Information reorganizing method, device, equipment and program product

Abstract

The invention provides an information reorganization method and device, equipment and a program product, and the method comprises the steps: calling a pre-trained large language model, retrieving an information document corresponding to an information reorganization demand from an information document library, retrieving an information abstract related to the information document from an information abstract library, and storing the information abstract related to the information document in the information abstract library; the information abstract library is used for storing information abstracts extracted from information documents; calling the large language model, determining an information reorganization outline, and reorganizing the retrieved information abstract according to the information reorganization outline to obtain an information reorganization result. According to the method, accurate information reorganization can be realized, so that a user or a reader can effectively obtain wanted information from mass information.

Description

Technical Field

[0001] The present application relates to the field of natural language processing technology, and in particular to an information compilation method, device, equipment and program product. Background Art

[0002] In today's era of information explosion, the popularity of the Internet and mobile devices allows people to access massive amounts of information anytime and anywhere. However, this information overload also brings new challenges: information fragmentation, duplication, redundancy, and difficulty in distinguishing true from false are becoming increasingly prominent. Faced with massive amounts of information, it is difficult for the media or readers to obtain the most valuable or desired information in a timely and accurate manner. Summary of the invention

[0003] Based on the above technical status, the present application proposes an information compilation method, device, equipment and program product, which can achieve accurate information compilation, thereby facilitating users or readers to effectively obtain desired information from massive information.

[0004] The first aspect of the present application provides an information compilation method, comprising:

[0005] Calling a pre-trained large language model, retrieving an information document corresponding to the information compilation requirement from an information document library, and retrieving an information summary related to the information document from an information summary library, wherein the information summary library is used to store information summaries extracted from the information document;

[0006] The large language model is called to determine an information compilation outline, and the retrieved information abstracts are compiled according to the information compilation outline to obtain an information compilation result.

[0007] In some implementations, calling the large language model to determine the information compilation outline includes:

[0008] The large language model is called to generate an information compilation outline that is adapted to the retrieved information summary, or the large language model is called to parse a preset compilation report template to obtain the information compilation outline.

[0009] In some implementations, calling the large language model to generate an information compilation outline adapted to the retrieved information summary includes:

[0010] The large language model is called to perform the following processing:

[0011] Clustering the retrieved information summaries to obtain at least one information summary group;

[0012] Extract key concepts and analyze the relevance of key concepts from the obtained information summary groups, and integrate the content of the information summary based on the relevance between the key concepts;

[0013] Sort the content display order of the information summary after content integration to generate an information outline;

[0014] The information outline is structured and integrated to obtain an information compilation outline.

[0015] In some implementations, the calling of the large language model to generate an information compilation outline adapted to the retrieved information summary further includes:

[0016] Calling the large language model to subject-classify the retrieved information abstracts, and acquiring a preset information compilation outline corresponding to the subject-classification result from an outline library according to the subject-classification result;

[0017] Furthermore, the large language model is called to compare the generated information compilation outline with the preset information compilation outline, and the generated information compilation outline is supplemented with missing content; the missing content includes outline content recorded in the preset information compilation outline and not recorded in the generated information compilation outline.

[0018] In some implementations, calling the large language model to organize the retrieved information summary according to the information organization outline to obtain an information organization result includes:

[0019] The large language model is called to perform the following processing:

[0020] Generate an information compilation title according to the information compilation requirement;

[0021] According to the outline title in the information compilation outline, extract information from the retrieved information summary to obtain information content corresponding to the outline title;

[0022] According to the structure of the information compilation outline, the information compilation titles and the information contents corresponding to the outline titles are integrated to obtain an information compilation result.

[0023] In some implementations, calling a pre-trained large language model, retrieving an information document corresponding to the information compilation requirement from an information document library, and retrieving an information summary related to the information document from an information summary library includes:

[0024] Call the pre-trained large language model to perform the following processing:

[0025] Retrieve information documents corresponding to the information compilation requirements from the information document library;

[0026] Analyze the information compilation requirements and determine information compilation restriction conditions;

[0027] Based on the information compilation restriction condition, selecting target information documents that meet the information compilation restriction condition from information documents corresponding to the information compilation requirement;

[0028] An information summary related to the target information document is retrieved from an information summary library.

[0029] In some implementations, the method further includes:

[0030] When an information document is acquired, the large language model is called to extract an information summary from the acquired information document, and the extracted information summary is stored in the information summary library.

[0031] In some implementations, the method further includes:

[0032] Clustering the retrieved information summaries to obtain at least one information summary group;

[0033] For each of the information summary groups, integrating the information summaries therein;

[0034] The step of organizing the retrieved information summaries according to the information organization outline to obtain information organization results includes organizing the information in each information summary group according to the information organization outline to obtain information organization results.

[0035] The second aspect of the present application provides an information compilation device, comprising:

[0036] An information retrieval unit, used to call a pre-trained large language model, retrieve information documents corresponding to information compilation requirements from an information document library, and retrieve information summaries related to the information documents from an information summary library, wherein the information summary library is used to store summaries extracted from information documents;

[0037] The information compilation unit is used to call the large language model, determine the information compilation outline, and compile the retrieved information abstracts according to the information compilation outline to obtain the information compilation result.

