Report generation method and device, equipment, medium and product

Abstract

The invention discloses a report generation method and device, equipment, a medium and a product, and relates to the technical field of artificial intelligence, and the method comprises the steps: responding to a received user input instruction, and assisting a user interface module to generate a target keyword and a query short sentence according to the user input instruction through a first agent role; performing document screening and sorting on the basis of the target keyword and the query short sentence to obtain a finely sorted document; performing semantic filtering on the finely arranged document through a semantic filtering module, and calling a second agent role to generate a final document; and calling a third agent role through the semantic processing module to perform semantic processing on the final document, and / or processing the final document through the tool calling module to generate a target report, thereby avoiding the dependence of the report generation process on manpower, realizing automatic report generation, effectively reducing the labor cost, and improving the report generation efficiency. The time for generating the report is saved, so that the efficiency of generating the report is improved, and the accuracy of generating the report can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a report generation method, device, equipment, medium and product. Background Art

[0002] In existing banking and financial business scenarios, business personnel often need comprehensive information such as recent data, recent development overviews, and public opinion about certain industries and companies when making relevant decisions. Therefore, it is necessary to collect relevant news, research reports, or public opinion information from all walks of life. Much of this information is obtained by searching recent online articles or reports stored on the Internet or company databases. After obtaining the search results, data and information are extracted from them, and then calculated, summarized, etc., to write specific reports, and then make judgments on the development of related industries based on the report information, so as to carry out loan financial business.

[0003] In the current report generation process, both information collection and information extraction require a lot of manual participation, sometimes supplemented by some traditional keywords, rules and traditional NLP (Natural Language Processing) models. However, these traditional methods are relatively weak and lack the ability to understand the semantics of articles, resulting in inaccurate report information.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a report generation method, device, equipment, medium and product, aiming to realize automatic report generation and improve the efficiency and accuracy of report generation.

[0006] To achieve the above-mentioned purpose, the present application provides a report generation method, which is applied to an industry information retrieval report system, wherein the industry information retrieval report system includes at least one of a data source module, a data rough screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summarization module, a tool interface library module, a user interface module, and a role management module, wherein the role management module is used to adjust the target macro model to at least one of a first agent role, a second agent role, and a third agent role according to the content of each prompt word, and the method includes:

[0007] In response to receiving a user input instruction, the user interface module is assisted by the first agent role to generate a target keyword and a query phrase according to the user input instruction;

[0008] Screening and sorting documents based on the target keywords and query phrases to obtain refined documents;

[0009] The semantic filtering module is used to perform semantic filtering on the refined document, and the second intelligent agent role is called to generate a final document; the third intelligent agent role is called to perform semantic processing on the final document through the semantic processing module, and / or the final document is processed through the tool calling module to generate a target report.

[0010] In one embodiment, in response to receiving a user input instruction, the step of assisting the user interface module in generating target keywords and query phrases according to the user input instruction by the first agent character comprises:

[0011] Receiving the user input instruction through the user interface module;

[0012] The first intelligent agent role is called, and the user input instruction is analyzed by the first intelligent agent role to generate the target keyword and query phrase.

[0013] In one embodiment, the step of calling the first agent role further includes:

[0014] The first agent role performs intent analysis on the user input instruction to generate a tool calling requirement, wherein the tool calling requirement is used to call a tool model to the tool interface library module.

[0015] In one embodiment, the step of screening and sorting documents based on the target keywords and query phrases to obtain finely sorted documents includes:

[0016] Inputting the target keyword and the first preset parameter into the data coarse screening module, and obtaining a coarse screening document through the data coarse screening module;

[0017] The target keyword, query phrase, coarsely screened document and second preset parameter are input into the data fine sorting module, and the fine sorting document is obtained through the data fine sorting module.

[0018] In one embodiment, the first preset parameter includes the number of first documents and / or the number of second documents, and the step of obtaining the coarsely screened documents by the data coarse screening module includes:

[0019] The data source module is called by the data coarse screening module, and the data source module performs weighted calculation and sorting on the documents in the local database according to the target keyword to obtain a first sorting result, and selects a first number of document results from the first sorting result; and / or, the data source module uses the target keyword to search for networked data and returns a second number of document results;

[0020] The coarse-screened documents are obtained by deduplicating the document results of the first number of documents and / or the document results of the second number of documents through local sensitive hashing.

[0021] In one embodiment, the second preset parameters include the third document quantity, the segmentation window size parameter, and the sliding distance, and the step of obtaining the refined document by the data refined sorting module includes:

[0022] The rough screening document is symbolized by the data fine sorting module to obtain a digital list; and the rough screening document is divided into individual text paragraphs according to the segmentation window size parameter and the sliding distance, and the similarity score between each text paragraph and the target keyword is calculated;

[0023] Sorting the text paragraphs according to the similarity scores to obtain a second sorting result, and updating the number list according to the second sorting result to obtain a target number list;

[0024] The document results of the third number of documents in the target digital list are used as the refined sorted documents.

[0025] In one embodiment, the step of performing semantic filtering on the refined document by the semantic filtering module and calling the second agent role to generate the final document includes:

[0026] Acquire the refined sorted documents, target keywords, query phrases and first preset instructions through the semantic filtering module;

[0027] Calling the second intelligent agent role, and judging whether each document in the refined arrangement document meets the target keyword and the query phrase through the second intelligent agent role according to the first preset instruction;

[0028] The document that meets the target keyword and the query phrase in the refined sorted documents is used as the final document.

[0029] In one embodiment, the step of calling the third agent role to perform semantic processing on the final document through the semantic processing module, and / or processing the final document through the tool calling module to generate a target report includes:

[0030] The final document and the second preset instruction are obtained through the semantic processing and aggregation module, the third intelligent agent role is called, and the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the final result corresponding to each document in the final document, or the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the intermediate result corresponding to each document in the final document, and the intermediate results corresponding to each document in the final document are aggregated to obtain the aggregated result, and the target report is generated according to the final result corresponding to each document in the final document or the aggregated result; and / or,

[0031] The target report is obtained by performing specific task processing on the final document through the tool calling module, wherein the specific task processing includes sentiment analysis and / or content extraction.

