching the first bounding box detection result with the second bounding box detection result to obtain a bounding box corresponding to each minimum layout unit includes: Constructing a bipartite graph corresponding to the first bounding box detection result and the second bounding box detection result; the left vertex in the bipartite graph is the bounding box in the first bounding box detection result, and the right vertex in the bipartite graph is the bounding box in the second bounding box detection result; Constructing a matching weight matrix; any element in the matching weight matrix represents the intersection-and-union ratio between the corresponding left vertex and the corresponding right vertex; Performing matching on the matching weight matrix based on the Hungarian algorithm to obtain a matching relationship between the first bounding box detection result and the second bounding box detection result; Based on the bounding boxes matched between the first bounding box detection result and the second bounding box detection result, a bounding box corresponding to each minimum layout unit is determined.

[0011] According to a method for generating an informatization-informatization fusion analysis report provided by the present invention, the deep perception layer and the shallow perception layer are constructed based on a multi-layer perception machine, and the number of hidden layers in the deep perception layer is greater than the number of hidden layers in the shallow perception layer.

[0012] According to a method for generating an industrialization-informatization fusion analysis report provided by the present invention, the image feature extraction module is constructed based on Swin Transformer, the position encoding module is constructed based on multiple convolutional layers, the fusion encoding module is constructed based on the encoder of the Transformer model, and the target detection module is constructed based on the Faster R-CNN model.

[0013] The present invention also provides a device for generating an analysis report of the integration of information technology and industrialization, comprising: A data determination unit is used to determine the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; A layout analysis unit, configured to retrieve the closest sample report based on the report target text, and perform layout analysis on the closest sample report to determine the content and form corresponding to each layout element; A content analysis unit, used to determine the key issues of the content corresponding to each layout element based on the large language model, and mark the indicators of various levels involved in the key issues based on the knowledge base; A key point matching unit, configured to determine a matching layout element of the report target text based on a matching result between the various levels of indicators in the report target text and the indicators in the key point questions of the contents corresponding to the various layout elements, and replace the matched indicators in the key point questions corresponding to the matching layout elements based on the various levels of indicators in the report target text; The report generating unit is used to generate an analysis report of the integration of information technology and industrialization based on the large language model, using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements and the data to be analyzed.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for generating an analysis report of the integration of information and information technology as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the method for generating an analysis report of the integration of information and information technology as described in any one of the above is implemented.

[0016] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for generating an analysis report on the integration of information and information technology.

[0017] The method, device, electronic device and storage medium for generating an analysis report for integration of industrialization and informationization provided by the present invention determine a report target text and data to be analyzed input by a user, retrieve the closest sample report based on the report target text, perform layout analysis on the closest sample report, determine the content and form corresponding to each layout element, determine the key issues of the content corresponding to each layout element based on a large language model, and mark the various levels of indicators involved in the key issues based on a knowledge base; determine the matching layout elements of the report target text based on the matching results of the various levels of indicators in the report target text and the indicators in the key issues of the content corresponding to each layout element, and replace the matching indicators in the key issues corresponding to the matching layout elements based on the various levels of indicators in the report target text; generate an analysis report for integration of industrialization and informationization based on a large language model using the key issues corresponding to the replaced matching layout elements, the form corresponding to the matching layout elements and the data to be analyzed, thereby improving the generation efficiency of the analysis report under the background of integration of industrialization and informationization and being compatible with the personalized requirements of different application scenarios for the report content and report form. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a flow chart of the method for generating an analysis report of the integration of information and industrialization provided by the present invention; Figure 2 It is a flow chart of the layout analysis method provided by the present invention; Figure 3 It is a structural schematic diagram of the device for generating an analysis report of the integration of information technology and industrialization provided by the present invention; Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Figure 1 It is a flow chart of the method for generating an analysis report of the integration of information and information technology provided by the present invention, such as Figure 1 As shown, the method includes: Step 110, determining the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; Step 120, retrieving the closest sample report based on the report target text, and performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element; Step 130, determining the key issues of the content corresponding to each layout element based on the large language model, and marking the indicators of various levels involved in the key issues based on the knowledge base; Step 140, determining the matching layout element of the report target text based on the matching results of the indicators at all levels in the report target text and the indicators in the key questions of the contents corresponding to the respective layout elements, and replacing the matching indicators in the key questions corresponding to the matching layout elements based on the indicators at all levels in the report target text; Step 150: Based on the large language model, an analysis report on the integration of information technology and industrialization is generated by using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements, and the data to be analyzed.