[0038] The third aspect of the present application provides an information compilation device, comprising:

[0039] Memory and processor;

[0040] The memory is connected to the processor and is used to store programs;

[0041] The processor is used to implement the above-mentioned information compilation method by running the program in the memory.

[0042] A fourth aspect of the present application provides a computer program product, comprising computer program instructions, which, when executed by a processor, enable the processor to execute the above-mentioned information compilation method.

[0043] The information compilation method proposed in this application uses a large language model to perform information compilation tasks. In the process of performing the information compilation task, the information document corresponding to the information compilation demand is first retrieved from the information document library, and the information summary related to the information document is retrieved from the information summary library, and then the information compilation outline is determined, and the retrieved information summary is compiled according to the information compilation outline to obtain the information compilation result. The above-mentioned information compilation method uses the powerful context understanding ability and natural language understanding ability of the large language model to comprehensively analyze the background, details and core elements in the information summary to compile the information summary, thereby ensuring that the information compilation result is more logical and coherent. This method can more efficiently and accurately generate information compilation results with clear logic and comprehensive content, so that users or readers can effectively obtain the desired information from massive information. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0045] Figure 1 A flowchart of the information compilation method provided in an embodiment of the present application.

[0046] Figure 2 Another processing flow diagram of the information compilation method provided in an embodiment of the present application.

[0047] Figure 3 A schematic diagram of the structure of an information compilation device provided in an embodiment of the present application.

[0048] Figure 4 A schematic diagram of the structure of an information compilation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In today's era of information explosion, the popularity of the Internet and mobile devices allows people to access massive amounts of information anytime and anywhere. However, this information overload also brings new challenges: information fragmentation, duplication, redundancy, and difficulty in distinguishing true from false are becoming increasingly prominent. Faced with massive amounts of information, it is difficult for the media or readers to obtain the most valuable or desired information in a timely and accurate manner.

[0050] For example, in the field of news, massive amounts of news information are generated every moment. Faced with massive amounts of news content, especially when encountering sudden hot events, it is difficult for the media or readers to obtain the most valuable information in a timely and accurate manner.

[0051] Information compilation is an effective means for readers to accurately obtain information of interest from massive information. Through information compilation, it is possible to extract and organize the main information of massive information, thereby improving the readability of massive information and enabling readers to obtain effective information according to their needs. Specifically, information compilation has the following advantages:

[0052] First, we need to deal with information overload and improve information acquisition efficiency. Millions of new articles, videos, and social media posts appear on the Internet every day, but people's time and attention are limited. Through information compilation, we can filter out the most important and relevant information, helping readers obtain key information in a short time and avoid getting lost in the flood of information.

[0053] Secondly, improve the readability and comprehensibility of information. The original information may contain a lot of professional terms, complex data and background information, which may be difficult for ordinary readers to fully understand. By summarizing and summarizing the information content, extracting the core points, and presenting them in easy-to-understand language, readers can enhance their understanding and absorption of the information.

[0054] Finally, it helps decision-making and research in professional fields. For enterprises and researchers, timely access to accurate hot spot information is crucial for decision-making and market analysis. The compiled information can provide in-depth insights and trend analysis to support their strategic planning and risk assessment.

[0055] To sum up, in the context of information overload, compiling information is not only a necessary means to meet personal information needs, but also an important way to improve the overall information literacy and decision-making efficiency of society. It can help readers quickly obtain high-quality information and promote the efficient dissemination of information and maximize its value.

[0056] Based on the above technical status and needs, the embodiment of the present application proposes an information compilation method that can achieve efficient and accurate information compilation, so that readers can compile information according to their own needs when faced with massive amounts of information, thereby obtaining the most valuable or desired information in a timely and accurate manner.

[0057] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] The embodiment of the present application first proposes an information compilation method, which can be applied to an information processing system, which can be run on a computer, an intelligent terminal, a server, or other equipment with data storage and processing functions. Exemplarily, the information processing system can adopt a client + server architecture, interact with the user through the client, obtain information compilation requirements, and then perform information compilation processing based on the information compilation requirements through the server, and finally output the information compilation results to the user through the client. The information compilation method proposed in this application is mainly based on information compilation requirements for information compilation processing, so the subsequent embodiments will focus on introducing the specific processing process of information compilation processing based on information compilation requirements.

[0059] See also Figure 1 As shown, the information compilation method proposed in the embodiment of the present application includes:

[0060] S101, calling a pre-trained large language model, retrieving information documents corresponding to information compilation requirements from an information document library, and retrieving information summaries related to the information documents from an information summary library.

[0061] The above-mentioned information compilation demand refers to the demand for compilation of specific information proposed by users or readers, which can be in the form of instructions or requests. For example, if a user needs to compile hot news events in August, he can input the information compilation demand "What are the hot events in August, compile them into news" through the client of the information processing system.

[0062] The above-mentioned information document library refers to a database for storing information documents. The above-mentioned information documents generally refer to documents used to record information, such as news manuscripts, papers, books, reports, notes, etc. In some embodiments, the above-mentioned information document library can be divided into information document libraries of different fields or types, such as news libraries, paper libraries, etc., according to the fields or types of the stored information documents.

[0063] The above-mentioned information document library can be a pre-built database containing information documents in various fields or various types, or a database containing information documents in certain specific fields or specific types. In some embodiments, the above-mentioned information document library can also be a database composed of information documents input by the user while inputting the information compilation requirements. For example, the user can input a large number of papers or books into the information processing system, and then input the information compilation requirements into the system, asking the system to compile the information of these papers or books. At this time, these papers or books constitute the information document library.