[0032] In addition, to achieve the above purpose, the present application also proposes a report generation device, which is applied to an industry information retrieval report system, and the industry information retrieval report system includes at least one of a data source module, a data coarse screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summarization module, a tool interface library module, a user interface module, and a role management module. The role management module is used to adjust the target macro model to at least one of a first agent role, a second agent role, and a third agent role according to the content of each prompt word. The report generation device includes:

[0033] A response module, configured to, in response to receiving a user input instruction, assist the user interface module in generating a target keyword and a query phrase according to the user input instruction through the first agent role;

[0034] A screening module is used to screen and sort documents based on the target keywords and query phrases to obtain refined documents;

[0035] A generation module is used to perform semantic filtering on the refined document through the semantic filtering module, and call the second intelligent agent role to generate a final document; call the third intelligent agent role to perform semantic processing on the final document through the semantic processing module, and / or process the final document through the tool calling module to generate a target report.

[0036] In addition, to achieve the above objectives, the present application also proposes a report generation device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the report generation method described above.

[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the report generation method described above are implemented.

[0038] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the report generation method described above are implemented.

[0039] One or more technical solutions proposed in this application have at least the following technical effects:

[0040] In response to receiving user input instructions, the first intelligent agent role assists the user interface module to generate target keywords and query phrases according to the user input instructions, and can directly connect with the user, analyze the target keywords and query phrases according to the user input instructions, and screen out documents related to the target keywords from different data sources based on the target keywords, and then combine the query phrases to further screen the related documents screened out from different data sources; by screening and sorting the documents based on the target keywords and query phrases, refined documents are obtained, so that documents with higher relevance to the user input instructions can be further selected from the initially screened documents; the refined documents are sorted by the semantic filtering module Semantic filtering, and calling the second intelligent agent role to generate the final document; calling the third intelligent agent role through the semantic processing module to perform semantic processing on the final document, and / or, processing the final document through the tool calling module to generate a target report, which can further determine the documents that better meet the user's needs from the screened documents with high relevance to the user input instructions, and obtain a target report with higher accuracy after processing, thereby avoiding the dependence of the report generation process on manual labor, and with the help of the various intelligent agent roles provided by the role management module, automatic report generation is realized, which effectively reduces labor costs and saves the time required for report generation, thereby improving the efficiency of report generation and at the same time improving the accuracy of report generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0043] Figure 1 A flowchart of the first embodiment of the method for generating a report for this application is provided;

[0044] Figure 2 A flowchart of the second embodiment of the method for generating a report for this application is provided;

[0045] Figure 3 is a schematic diagram showing a principle according to a second embodiment;

[0046] Figure 4 This is a flow chart of the third embodiment of the method for generating a report of this application;

[0047] Figure 5 A brief flowchart of the method for generating the report for this application;

[0048] Figure 6 A schematic diagram of the module structure of the report generating device provided in this application;

[0049] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the report generation method in the embodiment of the present application.

[0050] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0051] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0052] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0053] The main solution of the embodiment of the present application is: by responding to receiving user input instructions, the first intelligent agent role assists the user interface module to generate target keywords and query phrases according to the user input instructions, and can directly connect with the user, analyze the target keywords and query phrases according to the user's input instructions, and according to the target keywords, documents related to the target keywords can be screened out from different data sources, and then combined with the query phrases, the relevant documents screened out from different data sources can be further screened; by screening and sorting the documents based on the target keywords and query phrases, a refined document is obtained, so that documents with a higher relevance to the user input instructions can be further selected from the initially screened documents; through the semantic filtering module The block performs semantic filtering on the sorted document, and calls the second intelligent agent role to generate the final document; calls the third intelligent agent role through the semantic processing module to perform semantic processing on the final document, and / or processes the final document through the tool calling module to generate a target report. It can further determine the documents that better meet the user's needs from the screened documents with high relevance to the user input instructions, and obtain a target report with higher accuracy after processing, thereby avoiding the dependence of the report generation process on manual labor, and with the help of the various intelligent agent roles provided by the role management module, realizes automatic report generation, effectively reduces labor costs, saves the time required for report generation, thereby improving the efficiency of report generation, and at the same time can improve the accuracy of report generation.

[0054] In this embodiment, for the convenience of description, the report generating device is used as the execution subject for explanation below.

[0055] In existing banking and financial business scenarios, business personnel (such as corporate loans) often need comprehensive information about recent data of certain industries and companies, recent development overviews (whether it is favorable or whether it is a downward trend), public opinion, etc. when making relevant decisions. Therefore, it is necessary to collect relevant news or research reports or public opinion information from all walks of life. Much of this information comes from recent online articles (some institutions will purchase non-Internet data sources to obtain data from other sources) and reports stored on the Internet or company databases.

[0056] After obtaining the search results through manual retrieval, data and information are extracted from them, and then calculations and summaries are performed to write specific reports. Then, judgments are made on the development of related industries based on the information, so as to carry out loan financial business.

[0057] At present, this process has the following pain points:

[0058] 1) Search work requires a lot of manual participation. For example, to search for recent comments, news, reports or development trends of a specific industry or company, business personnel need to search from the Internet and the company's database. For online text, business personnel need to screen each web page one by one to eliminate useless junk information (such as some web pages without substantive content); and for data within the company, most medium-sized and small companies generally do not build their own search engines, or only have simple keyword search functions, and the search efficiency is low. In general, the first step of information collection and filtering requires more manpower and time;

[0059] 2) The information extraction process also requires a lot of manual participation. For most industry reports and articles related to industry development, the key content needs to be read, understood, extracted and summarized manually. Therefore, the information extraction process also requires a lot of manual participation.

[0060] In summary, when one wants to quickly track industry or company information, learn about the public opinion of an industry or company, and possible future developments, existing solutions require a large amount of manpower investment. If less manpower is invested, it is impossible to obtain comprehensive information in a timely manner.