[0022] Here, the user can input the report target text and the corresponding data to be analyzed for the current report to be generated through the human-computer interaction interface. Among them, the report target text contains the analysis targets corresponding to the indicators at all levels under the predetermined two-industry integration indicator system. In some embodiments, the two-industry integration indicator system can include two-level indicators, wherein the first-level indicators include new technologies, management, production, products, and services; the second-level indicators are the refinement of the first-level indicators. For example, the second-level indicators of the first-level indicator new technologies include information technology, industrial technology, and communication technology. The second-level indicators of the first-level indicator management include organizational management, comprehensive management, decision-making management, and supply chain management. The second-level indicators of the first-level indicator production include research and development design, production and manufacturing, production factors, and production equipment. The second-level indicators of the first-level indicator products include product functions, product performance, and product quality management. The second-level indicators of the first-level indicator services include service methods, service scope, and service value. In other embodiments, the two-industry integration indicator system can also include three or more levels of indicators, wherein the low-level indicators are the refinement of the high-level indicators. The specific design of the indicator system can be set according to actual needs, and the embodiments of the present invention do not specifically limit this.

[0023] For example, the report objective text entered by the user may be: "Conduct trend analysis and summary suggestions on enterprise production indicators, including: Manufacturing: Analyze the trends of manufacturing efficiency and yield rate; Production factors: Analyze the trends of human resource input-output ratio and raw material utilization rate; Production equipment: Analyze trends in equipment utilization and downtime rates. In addition, the user needs to input the data to be analyzed, which includes the structured data corresponding to the lowest level indicator. Taking the two-level indicator system as an example, the data to be analyzed is the structured indicator data of the second-level indicator.

[0024] Based on the above-mentioned report target text, a pre-built summary library is searched. Among them, the summary library contains the full text summary of each sample report. In some embodiments, keywords such as the indicator name and analysis target contained in the report target text can be extracted as search terms for retrieval, so as to retrieve the closest sample report that matches the report target text. Among them, the indicators and their analysis targets contained in the closest sample report match the report target text. In order to generate an analysis report that better fits the current application scenario, the closest sample report can be subjected to layout analysis to determine the content and form corresponding to each layout element, so as to generate an analysis report whose content focus and form are consistent with the closest sample report. Among them, the content corresponding to any layout element includes one or more of text, pictures and tables, and the form corresponding to any layout element includes plain text, a combination of text and pictures, a combination of text and tables, and a combination of text, pictures and tables.

[0025] In some embodiments, Figure 2 As shown, the layout analysis of the closest sample report can be performed in the following manner: Step 210, extracting primary image features of the image corresponding to the closest sample report based on the image feature extraction module of the text layout segmentation model; Step 220, extracting the position coding features of the image corresponding to the closest sample report based on the position coding module of the text layout segmentation model; Step 230, based on the fusion coding module of the text layout segmentation model, the primary image features and the position coding features of the image corresponding to the closest sample report are fused to obtain a position fusion feature map of the image corresponding to the closest sample report; Step 240, based on the target detection module of the text layout segmentation model, perform target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report; Step 250: Determine the content and form corresponding to each layout element of the closest sample report based on the bounding box corresponding to each layout element of the closest sample report.