[0064] The above-mentioned information summary library is used to store information summaries extracted from information documents. The above-mentioned information summary refers to summary information extracted from information documents. An information summary specifically includes a summary title and summary content. In an embodiment of the present application, for information documents stored in the above-mentioned information document library, information summaries are extracted respectively, and the extracted information summaries are stored in the information summary library.

[0065] Exemplarily, in the process of constructing the above-mentioned information document library, before storing the information document in the information document library, the information summary is first extracted from the information document to obtain at least one information summary extracted from the information document, and then the extracted information summary is stored in the information summary library, and the information document is stored in the information document library.

[0066] In some embodiments, the information summary of the information document can be extracted by calling the large language model. For example, when the information document is obtained, the large language model is called to extract the information summary from the obtained information document, and the extracted information summary is stored in the information summary library.

[0067] For example, for news, each news article may include multiple sub-topics, so we can extract the sub-topics in the news, generate the corresponding sub-topic titles and summaries, and obtain the news summary. Input the following prompt to the large language model to call the large language model to perform the above news summary extraction task:

[0068] You will receive a piece of news, which includes a title, a body of text and a time, where the time is the time when the news is edited.

[0069] In a news article, multiple events are described. These events may represent the same subtopic or different subtopics.

[0070] A subtopic can be defined as an independent subject area mentioned in the article, which focuses on a specific event, situation, behavior or issue. Subtopics usually revolve around a central point with a specific time, place, person or event.

[0071] Your task is to list all the subtopics covered in this article and give the title of the topic and a summary description of the topic.

[0072] Be careful not to make the sub-topics too granular or too detailed. The topics given should cover all the news content without any omissions.

[0073] Note that the sub-topic title should be detailed. The title should give specific time, place, people and other event parameters, so that people can have a preliminary understanding of the topic based on the title.

[0074] Enter the news as:

[0075] {query}

[0076] Now please analyze the above news and list all the sub-topics involved. No more explanation is needed. Please return the results in the standard format. Each line is a group of topics in the format of {"Title":"XX","Abstract":"XXX"}.

[0077] In the above prompt, "query" refers to the news document from which the news summary is to be extracted. The information processing system inputs the above prompt containing the news document into the large language model, so that the large language model can extract the news summary from the news document according to the instructions of the prompt.

[0078] Through the above processing, each group of topics output by the large language model is a news summary. Each news summary can be stored in a news summary library, and the original news document can be stored in the news library.

[0079] Based on the above-mentioned information document library and information summary library, when the information processing system obtains information compilation requirements, it calls the pre-trained large language model to perform information compilation processing corresponding to the information compilation requirements.

[0080] First, the large language model is called to retrieve information documents corresponding to the information compilation requirements from the information document library. Then, based on the retrieved information documents, information summaries related to the retrieved information documents are further retrieved from the information summary library.

[0081] Exemplarily, by inputting a prompt into the large language model for instructing the large language model to retrieve information documents corresponding to the information compilation requirements from the information document library, and further retrieve information summaries related to the retrieved information documents from the information summary library, the large language model can be called to perform the above-mentioned information document and information summary retrieval tasks. Among them, a prompt can be used to instruct the large language model to retrieve information documents corresponding to the information compilation requirements from the information document library, and further retrieve information summaries related to the retrieved information documents from the information summary library, or a prompt can be used to instruct the large language model to retrieve information documents corresponding to the information compilation requirements from the information document library, and then another prompt can be used to instruct the large language model to retrieve information summaries related to the retrieved information documents from the information summary library.

[0082] Taking news compilation as an example, the news library adopts the form of MySQL database. The information processing system can call the large language model to generate SQL query statements according to the news compilation requirements, and then retrieve news documents corresponding to the news compilation requirements from the news library through the SQL query statements.

[0083] For example, you can input the following prompt into the large language model to enable it to perform news retrieval tasks:

[0084] Now I give you a MySQL database named {database_name}, which has a table named {bw_table_name} that stores hot news.

[0085] The table contains the fields id (unique identifier string), name (file name string), title (news title), content (news text string), and time (date).

[0086] The table structure is:

[0087] CREATE TABLE`{bw_table_name}`(

[0088] `id`varchar(32)NOT NULL COMMENT'Primary key ID, unique',

[0089] `name`varchar(512)NOT NULL COMMENT'file name',

[0090] `time`datetime DEFAULT NULL COMMENT'report time',

[0091] `title`varchar(512)DEFAULT NULL COMMENT'Title',,

[0092] PRIMARY KEY (`id`)

[0093] )ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COMMENT='BW table';

[0094] I will give you a question and a list of news IDs. This question is associated with some news content in the database. I need you to understand the question and query all the news related to the question in this list of news IDs.

[0095] There are some constraints:

[0096] 1. You can only use the tables provided in the table structure information to generate SQL. It is prohibited to fabricate information at will.

[0097] 2. Be careful not to make mistakes in the relationship between tables and fields when generating SQL

[0098] 3. The current date is {current_datatime}. When the date in the question is unclear, such as tomorrow or today, you can infer the correct time based on this time. Note that if there is no time limit, do not consider the time.