[0061] The existing technology is sometimes supplemented by some traditional keywords, rules and traditional NLP models to complete the task, but these traditional methods are relatively weak and lack the ability to understand the semantics of articles, especially in the information extraction process. It is more difficult to complete the information extraction process through traditional technology.

[0062] Therefore, this application provides a solution, proposing a comprehensive information retrieval and analysis system based on a large language model agent and a traditional NLP model, aiming to combine traditional methods with new technologies to optimize the workflow in this field. The large model agent can determine the call of modules and the use of tools according to the user's specific instructions, meet the needs of automated report generation, and effectively assist business personnel in their work.

[0063] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a report generating device, etc. The following takes the report generating device as an example to illustrate this embodiment and the following embodiments.

[0064] The first embodiment of the report generation method of this application is proposed below. Figure 1 , Figure 1 This is a flowchart of the first embodiment of the report generation method of this application.

[0065] In this embodiment, the report generation method includes steps S10 to S30:

[0066] Step S10, in response to receiving a user input instruction, the first agent character assists the user interface module in generating a target keyword and a query phrase according to the user input instruction;

[0067] The report generation method in the embodiment of the present application is applied to an industry information retrieval reporting system, which includes multiple modules, such as a data source module, a data coarse screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summary module, a tool interface library module, a user interface module, and a role management module.

[0068] Optionally, the data source module is responsible for the management of local data and / or the network interface, and can provide documents containing keywords given a number of keywords. In the embodiment of the present application, elasticsearch is considered to be used as the management of local text data. The storage service will be responsible for functions such as word segmentation and fast indexing; the data source module can also return the full text of each document contained in the search results of the network search engine based on a given number of keywords.

[0069] Optionally, the data coarse screening module can call the data source module to obtain the document results of the keyword search. For the local database, the local elasticsearch storage retrieval keywords can be called, and the tf-idf or bm2.5 algorithm can be used to calculate the tf-idf value of each keyword in the document and use the weighted value as the score. The local database is preliminarily sorted and the first n1 documents are taken. For networked data, keyword search can be used and the first n2 articles are returned, and then the documents obtained from the local database and / or networked data are deduplicated to obtain coarse screening documents.

[0070] Optionally, the data sorting module can use more traditional NLP technology to sort documents according to the semantic requirements of target keywords.

[0071] Optionally, the semantic filtering module relies on the local language big model (target big model) to run. In this module, the target big model is adjusted to an intelligent role of a semantic filtering assistant through prompt engineering, which can understand the content and meaning of the article more deeply and filter and process the text according to the user's instructions.

[0072] Optionally, the semantic processing and summary module uses the big model to process and summarize the final filtered documents. In this module, the big model is adjusted to an intelligent role of a semantic processing assistant through prompt engineering. The big model traverses all articles, analyzes the articles according to user instructions, and obtains multiple results, or a summary result, and returns a final briefing. At the same time, after the big model is generated, the source of the article can be marked according to the article, which is convenient for users to query and verify the results or correctness of the report.

[0073] Optionally, the tool interface library module may provide an interface so that the large model can determine whether to call the tool model according to actual needs.

[0074] Optionally, the user interface module is the core module of the system, acting as a front end to directly interface with the user. The back end works on the big model, which plays the role of a central intelligent agent through prompt engineering guidance. It can generate target keywords, query phrases, and tool call requirements based on user input to complete the target report, and finally present the results to the front end user through the user interface module. The front end can collect user feedback information, that is, whether the results are good or bad, for future engineering algorithm personnel to optimize the process.

[0075] Optionally, the role management module is responsible for managing prompts (prompt word content) and is used to construct different large model Agents according to the content of each prompt word, that is, to adjust the target large model to each intelligent agent role.

[0076] Optionally, in response to receiving a user input instruction, the step of assisting the user interface module to generate a target keyword and a query phrase according to the user input instruction by the first agent role comprises:

[0077] Receiving the user input instruction through the user interface module;

[0078] The first intelligent agent role is called, and the user input instruction is analyzed by the first intelligent agent role to generate the target keyword and query phrase.

[0079] Optionally, the user's natural language input command is first received through the user interface module. The key to this step is to accurately parse the user's intention and convert the user's command into target keywords and query phrases that the system can understand and process. This usually involves natural language understanding (NLU) technology, which can parse the semantic content of the user's input and convert it into structured data, namely target keywords and query phrases.

[0080] Optionally, by generating target keywords and query phrases according to the user input instructions, it can be ensured that the system can accurately understand the user's query needs and extract key information for subsequent processing. Target keywords and query phrases are the basis for document retrieval and screening, and they will guide the system on how to operate in the next steps. By accurately parsing user instructions, the system can generate keywords and phrases that are highly relevant to the user's query intent, which provides a clear direction for subsequent document screening and sorting, thereby improving the efficiency and relevance of the entire information retrieval process.

[0081] Optionally, after the step of calling the first agent role, the step further includes:

[0082] The first agent role performs intent analysis on the user input instruction to generate a tool calling requirement, wherein the tool calling requirement is used to call a tool model to the tool interface library module.

[0083] Optionally, by performing intent analysis on user input instructions through the first intelligent agent role, calling parameters, i.e., tool calling requirements, can be provided to the tool calling module, so that the tool calling module can provide an interface for performing specific tasks according to the tool calling requirements, call the corresponding tool model, and perform tasks such as sentiment analysis and / or location extraction on the final filtered document to assist in generating a target report.

[0084] In the embodiment of the present application, the agent role (Agent) constructed by the role management module according to the prompt content (prompt) can perform user intent analysis, and the agent role adjusted based on the large model can understand natural language input and generate relevant target keywords, query phrases and tool call requirements. Prior to this, the large model needs to be adjusted through the prompt, and the resulting agent roles can adapt to specific query scenarios and the diversity of user instructions.

[0085] For example, the user inputs the command: "I need to know the latest developments in the steel industry in 2024." The target keywords generated by the system after parsing are: "2024", "steel industry", "latest developments". The query phrase is: "latest developments in the steel industry in 2024".

[0086] Step S20, screening and sorting documents based on the target keywords and query phrases to obtain refined sorted documents;

[0087] Furthermore, after generating target keywords and query phrases according to the user input instructions, the documents can be screened and sorted based on the target keywords and query phrases to obtain refined sorted documents.