[0026] Here, considering that the image form is helpful to extract complex layout information, and the image retains clear structural information such as hierarchy, title, paragraph, etc. in the structure, the processing of complex typesetting and nested structure is more intuitive. In addition, it does not depend on the encoding or language structure of the document, and can cross different language and platform restrictions. Therefore, in the embodiment of the present invention, the image corresponding to the closest sample report is obtained as the analysis object. Among them, if the sample report has multiple pages, the multiple pages are integrated into one image. Specifically, a text layout segmentation model can be constructed to perform layout analysis on the sample report. Among them, the text layout segmentation model includes an image feature extraction module, a position encoding module, a fusion encoding module and a target detection module.

[0027] The image feature extraction module is used to extract the primary image features of the image corresponding to the closest sample report. In some embodiments, the image feature extraction module can be a skeleton structure of Swin Transformer.

[0028] The position encoding module is used to extract the position encoding features of the image corresponding to the closest sample report. In some embodiments, the position encoding module can be a plurality of (eg, three) convolutional layers.

[0029] The fusion coding module is used to fuse the primary image features and position coding features of the image corresponding to the closest sample report to obtain a position fusion feature map of the image corresponding to the closest sample report. Since the position fusion feature map contains both the image semantics and position information in the image corresponding to the closest sample report, it helps to more accurately extract the layout information in the closest sample report. In some embodiments, the fusion coding module can be constructed based on the encoder of the Transformer model.

[0030] The target detection module is used to perform target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report. In some embodiments, a target detection model such as a Faster R-CNN model can be used to construct the target detection module. Based on the bounding box corresponding to each layout element of the closest sample report, the content and form corresponding to each layout element of the closest sample report can be obtained.

[0031] In some embodiments, considering the different image expressions of text and charts, it is relatively difficult for the model to detect associated paragraphs and charts. In order to reduce the complexity of the model, the target detection module of the text layout segmentation model can be used to perform target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box and category corresponding to each minimum layout unit of the closest sample report. Among them, the minimum layout unit is a natural paragraph, a title, a picture or a table, and its category is paragraph, title, picture or table.

[0032] Here, for paragraphs that span across pages or columns, since there is a certain distance in spatial position, deep perception layers and shallow perception layers can be added between the fusion coding module and the target detection module. The deep perception layer and the shallow perception layer are constructed based on multi-layer perceptrons, and the number of hidden layers in the deep perception layer (for example, 5 layers) is greater than the number of hidden layers in the shallow perception layer (for example, 3 layers). It can be seen that in this way, the global feature extraction capabilities of different multi-layer perceptrons can be comprehensively utilized (the deep perception layer is better at capturing the global contextual relationship of the input, and the shallow perception layer is more inclined to capture shallow features, such as edge contours), providing richer input features for the subsequent target detection module, and enhancing its performance in complex image tasks (such as association modeling of paragraphs that span across pages / columns).

[0033] Specifically, based on the deep perception layer of the text layout segmentation model, feature extraction can be performed on the position fusion feature map of the image corresponding to the closest sample report to obtain deep perception features, and based on the shallow perception layer of the text layout segmentation model, feature extraction can be performed on the position fusion feature map of the image corresponding to the closest sample report to obtain shallow perception features. Subsequently, the target detection module will perform target detection based on the deep perception features and the shallow perception features respectively to obtain a first bounding box detection result and a second bounding box detection result. By matching the above-mentioned first bounding box detection result and the second bounding box detection result, the bounding box corresponding to each minimum layout unit can be determined based on the matching result. In some embodiments, before matching the above-mentioned first bounding box detection result and the second bounding box detection result, a bipartite graph corresponding to the first bounding box detection result and the second bounding box detection result can be constructed. Among them, the left vertex in the bipartite graph corresponds to the bounding box in the first bounding box detection result, and the right vertex in the bipartite graph corresponds to the bounding box in the second bounding box detection result. Subsequently, a matching weight matrix can be constructed, and any element in the matching weight matrix represents the intersection-and-union ratio between the corresponding left vertex and the corresponding right vertex. Based on the Hungarian algorithm, matching is performed on the above matching weight matrix to obtain a matching relationship between the first bounding box detection result and the second bounding box detection result, and based on the bounding boxes matched between the first bounding box detection result and the second bounding box detection result, the bounding boxes corresponding to each minimum layout unit are determined. The bounding boxes in the first bounding box detection result and the bounding boxes matched in the second bounding box detection result can be fused to obtain the bounding boxes corresponding to each minimum layout unit.