[0099] 4. The time format includes date and time, such as 2024 / 3 / 1 12:13:46: So you need to pay attention to querying only when the date is given. For example, the query for October 1, 2024 should be time>='2024-10-01 00:00:00'AND time<'2024-10-02 00:00:00'

[0100] 5. When asking questions about the top topics, ignore this limit. For example, if you want to sort out the top 5 topics, all news will be queried because it only limits the topics, not the news.

[0101] 7. No need to explain too much, just give an executable MySQL query statement

[0102] Now, please analyze the following questions and news list based on the above cases and constraints. Please note that when there is no explicit time limit, please do not consider the current time limit. Ignore the time limit at this time. Also, note that only time has a sorting relationship, and other fields do not have a sorting relationship. Do not give any explanations and directly give the query statement.

[0103] The question is: {query}

[0104] The news list id is: {text_id}.

[0105] Through the above prompt, the large language model can generate a MySQL query statement, notify the execution of the query statement, and retrieve the news document corresponding to the question (news compilation demand) from the MySQL news library.

[0106] In some embodiments, the information summary library can also be in the form of a MySQL database. Referring to the above embodiment, by inputting a prompt for instructing the large language model to generate an SQL query statement, and then retrieving the information summary related to the retrieved information document from the information summary library through the SQL query statement, the large language model is called to perform the information summary retrieval task. The prompt can refer to the above news document retrieval prompt.

[0107] Through the above processing, the information summary related to the information compilation requirement is retrieved from the information summary library through the large language model, and then the information compilation processing is performed based on the retrieved information summary.

[0108] S102: calling the large language model to determine an information compilation outline, and compiling the retrieved information summaries according to the information compilation outline to obtain an information compilation result.

[0109] The above-mentioned information compilation outline refers to a template or outline used when compiling information. The information compilation outline can be a fixed information compilation outline of the information processing system, or an information compilation outline input by the user when inputting information compilation requirements, or an information compilation outline adaptively determined according to the content of the information summary to be compiled.

[0110] In some embodiments, calling the large language model to determine the information compilation outline may be to call the large language model to parse a preset compilation report template to obtain the information compilation outline, or to call the large language model to generate an information compilation outline that is compatible with the retrieved information summary.

[0111] The preset compilation report template may be a fixed compilation report template preset by the information processing system, or a compilation report template input by the user when inputting information compilation requirements into the information processing system.

[0112] If the user does not specify a compilation report template, the system calls the large language model to generate an information compilation outline that is compatible with the retrieved information summary; if the user specifies a compilation report template, the system calls the large language model to parse the user-specified compilation report template to obtain the information compilation outline.

[0113] After determining the information compilation outline, the large language model can be called to compile the retrieved information abstracts according to the information compilation outline to obtain the information compilation result. The information compilation result can be in the form of an information compilation report, and the content in the report conforms to the structure of the above information compilation outline.

[0114] Exemplarily, by inputting an information reorganization prompt into the large language model, the information processing system can call the large language model to reorganize the retrieved information summaries according to the above-mentioned information reorganization outline and output the information reorganization result.

[0115] In some embodiments, it is disclosed that a large language model is called to organize the retrieved information summary according to the information organization outline to obtain an information organization result. Specifically, the large language model can be called to perform the following processing A1-A3 to obtain the information organization result:

[0116] A1. Generate an information compilation title according to the information compilation requirement.

[0117] Specifically, since the characteristic of compilation is to summarize and integrate existing content, the clarity of the title is crucial. The embodiment of the present application generates an information compilation title based on the information compilation demand, which can more accurately reflect the topics or issues that users are concerned about, so that the main theme of the information compilation result is consistent with the user's expectations. At the same time, this method can avoid vague or inaccurate expressions, ensuring that the information compilation results are more targeted and reading-guided when published.

[0118] Exemplarily, by inputting a prompt for instructing the large language model to generate an information compilation title into the large language model, the large language model can be called to generate an information compilation title according to information compilation requirements.

[0119] Taking the above news compilation example, the large language model can be called to generate a news compilation title corresponding to the news compilation requirements through the following prompt:

[0120] I will give you a question. The purpose of this question is to compile some news. What I need you to do is to generate a headline for the compiled message based on this question.

[0121] Be careful not to make the title too complicated; just briefly summarize the message.

[0122] Here are a few examples:

[0123] The question is: sort out the relevant topics on March 1, 2024, and form a reorganized report

[0124] The generated title is: 22024-03-01 Hot Topics Compilation Report

[0125] The problem is: organize related topics into messages

[0126] The generated title is: Compilation report on related hot topics

[0127] Next, please generate relevant titles based on the questions below. No need to explain too much, just output the titles directly.

[0128] The problem is:

[0129] {input}

[0130] In the above prompt, "input" is the news compilation demand, such as "what are the hot events in August, compile them into news reports". By inputting this prompt into the big language model, the big language model can generate the news compilation title.

[0131] A2. According to the outline title in the information compilation outline, information is extracted from the retrieved information abstract to obtain information content corresponding to the outline title.

[0132] Specifically, the information compilation outline includes one or more outline titles, which define the structure of the information compilation outline. For example, the news compilation outline may include outline titles such as summary, event context, and trend analysis.

[0133] The information processing system calls the large language model to extract the specific information content corresponding to each outline title in the information compilation outline, that is, to extract the information content corresponding to each outline title in the information compilation outline from the retrieved information summary.