[0088] Optionally, based on the generated target keywords and query phrases, the documents in the document library are screened and sorted. This step usually includes two sub-processes: first, a rough screening based on keywords, and second, sorting the screened documents using a more sophisticated sorting algorithm.

[0089] Optionally, the purpose of filtering and sorting is to find the most relevant document set to the user's query from a large number of documents. By sorting, the system can put the documents that are most likely to meet the user's needs in the front, so that users can quickly find the information they need. By filtering and sorting, the number of documents that need to be browsed can be reduced, improving the efficiency of information retrieval. At the same time, sorting ensures that documents that are more relevant to user needs have higher priority, thereby improving the accuracy of subsequent report generation.

[0090] In the embodiments of the present application, a variety of information retrieval techniques can be used, including traditional keyword-based retrieval methods and modern machine learning-based sorting algorithms. For example, Elasticsearch can be used for preliminary document screening, and then a pre-trained sorting model (such as BM25 or a BERT-based model) can be used to sort the screened documents. For example, using the target keywords "2024", "steel industry", and "latest developments", the system filters out relevant documents from the document library, and then uses a BERT-based sorting model to sort these documents to obtain a refined document list.

[0091] Step S30, semantically filter the refined document through the semantic filtering module, and call the second intelligent agent role to generate a final document; call the third intelligent agent role through the semantic processing module to semantically process the final document, and / or process the final document through the tool calling module to generate a target report.

[0092] Furthermore, after selecting and sorting documents based on target keywords and query phrases, the refined documents can be subjected to deeper semantic analysis and processing. This includes using advanced semantic understanding technology to filter documents to ensure that only documents that are highly relevant to the user's query intent are retained, and then summarizing and processing the retained documents to generate the final report.

[0093] Optionally, the purpose of semantic filtering and processing is to ensure that the report presented to the user not only contains relevant documents, but also that these documents can accurately reflect the user's query intent. Through semantic processing, the system can extract key information from the document and present it in a user-friendly manner. Through semantic filtering and processing, high-quality reports can be provided, which not only contain the information required by the user, but also have been optimized at the semantic level, so that users can obtain the information they need faster, and the accuracy and relevance of the information are guaranteed.

[0094] The semantic filtering module can use the big model to conduct in-depth analysis of the document content and determine the degree of match with the target keywords and query phrases. The semantic processing and summary module can use the big model to analyze the documents one by one, extract key information, and may call specific tool interfaces for additional task processing. Finally, the system summarizes this information and generates a target report.

[0095] This embodiment adopts the above scheme, specifically by responding to receiving user input instructions, the first agent role assists the user interface module to generate target keywords and query phrases according to the user input instructions, can directly connect with the user, analyze the target keywords and query phrases according to the user input instructions, and can filter out documents related to the target keywords from different data sources based on the target keywords, and then combine the query phrases to further filter the related documents filtered out from different data sources; by filtering and sorting the documents based on the target keywords and query phrases, a refined document is obtained, thereby further selecting documents with a higher relevance to the user input instructions from the initially filtered documents; through the semantic filtering module Semantic filtering is performed on the sorted documents, and the second intelligent agent role is called to generate the final document; the third intelligent agent role is called through the semantic processing module to perform semantic processing on the final document, and / or the final document is processed through the tool calling module to generate a target report. From the screened documents with a high correlation with the user input instructions, documents that better meet the user's needs can be further determined, and a target report with higher accuracy can be obtained after processing. This can avoid the dependence of the report generation process on manual labor, and with the help of the various intelligent agent roles provided by the role management module, automatic report generation is realized, which effectively reduces labor costs and saves the time required for report generation, thereby improving the efficiency of report generation and at the same time improving the accuracy of report generation.

[0096] Based on the above first embodiment, a second embodiment of the method for generating a report of this application is proposed. In this embodiment, the same or similar contents as those of the above first embodiment can be referred to the above introduction, and will not be described in detail later. Figure 2 , Figure 2This is a flow chart of the second embodiment of the method for generating a report for this application. In this embodiment, step S20 includes step S201 and step S202:

[0097] Step S201: inputting the target keyword and the first preset parameter into the data coarse screening module, and obtaining a coarse screening document through the data coarse screening module;

[0098] Optionally, the first preset parameter includes the first document quantity n1 and / or the second document quantity n2.

[0099] Optionally, the step of obtaining a rough screening document through the data rough screening module includes:

[0100] The data source module is called by the data coarse screening module, and the data source module performs weighted calculation and sorting on the documents in the local database according to the target keyword to obtain a first sorting result, and selects a first number of document results from the first sorting result; and / or, the data source module uses the target keyword to search for networked data and returns a second number of document results;

[0101] The coarse-screened documents are obtained by deduplicating the document results of the first number of documents and / or the document results of the second number of documents through local sensitive hashing.

[0102] Optionally, the input of the data coarse screening module is the target keyword, and the first preset parameter used includes the returned first document number n1 and / or second document number n2. It should be noted that the first document number n1 and the second document number n2 can be customized by the user, or can be automatically determined by the data coarse screening module based on empirical values ​​or actual needs. This is not specifically limited in the embodiments of the present application.

[0103] Optionally, the data coarse screening module calls the data source module to obtain the document results retrieved by the target keyword. For the local database, the local elasticsearch storage retrieval keyword can be called, and the tf-idf or bm2.5 algorithm is used to calculate the tf-idf value of each keyword in the article and use the weighted value as the score. The local database is preliminarily sorted and the first n1 documents are taken; for networked data, the target keyword is used for search and the first n2 documents are returned; local sensitive hashing (simhash) is used to deduplicate n1+n2 documents, and finally n coarse screening documents are obtained.