[0034] After obtaining the bounding boxes and their categories corresponding to each minimum layout unit, the picture bounding box (i.e., the bounding box with the category of picture) and the table bounding box (i.e., the bounding box with the category of table) can be screened out, and the candidate paragraph bounding boxes corresponding to each picture bounding box and table bounding box can be determined. In some embodiments, based on a window of preset size, the center of any picture bounding box or table bounding box can be taken as the center of the window, and the paragraph bounding box (i.e., the bounding box with the category of paragraph) whose area in the window or overlapping area with the window is greater than the preset area can be obtained as the candidate paragraph bounding box corresponding to the picture bounding box and the table bounding box. Based on the text matching results in any picture bounding box or table bounding box and its corresponding candidate paragraph bounding box, the paragraph bounding box associated with the picture bounding box or table bounding box is determined. Subsequently, each picture bounding box or table bounding box and its associated paragraph bounding box are merged to obtain the bounding box corresponding to each layout element.

[0035] It should be noted that a training set can be constructed based on pre-collected sample documents, and the bounding box of the smallest layout unit in each sample document is manually annotated as an annotated bounding box to train the above-mentioned text layout segmentation model. In some embodiments, the model loss can be calculated based on the intersection-over-union ratio between the bounding box output by the model and the annotated bounding box, and the Adam optimizer is used for iterative training for 100 epochs, where the learning rate is set to 0.001, and the first-order momentum decay coefficient and the second-order momentum decay coefficient are set to 0.9 and 0.999 respectively, so as to achieve model training.

[0036] After obtaining the content and form corresponding to each layout element in the closest sample report, the key issues of the content corresponding to each layout element can be obtained based on the pre-trained large language model. Here, the large language model (LLM) can adopt any currently public large language model and use the existing large language model's natural language processing capabilities to automatically extract the key issues of the content corresponding to each layout element, where the key issues of the content corresponding to any layout element extract the key elements of the content corresponding to the layout element and describe them in the form of questions. Subsequently, considering that the description of the same indicator in different sample reports may be slightly different, the indicators at all levels involved in the key issues can be marked based on the pre-built knowledge base in the context of the integration of industrialization and informationization, that is, the keywords corresponding to each indicator in the key issues are marked as the unified names of the corresponding indicators.

[0037] Considering that there may be descriptions of other indicators that are not related to the report target text in the sample report, the indicators at all levels in the report target text can be matched with the indicators in the key questions of the content corresponding to each layout element based on the knowledge base, and the layout elements associated with the report target text, that is, the matching layout elements, are screened out. Subsequently, the description of the indicators at all levels in the report target text is used to replace the description of the indicators matched in the key questions corresponding to the matching layout elements to ensure the consistency of the description of the generated report and the report target text. In some embodiments, if the key questions corresponding to the matching layout elements also contain other indicators that are not related to the report target text, the other irrelevant indicators are deleted, and the description of the key questions is adjusted manually or through a large language model to ensure that its sentences are coherent and smooth.

[0038] Subsequently, based on the large language model, the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements, and the data to be analyzed can be used to generate an analysis report on the integration of industrialization and informationization. In some embodiments, the key issues corresponding to the replaced matching layout elements can be organized in the order of the corresponding matching layout elements in the sample report, and then input into the large language model together with the data to be analyzed, so as to generate new text content corresponding to each matching layout element, forming the text part of the analysis report on the integration of industrialization and informationization. Subsequently, based on the forms corresponding to each matching layout element, when it is necessary to insert pictures and / or tables, the Python tool library is used to generate graphics and / or tables of the corresponding data, and combined with the new text content corresponding to the corresponding matching layout elements. Then, the typesetting style of text, paragraphs, tables, pictures, etc. is set based on the typesetting of the sample report, so as to form an analysis report on the integration of industrialization and informationization.