[0134] Exemplarily, the information processing system inputs a prompt to the large language model to instruct the large language model to extract the information content corresponding to each outline title in the information compilation outline from the retrieved information summary, and the large language model can be used to obtain the specific information content corresponding to each outline title in the information compilation outline.

[0135] Still taking the above news compilation as an example, the information processing system inputs the following prompt to the large language model, which enables the large language model to extract the event context from the news summary:

[0136] I will give you a set of news texts. These news are related and describe different aspects of the same topic.

[0137] Your task is to extract the events described in the news based on the content, and then arrange them in chronological order to form a context of events.

[0138] Note: The event description in the output context should be detailed. Event-related parameters such as time, place, person, action, etc. need to be described in detail, especially time. If it is mentioned, it must be given.

[0139] The sequence of events needs to be arranged in chronological order from small to large. If some events do not have time, then the events that are not mentioned in time should be arranged according to the logic you think is appropriate.

[0140] Now please analyze the following news:

[0141] {input}

[0142] Please extract the events described therein and arrange them in chronological order to form an event context. There is no need for excessive explanation, just directly output the event context results.

[0143] Notice:

[0144] The event description in the output context should be detailed. Event-related parameters such as time, place, person, action, etc. need to be described in detail. In particular, time must be given if it is mentioned, and ignored if it is not mentioned.

[0145] The sequence of events needs to be arranged in chronological order from small to large. If some events do not have time, then the events that are not mentioned in time should be arranged according to the logic you think is appropriate.

[0146] No need for too much explanation, just list the events and output the event context results.

[0147] In the above prompt, "input" is the news summary retrieved from the news summary library. The large language model processes according to the above prompt to extract the event context from the news summary.

[0148] Similarly, for other outline titles, the corresponding prompt can be input into the large language model in the above manner to obtain the information content corresponding to the outline title.

[0149] A3. According to the structure of the information compilation outline, the information compilation titles and the information contents corresponding to each outline title are integrated to obtain an information compilation result.

[0150] Specifically, after obtaining the information content corresponding to each outline title in the information compilation outline, the information compilation title, each outline title, and the information content corresponding to each outline title are integrated according to the structure of the compilation outline to obtain the information compilation result.

[0151] Exemplarily, by inputting into the large language model a prompt for instructing the large language model to integrate the information compilation title, each outline title, and the information content corresponding to each outline title according to the structure of the compilation outline, the large language model can perform the above-mentioned information integration processing and output the information compilation result.

[0152] From the above introduction, it can be seen that the information compilation method proposed in the embodiment of the present application performs the information compilation task with the help of a large language model. In the process of performing the information compilation task, the information document corresponding to the information compilation requirement is first retrieved from the information document library, and the information summary related to the information document is retrieved from the information summary library, and then the information compilation outline is determined, and the retrieved information summary is compiled according to the information compilation outline to obtain the information compilation result. The above-mentioned information compilation method, with the help of the powerful context understanding ability and natural language understanding ability of the large language model, can comprehensively analyze the background, details and core elements in the information summary, so as to ensure that the information compilation result is more logical and coherent, so that the method can more efficiently and accurately generate information compilation results with clear logic and comprehensive content.

[0153] In some other embodiments, calling a pre-trained large language model, retrieving information documents corresponding to information compilation requirements from an information document library, and retrieving information summaries related to the information documents from an information summary library, specifically includes calling a pre-trained large language model to perform the following processing implementations B1-B4:

[0154] B1. Retrieve information documents corresponding to the information compilation requirements from the information document library.

[0155] Exemplarily, by inputting a prompt into the large language model for instructing the large language model to retrieve information documents corresponding to the information compilation requirements from the information document library, the large language model can be called to perform the above-mentioned information document retrieval task to obtain information documents corresponding to the information compilation requirements.

[0156] The specific prompt design method can refer to the prompt design method described in the specific introduction of step S101 in the above embodiment, which will not be repeated here.

[0157] B2. Analyze the information compilation requirements and determine information compilation restriction conditions.

[0158] Specifically, the information compilation requirements input by the user to the information processing system can restrict the user's requirements in many aspects, such as limiting the information compilation conditions in terms of time, subject, region, quantity, etc. Only by accurately understanding the information compilation restriction conditions in the information compilation requirements can the information be compiled more accurately so that the information compilation results meet the user's requirements.

[0159] Therefore, in the embodiment of the present application, when the large language model performs information compilation, the user's information compilation requirements are analyzed and understood, and information compilation restriction conditions are extracted and mined therefrom.

[0160] Exemplarily, by inputting into the large language model a prompt for instructing the large language model to understand the information compilation requirement and extract the information compilation restriction condition therefrom, the large language model can be called to extract the information compilation restriction condition from the information compilation requirement.

[0161] For example, for news compilation needs, by inputting the following prompt into the big language model, the big language model can be called to parse the news compilation needs and determine the news compilation restrictions:

[0162] I will give you a news compilation requirement instruction. This instruction will involve some news / topic information. You need to extract some restrictions on news / topics in this instruction. The restriction parameters are as follows:

[0163] 1. Topic: This part mainly includes some topics restricted in the directive, such as the opening of the Winter Olympics.