[0104] Alternatively, SimHash is a locality-sensitive hashing technology that is mainly used to detect duplicate content or find similar items in large-scale data sets. The core idea of ​​SimHash is to transform the similarity detection problem of high-dimensional data into a similarity comparison problem of hash values. By converting the input data (such as text) into feature vectors, and then performing a series of hash operations on these feature vectors, a hash signature (usually a fixed-length binary string) is finally generated for each data. If a large number of bits of the SimHash signatures of two data are the same, then the two data can be considered similar. SimHash is very effective in processing massive data because it can quickly screen out potentially similar data pairs, and then perform more detailed comparisons to confirm their similarity.

[0105] Optionally, the data rough screening module eventually returns n documents as rough screening documents, and n is generally much smaller than the sum of the number of documents in the database and available on the network, that is, n is much smaller than n1+n2.

[0106] Step S202: input the target keyword, query phrase, coarsely screened documents and second preset parameters into the data fine-ranking module, and obtain the fine-ranked documents through the data fine-ranking module.

[0107] Optionally, the second preset parameters include the third document quantity, a segmentation window size parameter (window) and a sliding distance (slide).

[0108] Optionally, the step of obtaining the refined document through the data refined arrangement module includes:

[0109] The rough screening document is symbolized by the data fine sorting module to obtain a digital list; and the rough screening document is divided into individual text paragraphs according to the segmentation window size parameter and the sliding distance, and the similarity score between each text paragraph and the target keyword is calculated;

[0110] Sorting the text paragraphs according to the similarity scores to obtain a second sorting result, and updating the number list according to the second sorting result to obtain a target number list;

[0111] The document results of the third number of documents in the target digital list are used as the refined sorted documents.

[0112] Optionally, the input of the data sorting module includes coarse-screened documents, target keywords and query phrases, and the second preset parameters used include the number of retained articles N, the segmentation window size window and the sliding distance slide.

[0113] Optionally, in the data ranking module, more traditional NLP technology will be used to sort documents according to the semantic requirements of the query keywords. The sorting models are mostly Bert-type models pre-trained for specific retrieval scenarios. The input is the concatenation of the query phrase, target keywords and text end, and the output is a relevance score between 0 and 1.

[0114] Reference Figure 3 , Figure 3 FIG. 1 is a schematic diagram showing a principle according to the second embodiment, as shown in FIG. Figure 3 As shown, the data sorting module runs. First, all input documents are tokenized to obtain a digital list. According to the window size and slide distance, they are divided into multiple text paragraphs of the same size as the window (for example, the first paragraph is text[0:window], and the second paragraph is text[slide:window+slide]). For each document, the similarity score between each paragraph and the query keyword and the text paragraph is calculated, and all article paragraphs are taken out and sorted. Then, from high to low scores, the article number to which this paragraph belongs is put into the list. If the number already exists in the list, it will not be put in. Finally, the first N documents in the digital list are returned, which are the sorted documents.

[0115] This embodiment adopts the above scheme, specifically by inputting the target keyword and the first preset parameter into the data coarse screening module, and obtaining the coarse screening document through the data coarse screening module; inputting the target keyword, the query phrase, the coarse screening document and the second preset parameter into the data fine sorting module, and obtaining the fine sorting document through the data fine sorting module, it is possible to screen out relevant coarse screening documents from different data sources, and then further screen the relevant documents screened out from different data sources in combination with the query phrase; by screening and sorting documents based on the target keyword and the query phrase, fine sorting documents are obtained, and fine sorting documents with high relevance to the user input instructions are further selected from the screened coarse screening documents, ensuring that a target report with high accuracy is obtained after subsequent processing, thereby avoiding the dependence of the report generation process on manual labor, realizing automatic report generation, effectively reducing labor costs, saving the time required for report generation, thereby improving the efficiency of report generation, and at the same time improving the accuracy of report generation.

[0116] Based on the above first embodiment and / or second embodiment, a third embodiment of the method for generating a report of the present application is proposed. In this embodiment, the same or similar contents as those of the above first embodiment and / or second embodiment can be referred to the above introduction, and will not be described in detail later. Figure 4 , Figure 4This is a flow chart of the third embodiment of the method for generating a report for this application. In this embodiment, step S30 includes step S301 and step S302:

[0117] Step S301: semantically filter the refined document through the semantic filtering module, and call the second agent role to generate a final document;

[0118] Optionally, the step of performing semantic filtering on the refined document by the semantic filtering module and calling the second agent role to generate a final document includes:

[0119] Acquire the refined sorted documents, target keywords, query phrases and first preset instructions through the semantic filtering module;

[0120] Calling the second intelligent agent role, and judging whether each document in the refined arrangement document meets the target keyword and the query phrase through the second intelligent agent role according to the first preset instruction;

[0121] The document that meets the target keyword and the query phrase in the refined sorted documents is used as the final document.

[0122] Optionally, the input of the semantic filtering module is the refined documents, target keywords and query phrases selected in the above-mentioned embodiments and the first preset instruction. The semantic filtering module relies on the local language large model (i.e., the target large model) to run (e.g., a model of level 14B and above). In this module, the large model is adjusted to the role of a semantic filtering assistant, i.e., the second intelligent agent role, through prompt engineering, which can more deeply understand the content and meaning of the article, and filter and process the text according to the user's instructions.

[0123] Optionally, the semantic filtering module connects the first preset instruction (such as a filtering instruction), the target keyword, the query phrase and the document content, and inputs them into the second intelligent agent role. It should be noted that the second preset instruction can be customized by the user, or it can be automatically generated by the semantic filtering module according to actual needs, and this is not specifically limited in the embodiments of the present application. For example, in this embodiment, the first preset instruction is a filtering instruction, which requires the second intelligent agent role to read the full text, and determine whether the document is consistent with the target keyword and the content of the query phrase, and return one of the three reply results ["yes", "no", "uncertain"]. Filter out documents with a no result. At the same time, you can choose to retain or filter uncertain documents according to actual needs to obtain the final document. The final document will be input into the semantic processing and summary module as a carefully selected result. Generally speaking, the number of final documents will range from dozens to one hundred, and this is not specifically limited in this embodiment.

[0124] Step S302: calling the third agent role through the semantic processing module to perform semantic processing on the final document, and / or processing the final document through the tool calling module to generate a target report.