[0039] In summary, the method provided by the embodiment of the present invention determines the report target text and the data to be analyzed input by the user, retrieves the closest sample report based on the report target text, performs layout analysis on the closest sample report, determines the content and form corresponding to each layout element, determines the key issues of the content corresponding to each layout element based on the large language model, and marks the various levels of indicators involved in the key issues based on the knowledge base; determines the matching layout elements of the report target text based on the matching results of the various levels of indicators in the report target text and the indicators in the key issues of the content corresponding to each layout element, and replaces the matching indicators in the key issues corresponding to the matching layout elements based on the various levels of indicators in the report target text; based on the large language model, generates an analysis report on the integration of industrialization and information technology using the key issues corresponding to the replaced matching layout elements, the form corresponding to the matching layout elements, and the data to be analyzed, thereby improving the generation efficiency of the analysis report under the background of the integration of industrialization and information technology and being compatible with the personalized needs of different application scenarios for report content and report form.

[0040] The following is a description of the device for generating an analysis report for the integration of information and industrialization provided by the present invention. The device for generating an analysis report for the integration of information and industrialization described below and the method for generating an analysis report for the integration of information and industrialization described above can refer to each other.

[0041] Based on any of the above embodiments, Figure 3 It is a structural schematic diagram of the device for generating an analysis report of the integration of industrialization and informationization provided by the present invention. Figure 3 As shown, the device comprises: The data determination unit 310 is used to determine the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; A layout analysis unit 320 is used to retrieve the closest sample report based on the report target text, and perform layout analysis on the closest sample report to determine the content and form corresponding to each layout element; The content analysis unit 330 is used to determine the key issues of the content corresponding to each layout element based on the large language model, and mark the indicators of various levels involved in the key issues based on the knowledge base; A key point matching unit 340 is used to determine the matching layout element of the report target text based on the matching results of the indicators at all levels in the report target text and the indicators in the key point questions of the contents corresponding to the respective layout elements, and replace the matched indicators in the key point questions corresponding to the matching layout elements based on the indicators at all levels in the report target text; The report generating unit 350 is used to generate an analysis report of the integration of information technology and industrialization based on the large language model by using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements and the data to be analyzed.

[0042] The device provided by the embodiment of the present invention determines the report target text and the data to be analyzed input by the user, retrieves the closest sample report based on the report target text, performs layout analysis on the closest sample report, determines the content and form corresponding to each layout element, determines the key issues of the content corresponding to each layout element based on the large language model, and marks the various levels of indicators involved in the key issues based on the knowledge base; determines the matching layout elements of the report target text based on the matching results of the various levels of indicators in the report target text and the indicators in the key issues of the content corresponding to each layout element, and replaces the matching indicators in the key issues corresponding to the matching layout elements based on the various levels of indicators in the report target text; generates an analysis report on the integration of industrialization and information technology based on the large language model, using the key issues corresponding to the replaced matching layout elements, the form corresponding to the matching layout elements, and the data to be analyzed, thereby improving the generation efficiency of the analysis report under the background of the integration of industrialization and information technology and being compatible with the personalized needs of different application scenarios for report content and report form.

[0043] Based on any of the above embodiments, performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element includes: An image feature extraction module based on a text layout segmentation model extracts primary image features of an image corresponding to the closest sample report; Extracting the position coding features of the image corresponding to the closest sample report based on the position coding module of the text layout segmentation model; Based on the fusion coding module of the text layout segmentation model, the primary image features and the position coding features of the image corresponding to the closest sample report are fused to obtain a position fusion feature map of the image corresponding to the closest sample report; Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report; Based on the bounding boxes corresponding to the respective layout elements of the closest sample report, the content and form corresponding to the respective layout elements of the closest sample report are determined.