[0164] 2. Country / Region area: This refers to the restrictions of some regions or countries in the directive

[0165] 3. Time: This refers to the time limit in the instruction. Please note that a start and end time range needs to be given. For example, if it is August 3, 2024, the format is ['2024-08-03 00:00:00', '2024-08-04 00:00:00']

[0166] 4. Quantity: This refers to the number of news / topics that may be limited in the command. Please list all possible index values ​​in order. The index starts from 0. For example, the output of the top 5 topics should be [0, 1, 2, 3, 4], and the output of the last three topics should be [-3, -2, -1].

[0167] Please note that there is no need for further explanation, and the value is returned directly in json format. If there is no restriction parameter mentioned in the question, it will be "None"

[0168] Here is an example:

[0169] Case 1:

[0170] Question: Report the top 5 topics

[0171] Result: {"topic":"None","area":"None","time":"None","quantity":[0,1,2,3,4]}

[0172] Next, please analyze the following problem and output the json result directly.

[0173] Question: {query}.

[0174] In the above prompt, "query" refers to the news compilation demand to be parsed, such as "What are the hot events in August, compile them into news reports?" By parsing the news compilation demand through a large language model, we can understand and mine the news compilation restrictions contained in the news compilation demand.

[0175] For other information compilation requirements, the above prompt can be written and input into the large language model, so that the large language model can perform the information compilation requirement parsing task according to the prompt and determine the information compilation restriction conditions.

[0176] B3. Based on the information compilation restriction condition, select target information documents that meet the information compilation restriction condition from the information documents corresponding to the information compilation requirement.

[0177] Specifically, usually, the information document library uses a specific data structure to store information documents, such as a MySQL database. Accordingly, when retrieving information documents from the information document library, a search instruction that is compatible with the information document library needs to be used, such as implementing information document retrieval through an SQL search instruction.

[0178] Due to the limitation of the database table structure, the information compilation requirements and the table structure cannot be fully matched, and the information compilation requirements input by the user are often not standardized enough. Directly performing string matching based on information documents will result in a large amount of information loss. Therefore, in the embodiment of the present application, the information documents are first preliminarily and roughly screened from the information document library through step B1 to retrieve more information documents and avoid missing information documents. Then, the information compilation restriction conditions are extracted from the information compilation requirements through step B2, so that the information compilation restriction conditions can be further used in step B3 to accurately screen the information documents retrieved in step B1 to obtain information documents that meet the information compilation restriction conditions. This information document screening method not only improves the accuracy of information document screening and avoids missing information documents, but also can eliminate redundant information documents, reducing the subsequent information compilation workload.

[0179] Exemplarily, by inputting into the large language model a prompt for instructing the large language model to filter out target information documents that meet the information reorganization restrictions from information documents corresponding to the information reorganization requirements based on the information reorganization restrictions, the large language model can filter out information documents that meet the information reorganization restrictions from the information documents retrieved in step B1 to obtain the target information document.

[0180] B4. Retrieve information summaries related to the target information document from the information summary library.

[0181] Exemplarily, by inputting a prompt to the large language model for instructing the large language model to retrieve an information summary related to the target information document from the information summary library, the large language model can be called to perform the above-mentioned information summary retrieval task to obtain an information summary related to the target information document.

[0182] The specific prompt design method can refer to the prompt design method described in the specific introduction of step S101 in the above embodiment, which will not be repeated here.

[0183] In some other embodiments, calling the large language model to generate an information compilation outline that is adapted to the retrieved information summary can be specifically implemented by calling the large language model to perform the following steps C1-C4:

[0184] C1. Cluster the retrieved information summaries to obtain at least one information summary group.

[0185] Specifically, in the process of information compilation, "information summary clustering" is a crucial step. Due to the diversity of information content, different information may involve different topics. Directly processing these contents can easily lead to fragmented and incomplete information. Through cluster analysis, information with similar or related content can be classified into the same group, helping to better identify and summarize key information under different topics. This not only improves the efficiency of information compilation, but also ensures that the compilation results are clearer and more logical in structure, ensuring that key information is fully covered.

[0186] Using a large language model for information clustering has significant advantages over traditional small models. It can not only better understand complex semantic relationships and contexts, but also process large-scale data and improve the accuracy and efficiency of clustering. Large language models can automatically learn features in information, reduce human intervention, and adapt to different types of information, thereby improving the automation level and quality of information compilation.

[0187] Exemplarily, a prompt for instructing the large language model to cluster information summaries is input into the large language model, and at the same time, information summaries to be classified are input into the large language model, so that the large language model can cluster the input information summaries to obtain at least one information summary grouping.

[0188] Taking the clustering of news summaries as an example, inputting the following prompt into the large language model can enable the large language model to cluster the input news summaries and obtain at least one news summary grouping:

[0189] I will provide you with multiple groups of news summaries. Each group of news summaries contains the event ID, title and summary. The ID is the ID of the news summary and must be consistent. Now you need to merge and group these news summaries.

[0190] First, a definition:

[0191] News Summary: It focuses on a specific issue, event, situation, behavior or concept. News summary usually revolves around a central point, with a specific time, place, person or event.

[0192] A news summary is different from a topic. The event itself, the background of the event, the cause of the event, the consequences of the event, external evaluations, and subsequent developments of the event are all independent news summaries.