[0125] Optionally, the step of calling the third agent role to perform semantic processing on the final document through the semantic processing module, and / or processing the final document through the tool calling module to generate a target report includes:

[0126] The final document and the second preset instruction are obtained through the semantic processing and aggregation module, the third intelligent agent role is called, and the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the final result corresponding to each document in the final document, or the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the intermediate result corresponding to each document in the final document, and the intermediate results corresponding to each document in the final document are aggregated to obtain the aggregated result, and the target report is generated according to the final result corresponding to each document in the final document or the aggregated result; and / or,

[0127] The target report is obtained by performing specific task processing on the final document through the tool calling module, wherein the specific task processing includes sentiment analysis and / or content extraction.

[0128] Optionally, the input of the semantic processing and summarization module is the final document obtained by the filtering process and the second preset instruction. The semantic processing and summarization module uses the third agent role adjusted by the large model to process and summarize the final filtering results. In this module, the large model is adjusted to the role of a semantic processing assistant, that is, the third agent role, through prompt engineering. The second preset instruction will contain specific tasks, such as: "Find out the data related to the steel industry in 2024", "What specific information does this article talk about related to the furniture industry? Please help me extract the summary and return it to me in the xxx format, and strive to be detailed."

[0129] Optionally, the semantic processing and aggregation module has two processing modes, one is to operate document by document and output the results of multiple documents; the other is to operate document by document, output intermediate results, and then aggregate the results. Optionally, the third intelligent agent role traverses all the final documents, analyzes the documents according to the second preset instructions, and obtains multiple results, or a summary result, and returns a final result document, which is the target report; and / or, the tool calling module performs specific task processing on the final document, such as sentiment analysis and / or location extraction, to obtain the final target report. Optionally, after the third intelligent agent role completes the processing, the document source can be marked on the result document to facilitate the user to query and verify the results or correctness of the report.

[0130] This embodiment adopts the above scheme, specifically filters the sorted documents through the semantic filtering module to obtain the final document, which can more deeply understand the content and meaning of the article, and filters and processes the text according to the first preset instruction, so that the obtained final document better meets the needs of users; the final document is semantically processed through the semantic processing and aggregation module, and / or the final document is processed for specific tasks through the tool calling module to obtain the target report; wherein, the specific task processing includes sentiment analysis and / or content extraction, and specific processing tasks can be performed according to the second preset instruction to process the final document into a form that meets the user's needs, which can improve the user experience.

[0131] For example, to help understand the implementation process of the report generation method obtained by combining this embodiment with the first embodiment and / or the second embodiment, please refer to Figure 5 , Figure 5 A brief flowchart of a report generation method is provided, specifically:

[0132] The overall workflow of the industry information retrieval reporting system in the embodiment of the present application is as follows:

[0133] 1. Users give instructions in the interface template using natural language;

[0134] 2. The user interface module uses the intelligent agent built by the role management module to analyze user intentions and generate target keywords, query phrases, and tool call requirements;

[0135] 3. Go all the way down from the right path to complete data acquisition, document rough screening and document fine sorting, and obtain fine sorted documents. In this process, the intelligent agent role is used to autonomously generate target keywords and complete the retrieval work;

[0136] 4. Filter and process the documents through the semantic filtering module and the semantic processing module. At the same time, the tool calling module can process the final document according to the parameters generated by the analysis (tool calling requirements) to obtain the result document, that is, the target document, and return the target document to the front end for users to view and use.

[0137] For example, the following examples are provided in the embodiments of the present application to further illustrate the overall workflow of the industry information retrieval report system:

[0138] 1. Application scenario: A business person wants to find out the positive or negative news about the domestic steel industry in the near future, and wants to find analytical articles related to the industry and analyze the positive and negative emotions of these articles.

[0139] Therefore, users can fill in the format in the system interface:

[0140] Please enter the content you want to query:

[0141] I want to find out the reports related to the domestic steel industry in the first four months of 2024, which can reflect whether the overall situation of the industry is good or bad.

[0142] Semantic processing instructions: Summarize the main content of the article in one or two sentences.

[0143] 2. A large model of the central role of the user interface, based on user input, generates:

[0144] Target keywords: 2024, first quarter, first four months, China, steel industry, report, research report, positive, rise, growth, fall, sudden collapse, bankruptcy, restructuring.

[0145] Query phrase: China's steel industry in the first quarter of 2024, positive or negative reports, industry conditions.

[0146] Tool call requirements: sentiment analysis tool.

[0147] 3. Based on the above generated results, data is extracted and filtered according to the above process. During this period, many irrelevant documents will be filtered out, such as expired documents (such as last year), documents containing target keywords but whose content does not meet the requirements, etc.

[0148] 4. Summarize each article based on the semantic processing instructions provided by the user.

[0149] 5. At the same time, the sentiment analysis tool can be automatically called to determine whether the sentiment of each document in the result is positive or negative.

[0150] 6. Return the results of 4 and 5. Users can read an overall overview and sentiment with very little time cost and have a general understanding of the industry.

[0151] Through the above solution, this embodiment enables business personnel to obtain the following advantages in the same scenario:

[0152] 1. Savings in labor costs: This application transfers most of the entire process to intelligent systems and large model agents for autonomous retrieval and aggregation, without manual retrieval, sorting, and aggregation, allowing business personnel to focus on the data and information in industry reports, effectively saving labor costs;

[0153] 2. Savings in time costs: In actual use, an industry report that integrates multiple information sources can be completed in a short time (several minutes to ten minutes). Compared with manual and traditional technical solutions, it can significantly improve efficiency and provide reports in a timely manner;

[0154] 3. Rich functions, scalability and customization: New tools can be continuously introduced into the toolbox, and different information retrieval and summary requirements can be customized using natural language. It is easy to use, has low implementation difficulty for business personnel, and has strong operability.