[0044] Based on any of the above embodiments, the target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report, including: Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box and its category corresponding to each minimum layout unit of the closest sample report; the minimum layout unit is a natural paragraph, a title, a picture or a table; Determine a candidate paragraph bounding box corresponding to any image bounding box or table bounding box; Determine a paragraph bounding box associated with any one of the picture bounding boxes or the table bounding boxes based on a result of a text match between any one of the picture bounding boxes or the table bounding boxes and the text in the candidate paragraph bounding boxes; Each picture bounding box or table bounding box is merged with the associated paragraph bounding box to obtain the bounding box corresponding to each layout element.

[0045] Based on any of the above embodiments, the target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each minimum layout unit of the closest sample report, including: Based on the deep perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain deep perception features; Based on the shallow perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain shallow perception features; Based on the object detection module of the text layout segmentation model, performing object detection on the deep perception feature and the shallow perception feature respectively to obtain a first bounding box detection result and a second bounding box detection result; The first bounding box detection result and the second bounding box detection result are matched to obtain a bounding box corresponding to each minimum layout unit.

[0046] Based on any of the foregoing embodiments, matching the first bounding box detection result and the second bounding box detection result to obtain a bounding box corresponding to each minimum layout unit includes: Constructing a bipartite graph corresponding to the first bounding box detection result and the second bounding box detection result; the left vertex in the bipartite graph is the bounding box in the first bounding box detection result, and the right vertex in the bipartite graph is the bounding box in the second bounding box detection result; Constructing a matching weight matrix; any element in the matching weight matrix represents the intersection-and-union ratio between the corresponding left vertex and the corresponding right vertex; Performing matching on the matching weight matrix based on the Hungarian algorithm to obtain a matching relationship between the first bounding box detection result and the second bounding box detection result; Based on the bounding boxes matched between the first bounding box detection result and the second bounding box detection result, a bounding box corresponding to each minimum layout unit is determined.

[0047] Based on any of the above embodiments, the deep perception layer and the shallow perception layer are constructed based on a multi-layer perceptron, and the number of hidden layers in the deep perception layer is greater than the number of hidden layers in the shallow perception layer.

[0048] Based on any of the above embodiments, the image feature extraction module is constructed based on Swin Transformer, the position encoding module is constructed based on multiple convolutional layers, the fusion encoding module is constructed based on the encoder of the Transformer model, and the target detection module is constructed based on the Faster R-CNN model.

[0049] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor (processor) 410, a memory (memory) 420, a communication interface (Communications Interface) 430 and a communication bus 440, wherein the processor 410, the memory 420, and the communication interface 430 communicate with each other through the communication bus 440. The processor 410 can call the logic instructions in the memory 420 to execute the method for generating an analysis report for the integration of industrialization and information technology, which includes: determining a report target text and data to be analyzed input by a user; wherein the report target text contains analysis targets corresponding to indicators at all levels under the indicator system for the integration of industrialization and information technology, and the data to be analyzed includes structured data corresponding to the lowest level indicator; retrieving the closest sample report based on the report target text, and performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element; determining key issues of the content corresponding to each layout element based on a large language model, and marking the indicators at all levels involved in the key issues based on a knowledge base; determining matching layout elements of the report target text based on matching results between the indicators at all levels in the report target text and the indicators in the key issues of the content corresponding to each layout element, and replacing the matching indicators in the key issues corresponding to the matching layout elements based on the indicators at all levels in the report target text; and generating an analysis report for the integration of industrialization and information technology based on the large language model using the key issues corresponding to the replaced matching layout elements, the form corresponding to the matching layout elements, and the data to be analyzed.