[0193] There are several requirements for merging groups:

[0194] 1. The granularity of the merged grouping does not need to be too fine. The merged news summaries all belong to the same theme. The event itself, the background of the event, the cause of the event, the consequences of the event, external evaluation, the subsequent development of the event, and the reflection caused by the event all belong to the same theme and can be merged into one group.

[0195] 2. You need to provide the title of the news summary after the merger and the original news summary corresponding to the topic. The merged news summary needs to fully describe all the news summaries under the merged group, and the titles of the original news summaries listed must be consistent

[0196] 3. Be careful not to miss the news summary. All the provided event sub-topics must be grouped. The clustering result may be only one group or multiple groups. All the results need to be listed.

[0197] 4. Note that news summary clustering should be analyzed based on the news summary title and summary content. Note that the news summary titles listed and the ID numbers in front of the titles should be consistent.

[0198] Note that the clustering categories should be detailed and not explained too much. All clustering groups are listed and output, and the format of each clustered group is output in the following format:

[0199] 'Cluster 1: Title after clustering

[0200] Event ID: the news summary ID of this cluster; Title: the news summary title of this cluster

[0201] Event ID: the news summary ID of this cluster; Title: the news summary title of this cluster

[0202] Cluster 2: Titles after clustering

[0203] Event ID: the news summary ID of this cluster; Title: the news summary title of this cluster

[0204] Event ID: the news summary ID of this cluster; Title: the news summary title of this cluster

[0205] ...'

[0206] Please analyze the following news summary and cluster the news summary.

[0207] {input}

[0208] In the above prompt, "input" is the news summary to be clustered. By inputting the above prompt into the large language model, the large language model can cluster and group the input news summary and divide the input news summary into at least one news summary group.

[0209] According to the prompt example above, a response prompt may be generated for the retrieved information summary and the prompt may be input into the large language model so that the large language model clusters the input information summary to obtain at least one information summary group.

[0210] C2. extracting key concepts and analyzing the relevance of key concepts from the obtained information summary groups, and integrating the content of the information summary based on the relevance between the key concepts.

[0211] Specifically, the core task of extracting key concepts is to find the most representative and informative elements in the information summary, help focus on the core content of the information summary, and ensure that the key points of the compilation results are clear. Using the natural language processing capabilities of the large language model, the system can not only identify keywords in the literal sense, but also understand the semantic relationship behind them. For example, in a news article about global climate change, key concepts that may be extracted include "carbon emissions", "climate agreement", "extreme weather", etc. These concepts will form the basis of the subsequent compilation outline to ensure that the compilation results can comprehensively cover the core themes and important information of the news and avoid missing key content.

[0212] By inputting a prompt for instructing the large language model to extract key concepts from information summaries, the large language model can be used to perform the task of extracting key concepts from information summaries in each information summary group.

[0213] Taking the extraction of key concepts from news summaries as an example, the following prompt is input to the large language model, so that the large language model is called to extract key concepts from the news summaries:

[0214] I will provide a news article. Please extract the key concepts from it. Key concepts are words or phrases that have important meanings in the event, including:

[0215] 1. **Main actors**: individuals, organizations or countries involved in the incident.

[0216] 2. **Key Incidents**: The major events or actions that occurred during an incident, especially those that directly affected the outcome.

[0217] 3. **Cause**: Factors that led to the event, including economic, social or political reasons.

[0218] 4. **Impact**: The impact or consequences of the event on relevant parties (individuals, enterprises, society, country, etc.).

[0219] 5. **Response measures**: Measures or strategies taken to respond to incidents.

[0220] Please provide the following information for each extracted key concept:

[0221] - **Concept Name**: Short concept name.

[0222] - **Description**: A detailed description of the concept, including its role and importance in the event.

[0223] Here is the news text:

[0224] [Insert news text here]

[0225] Start extracting key concepts.

[0226] In the above prompt, the news text is the news summary of the key concepts to be extracted. The above prompt is input into the large language model, so that the large language model can extract the key concepts of the news summary in the prompt.

[0227] According to the above method, for the information summaries in the information summary group obtained in step C1, the above prompt can be designed to enable the large language model to extract key concepts from the information summaries.

[0228] After extracting key concepts from the information summary, the next step is to conduct a correlation analysis on the extracted key concepts to analyze the correlation between these concepts. Through the semantic analysis function of the large language model, the inherent relationship between different key points in the key concepts can be identified, and related key points can be combined to form a larger topic group. For example, in the above-mentioned news on climate change, "carbon emissions" may be closely related to "climate agreements", while "extreme weather" may be related to the direct consequences of climate change. Through correlation analysis, information isolation and fragmentation can be avoided, related content can be organically integrated, and the overall coherence and logical fluency of the compilation results can be improved.

[0229] By inputting a prompt for instructing the large language model to perform relevance analysis on key concepts into the large language model, the task of performing relevance analysis on key concepts can be performed with the help of the large language model.

[0230] Taking the key concept relevance analysis of news summaries as an example, the following prompt can be input into the large language model, thereby invoking the large language model to perform relevance analysis on the key concepts extracted from the news summaries:

[0231] I will provide a list of key concepts (including concept names and brief descriptions). Please conduct a correlation analysis on these key concepts, identify and integrate the relationships between them, and form new topics. The specific requirements are as follows:

[0232] 1. **Input content**:

[0233] [Key concept list, extracted results]:

[0234] 2. Relationship Identification: Analyze the relationships between key concepts, focusing on the following aspects:

[0235] - **Causation**: Which concepts cause other concepts?