[0155] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the report generation method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0156] This application also provides a report generation device, please refer to Figure 6 , Figure 6 The module structure diagram of the report generation device provided in the present application is applied to an industry information retrieval report system, and the industry information retrieval report system includes at least one of a data source module, a data rough screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summarization module, a tool interface library module, a user interface module, and a role management module. The role management module is used to adjust the target macro model to at least one of a first agent role, a second agent role, and a third agent role according to the content of each prompt word. The report generation device includes:

[0157] A response module, configured to, in response to receiving a user input instruction, assist the user interface module in generating a target keyword and a query phrase according to the user input instruction through the first agent role;

[0158] A screening module is used to screen and sort documents based on the target keywords and query phrases to obtain refined documents;

[0159] A generation module is used to perform semantic filtering on the refined document through the semantic filtering module, and call the second intelligent agent role to generate a final document; call the third intelligent agent role to perform semantic processing on the final document through the semantic processing module, and / or process the final document through the tool calling module to generate a target report.

[0160] Optionally, in response to receiving a user input instruction, the step of assisting the user interface module to generate a target keyword and a query phrase according to the user input instruction by the first agent role comprises:

[0161] Receiving the user input instruction through the user interface module;

[0162] The first intelligent agent role is called, and the user input instruction is analyzed by the first intelligent agent role to generate the target keyword and query phrase.

[0163] Optionally, after the step of calling the first agent role, the step further includes:

[0164] The first agent role performs intent analysis on the user input instruction to generate a tool calling requirement, wherein the tool calling requirement is used to call a tool model to the tool interface library module.

[0165] Optionally, the step of screening and sorting documents based on the target keywords and query phrases to obtain refined sorted documents includes:

[0166] Inputting the target keyword and the first preset parameter into the data coarse screening module, and obtaining a coarse screening document through the data coarse screening module;

[0167] The target keyword, query phrase, coarsely screened document and second preset parameter are input into the data fine sorting module, and the fine sorting document is obtained through the data fine sorting module.

[0168] Optionally, the first preset parameter includes the first document quantity and / or the second document quantity, and the step of obtaining the coarsely screened documents by the data coarse screening module includes:

[0169] The data source module is called by the data coarse screening module, and the data source module performs weighted calculation and sorting on the documents in the local database according to the target keyword to obtain a first sorting result, and selects a first number of document results from the first sorting result; and / or, the data source module uses the target keyword to search for networked data and returns a second number of document results;

[0170] The coarse-screened documents are obtained by deduplicating the document results of the first number of documents and / or the document results of the second number of documents through local sensitive hashing.

[0171] Optionally, the second preset parameters include the third document quantity, a segmentation window size parameter, and a sliding distance, and the step of obtaining the refined document through the data refined sorting module includes:

[0172] The rough screening document is symbolized by the data fine sorting module to obtain a digital list; and the rough screening document is divided into individual text paragraphs according to the segmentation window size parameter and the sliding distance, and the similarity score between each text paragraph and the target keyword is calculated;

[0173] Sorting the text paragraphs according to the similarity scores to obtain a second sorting result, and updating the number list according to the second sorting result to obtain a target number list;

[0174] The document results of the third number of documents in the target digital list are used as the refined sorted documents.

[0175] Optionally, the step of performing semantic filtering on the refined document by the semantic filtering module and calling the second agent role to generate a final document includes:

[0176] Acquire the refined sorted documents, target keywords, query phrases and first preset instructions through the semantic filtering module;

[0177] Calling the second intelligent agent role, and judging whether each document in the refined arrangement document meets the target keyword and the query phrase through the second intelligent agent role according to the first preset instruction;

[0178] The document that meets the target keyword and the query phrase in the refined sorted documents is used as the final document.

[0179] Optionally, the step of calling the third agent role to perform semantic processing on the final document through the semantic processing module, and / or processing the final document through the tool calling module to generate a target report includes:

[0180] The final document and the second preset instruction are obtained through the semantic processing and aggregation module, the third intelligent agent role is called, and the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the final result corresponding to each document in the final document, or the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the intermediate result corresponding to each document in the final document, and the intermediate results corresponding to each document in the final document are aggregated to obtain the aggregated result, and the target report is generated according to the final result corresponding to each document in the final document or the aggregated result; and / or,

[0181] The target report is obtained by performing specific task processing on the final document through the tool calling module, wherein the specific task processing includes sentiment analysis and / or content extraction.

[0182] The report generation device provided by the present application adopts the report generation method in the above embodiment, which can solve the technical problem of report generation. Compared with the prior art, the beneficial effects of the report generation device provided by the present application are the same as the beneficial effects of the report generation method provided by the above embodiment, and other technical features in the report generation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0183] The present application provides a report generating device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the report generating method in the above-mentioned embodiment 1.

[0184] Reference below Figure 7 , which shows a schematic diagram of the structure of a report generating device suitable for implementing the embodiment of the present application. The report generating device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7 The report generating device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0185] like Figure 7As shown, the report generating device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the report generating device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the report generating device to communicate with other devices wirelessly or by wire to exchange data. Although the report generating device with various systems is shown in the figure, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0186] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0187] The report generation device provided by the present application adopts the report generation method in the above embodiment to solve the technical problem of report generation. Compared with the prior art, the beneficial effects of the report generation device provided by the present application are the same as the beneficial effects of the report generation method provided by the above embodiment, and other technical features in the report generation device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0188] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0189] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0190] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the report generating method in the above-mentioned embodiment.

[0191] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0192] The computer-readable storage medium may be included in the report generating device; or may exist independently without being assembled into the report generating device.

[0193] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the report generating device, the report generating device: by responding to receiving user input instructions, the first intelligent agent role assists the user interface module to generate target keywords and query phrases according to the user input instructions, can directly connect with the user, analyze the target keywords and query phrases according to the user's input instructions, and can filter out documents related to the target keywords from different data sources based on the target keywords, and then combine with the query phrases to further filter the relevant documents filtered out from different data sources; by filtering and sorting the documents based on the target keywords and query phrases, a refined document is obtained, thereby further selecting documents related to the user input from the initially filtered documents. Input instructions with high relevance; semantic filtering is performed on the sorted documents through the semantic filtering module, and the second intelligent agent role is called to generate a final document; the final document is semantically processed by calling the third intelligent agent role through the semantic processing module, and / or the final document is processed by the tool calling module to generate a target report. From the screened documents with high relevance to the user input instructions, documents that better meet the user's needs can be further determined, and a target report with higher accuracy can be obtained after processing. Thereby, the report generation process can avoid dependence on manual labor, and with the help of the intelligent agent roles provided by the role management module, automatic report generation is realized, which effectively reduces labor costs and saves the time required for report generation, thereby improving the efficiency of report generation and at the same time improving the accuracy of report generation.