[0050] In addition, the logic instructions in the above-mentioned memory 420 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0051] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the method for generating an analysis report for the integration of industrialization and information technology provided by the above methods, and the method includes: determining a report target text and data to be analyzed input by a user; wherein the report target text contains analysis targets corresponding to indicators at all levels under the integration of industrialization and information technology indicator system, and the data to be analyzed includes structured data corresponding to the lowest level indicators; based on the report target text, the closest sample report is retrieved, and the closest sample report is layout analyzed. Analysis is performed to determine the content and form corresponding to each layout element; based on the large language model, key issues of the content corresponding to each layout element are determined, and indicators at all levels involved in the key issues are marked based on the knowledge base; based on the matching results of the indicators at all levels in the report target text and the indicators in the key issues of the content corresponding to each layout element, the matching layout elements of the report target text are determined, and the matching indicators in the key issues corresponding to the matching layout elements are replaced based on the indicators at all levels in the report target text; based on the large language model, the key issues corresponding to the replaced matching layout elements, the form corresponding to the matching layout elements and the data to be analyzed are used to generate an analysis report on the integration of industrialization and informationization.

[0052] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the above-mentioned two-in-one integration analysis report generation method, the method comprising: determining a report target text and data to be analyzed input by a user; wherein the report target text contains analysis targets corresponding to indicators at all levels under the two-in-one integration indicator system, and the data to be analyzed includes structured data corresponding to the lowest level indicator; based on the report target text, the closest sample report is retrieved, and the layout analysis of the closest sample report is performed to determine the content and form corresponding to each layout element; Determine the key issues of the content corresponding to each layout element based on the large language model, and mark the various levels of indicators involved in the key issues based on the knowledge base; determine the matching layout elements of the report target text based on the matching results of the various levels of indicators in the report target text and the indicators in the key issues of the content corresponding to each layout element, and replace the matching indicators in the key issues corresponding to the matching layout elements based on the various levels of indicators in the report target text; based on the large language model, generate an analysis report on the integration of industrialization and informationization using the replaced key issues corresponding to the matching layout elements, the forms corresponding to the matching layout elements, and the data to be analyzed.

[0053] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0054] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0055] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating an analysis report of the integration of information technology and industrialization, characterized in that: include: Determine the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; Retrieving the closest sample report based on the report target text, and performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element; Determine the key issues of the content corresponding to each layout element based on the large language model, and mark the indicators at all levels involved in the key issues based on the knowledge base; Based on the matching results of the indicators at all levels in the report target text and the indicators in the key questions of the contents corresponding to the various layout elements, the matching layout elements of the report target text are determined, and the matching indicators in the key questions corresponding to the matching layout elements are replaced based on the indicators at all levels in the report target text; Based on the large language model, an analysis report on the integration of information and industrialization is generated by using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements, and the data to be analyzed.

2. The method for generating an analysis report of the integration of information technology and industrialization according to claim 1 is characterized in that: The performing layout analysis on the closest sample report to determine the content and form corresponding to each layout element includes: An image feature extraction module based on a text layout segmentation model extracts primary image features of an image corresponding to the closest sample report; Extracting the position coding features of the image corresponding to the closest sample report based on the position coding module of the text layout segmentation model; Based on the fusion coding module of the text layout segmentation model, the primary image features and the position coding features of the image corresponding to the closest sample report are fused to obtain a position fusion feature map of the image corresponding to the closest sample report; Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report; Based on the bounding boxes corresponding to the respective layout elements of the closest sample report, the content and form corresponding to the respective layout elements of the closest sample report are determined.

3. The method for generating an analysis report of the integration of information and industrialization according to claim 2 is characterized in that: The target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each layout element of the closest sample report, including: Based on the target detection module of the text layout segmentation model, target detection is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box and its category corresponding to each minimum layout unit of the closest sample report; the minimum layout unit is a natural paragraph, a title, a picture or a table; Determine a candidate paragraph bounding box corresponding to any image bounding box or table bounding box; Determine a paragraph bounding box associated with any one of the picture bounding boxes or the table bounding boxes based on a result of a text match between any one of the picture bounding boxes or the table bounding boxes and the text in the candidate paragraph bounding boxes; Each picture bounding box or table bounding box is merged with the associated paragraph bounding box to obtain the bounding box corresponding to each layout element.