[0194] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0195] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0196] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0197] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned report generation method, and can solve the technical problem of report generation. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the report generation method provided in the above-mentioned embodiment, and will not be repeated here.

[0198] The present application also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the steps of the report generation method as described above are implemented.

[0199] The computer program product provided by the present application can solve the technical problem of report generation. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the report generation method provided by the above embodiment, which will not be described in detail here.

[0200] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A report generation method, characterized in that: The method is applied to an industry information retrieval and reporting system, which includes at least one of a data source module, a data rough screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summarization module, a tool interface library module, a user interface module, and a role management module. The role management module is used to adjust the target macro model to at least one of a first agent role, a second agent role, and a third agent role according to the content of each prompt word. The method includes: In response to receiving a user input instruction, the user interface module is assisted by the first agent role to generate a target keyword and a query phrase according to the user input instruction; Screening and sorting documents based on the target keywords and query phrases to obtain refined documents; The semantic filtering module is used to perform semantic filtering on the refined document, and the second intelligent agent role is called to generate a final document; the third intelligent agent role is called to perform semantic processing on the final document through the semantic processing module, and / or the final document is processed through the tool calling module to generate a target report.

2. The method according to claim 1, characterized in that In response to receiving the user input instruction, the step of assisting the user interface module to generate target keywords and query phrases according to the user input instruction by the first agent role comprises: Receiving the user input instruction through the user interface module; The first intelligent agent role is called, and the user input instruction is analyzed by the first intelligent agent role to generate the target keyword and query phrase.

3. The method according to claim 2, characterized in that The step of calling the first agent role also includes: The first agent role performs intent analysis on the user input instruction to generate a tool calling requirement, wherein the tool calling requirement is used to call a tool model to the tool interface library module.

4. The method according to claim 1, characterized in that The step of screening and sorting documents based on the target keywords and query phrases to obtain finely sorted documents comprises: Inputting the target keyword and the first preset parameter into the data coarse screening module, and obtaining a coarse screening document through the data coarse screening module; The target keyword, query phrase, coarsely screened document and second preset parameter are input into the data fine sorting module, and the fine sorting document is obtained through the data fine sorting module.

5. The method according to claim 4, characterized in that The first preset parameter includes the first document quantity and / or the second document quantity, and the step of obtaining the rough screening documents through the data rough screening module includes: The data source module is called by the data coarse screening module, and the data source module performs weighted calculation and sorting on the documents in the local database according to the target keyword to obtain a first sorting result, and selects a first number of document results from the first sorting result; and / or, the data source module uses the target keyword to search for networked data and returns a second number of document results; The coarse-screened documents are obtained by deduplicating the document results of the first number of documents and / or the document results of the second number of documents through local sensitive hashing.

6. The method according to claim 4, characterized in that The second preset parameters include the third document quantity, the segmentation window size parameter and the sliding distance, and the step of obtaining the refined document through the data refined sorting module includes: The rough screening document is symbolically converted by the data fine sorting module to obtain a digital list; and the rough screening document is segmented into individual text paragraphs according to the segmentation window size parameter and the sliding distance, and the similarity score between each text paragraph and the target keyword is calculated; Sorting the text paragraphs according to the similarity scores to obtain a second sorting result, and updating the number list according to the second sorting result to obtain a target number list; The document results of the third number of documents in the target digital list are used as the refined sorted documents.

7. The method according to claim 1, characterized in that The step of performing semantic filtering on the refined document through the semantic filtering module and calling the second agent role to generate a final document comprises: Acquire the refined sorted documents, target keywords, query phrases and first preset instructions through the semantic filtering module; Calling the second intelligent agent role, and judging whether each document in the refined arrangement document meets the target keyword and the query phrase through the second intelligent agent role according to the first preset instruction; The document that meets the target keyword and the query phrase in the refined sorted documents is used as the final document.

8. The method according to claim 1, characterized in that The step of calling the third agent role to perform semantic processing on the final document through the semantic processing module, and / or processing the final document through the tool calling module to generate a target report includes: The final document and the second preset instruction are obtained through the semantic processing and aggregation module, the third intelligent agent role is called, and the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the final result corresponding to each document in the final document, or the third intelligent agent role analyzes each document in the final document one by one according to the second preset instruction to obtain the intermediate result corresponding to each document in the final document, and the intermediate results corresponding to each document in the final document are aggregated to obtain the aggregated result, and the target report is generated according to the final result corresponding to each document in the final document or the aggregated result; and / or, The target report is obtained by performing specific task processing on the final document through the tool calling module, wherein the specific task processing includes sentiment analysis and / or content extraction.

9. A report generating device, characterized in that: The device is applied to an industry information retrieval and reporting system, which includes at least one of a data source module, a data rough screening module, a data fine sorting module, a semantic filtering module, a semantic processing and summarization module, a tool interface library module, a user interface module, and a role management module. The role management module is used to adjust the target macro model to at least one of a first agent role, a second agent role, and a third agent role according to the content of each prompt word. The device includes: A response module, configured to, in response to receiving a user input instruction, assist the user interface module in generating a target keyword and a query phrase according to the user input instruction through the first agent role; A screening module is used to screen and sort documents based on the target keywords and query phrases to obtain refined documents; A generation module is used to perform semantic filtering on the refined document through the semantic filtering module, and call the second intelligent agent role to generate a final document; call the third intelligent agent role to perform semantic processing on the final document through the semantic processing module, and / or process the final document through the tool calling module to generate a target report.

10. A report generating device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the report generation method according to any one of claims 1 to 8.

11. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the report generation method according to any one of claims 1 to 8 are implemented.

12. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the report generation method according to any one of claims 1 to 8 are implemented.

Images