4. The method for generating an analysis report of the integration of information and industrialization according to claim 3 is characterized in that: The target detection module based on the text layout segmentation model performs target detection on the position fusion feature map of the image corresponding to the closest sample report to obtain the bounding box corresponding to each minimum layout unit of the closest sample report, including: Based on the deep perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain deep perception features; Based on the shallow perception layer of the text layout segmentation model, feature extraction is performed on the position fusion feature map of the image corresponding to the closest sample report to obtain shallow perception features; Based on the object detection module of the text layout segmentation model, performing object detection on the deep perception feature and the shallow perception feature respectively to obtain a first bounding box detection result and a second bounding box detection result; The first bounding box detection result and the second bounding box detection result are matched to obtain a bounding box corresponding to each minimum layout unit.

5. The method for generating an analysis report of the integration of information and industrialization according to claim 4 is characterized in that: The matching the first bounding box detection result and the second bounding box detection result to obtain a bounding box corresponding to each minimum layout unit includes: Constructing a bipartite graph corresponding to the first bounding box detection result and the second bounding box detection result; the left vertex in the bipartite graph is the bounding box in the first bounding box detection result, and the right vertex in the bipartite graph is the bounding box in the second bounding box detection result; Constructing a matching weight matrix; any element in the matching weight matrix represents the intersection-and-union ratio between the corresponding left vertex and the corresponding right vertex; Performing matching on the matching weight matrix based on the Hungarian algorithm to obtain a matching relationship between the first bounding box detection result and the second bounding box detection result; Based on the bounding boxes matched between the first bounding box detection result and the second bounding box detection result, a bounding box corresponding to each minimum layout unit is determined.

6. The method for generating an analysis report of the integration of information and industrialization according to claim 4 is characterized in that: The deep perception layer and the shallow perception layer are constructed based on a multi-layer perception machine, and the number of hidden layers in the deep perception layer is greater than the number of hidden layers in the shallow perception layer.

7. The method for generating an analysis report of the integration of information and industrialization according to any one of claims 2 to 6, characterized in that: The image feature extraction module is constructed based on Swin Transformer, the position encoding module is constructed based on multiple convolutional layers, the fusion encoding module is constructed based on the encoder of the Transformer model, and the target detection module is constructed based on the Faster R-CNN model.

8. A device for generating an analysis report of the integration of information technology and industrialization, characterized in that: include: A data determination unit is used to determine the report target text and the data to be analyzed input by the user; wherein the report target text contains the analysis targets corresponding to the indicators at all levels under the indicator system of integration of industrialization and informationization, and the data to be analyzed includes the structured data corresponding to the lowest level indicators; A layout analysis unit, configured to retrieve the closest sample report based on the report target text, and perform layout analysis on the closest sample report to determine the content and form corresponding to each layout element; A content analysis unit, used to determine the key issues of the content corresponding to each layout element based on the large language model, and mark the indicators of various levels involved in the key issues based on the knowledge base; A key point matching unit, configured to determine a matching layout element of the report target text based on a matching result between the various levels of indicators in the report target text and the indicators in the key point questions of the contents corresponding to the various layout elements, and replace the matched indicators in the key point questions corresponding to the matching layout elements based on the various levels of indicators in the report target text; The report generating unit is used to generate an analysis report of the integration of information technology and industrialization based on the large language model, using the key issues corresponding to the replaced matching layout elements, the forms corresponding to the matching layout elements and the data to be analyzed.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for generating an informatization and informationization integration analysis report as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating an informatization and informationization integration analysis report as described in any one of claims 1 to 7 is implemented.

Images