[0236] -**Influence relationships**: Which concepts influence each other? Describe the direction and degree of influence.

[0237] -**Similarities and Contrast**: Are there concepts that are similar or contrasting in some way?

[0238] 3. **Theme integration**: Based on the above analysis, identify concepts that can be integrated and generate new themes. For example, combine "main factors" and "country response policies" to form "causes and responses to economic recession."

[0239] 4. **Final output**: List the generated new topics and their descriptions, ensuring their logic and relevance.

[0240] Please start the association analysis and generate new themes.

[0241] In the above prompt, the "key concept list" is a list of key concepts extracted from the news summary. By inputting the above prompt into the large language model, the large language model can perform a correlation analysis on the key concepts in the key concept list, and based on the analysis results, merge the related key concepts to obtain a new concept theme.

[0242] Similarly, for information summaries in any scenario, the above method can be used to perform correlation analysis on the key concepts extracted from the information summary with the help of a large language model, and based on the correlation analysis results, the related key concepts can be merged to obtain new concepts, thereby achieving the purpose of content integration of the information summary based on the correlation relationship between the key concepts.

[0243] After the above processing, the content of the summary information is more prominent and the information logic is clearer, which is conducive to the reorganization of the information content in the summary information.

[0244] C3. Sort the content display order of the information summary after content integration to generate an information outline.

[0245] Specifically, after clarifying the relationship between the key concepts in the information summary, the next step is to determine the order of display of each part of the information through "priority sorting". The core of this step is to display the most important information first according to the main theme of the information summary and the needs of the readers, and arrange the secondary content appropriately at the back. Priority sorting not only ensures the logical clarity of the compilation results, but also helps readers quickly grasp the key information and gradually understand the main content and secondary details of the information.

[0246] For example, if the focus of the news is on the progress of the climate agreement, then the content related to the "climate agreement" should be displayed first, because this is the most influential core point. Secondly, other topics related to the climate agreement can be displayed, such as the potential impact of "extreme weather" on the implementation of the agreement. This arrangement can naturally guide readers to think about the overall context of the problem.

[0247] This prioritization process makes the compilation results more organized in structure, improves logic and reading experience, and ensures that readers start from the most important information and gradually understand the broader context and details behind the information.

[0248] By inputting a prompt into the large language model for instructing the large language model to sort the content display order of the information summary and generate an information outline, the large language model can be used to sort the content display order of the information summary and generate an information outline.

[0249] Taking the priority sorting of key concepts in news summaries as an example, the following prompt can be input into the large language model, so that the large language model is called to prioritize the key concepts extracted from the news summaries and generate an information outline based on the sorting results:

[0250] I will provide a list of key concepts after extraction. Please sort these concepts according to priority to help generate a logical and clear information outline. The sorting should follow the following specifications:

[0251] 1. **Core Points First**: Sort the most critical and core concepts first. These are the things that readers need to know most. Usually include information that directly answers "why" and "main reasons" of the event.

[0252] 2. **Cause and effect first**: If there is a causal relationship between certain concepts, list the "cause" part first, and then the "result" or "impact" part to ensure a clear logical order.

[0253] 3. **Reader attention priority**: Consider the needs and concerns of readers, and give priority to displaying content that readers care about most, such as the direct impact of the event on individuals, companies or society.

[0254] 4. **Chronological order**: If the events have a chronological order, arrange them in the order in which they occurred to ensure narrative coherence.

[0255] 5. **Background information is secondary**: Background information helps readers understand the context of the event, but it is usually not the most important and should be placed after the main information.

[0256] 6. **Future Outlook or Conclusion**: Finally, place content related to future predictions, outlooks, or conclusions to help readers think about the future development of events.

[0257] Based on the above sorting criteria, please prioritize the following key concepts and generate a logically clear information outline.

[0258] List of key concepts:

[0259] [Key concept results after correlation analysis]

[0260] Please prioritize these concepts and generate a detailed information outline.

[0261] In the above prompt, "key concept results after correlation analysis" are the key concepts obtained after correlation analysis and merging in step C2. By inputting the above prompt into the large language model, the large language model can sort these concepts and generate an information outline.

[0262] In the above-mentioned priority sorting process, prioritizing key concepts is equivalent to prioritizing the summary contents corresponding to the key concepts. Therefore, based on the above-mentioned priority sorting results, the display order of each part of the summary contents in the information summary in the information outline can be determined.

[0263] According to the above method, for each information summary group, the corresponding information outline can be obtained through the processing of steps C2 and C3.

[0264] C4. Performing structured integration processing on the information outline to obtain an information compilation outline.

[0265] Specifically, structural integration of information outlines means merging the independent information outlines corresponding to each information summary group according to the logical order and importance of their content, ensuring that information of different topics and categories is presented in an orderly manner. This integration method can not only cover the core points of each information summary group, but also make the compilation results clear and thematic through the overall structured layout, providing readers with a coherent and in-depth analysis framework. The final information compilation outline will help to fully and accurately convey the information content and the logical relationship behind it.

[0266] By inputting a prompt into the large language model for instructing the large language model to perform structural integration on the information outlines corresponding to the information summary groups, the large language model can be used to perform structural integration on the information outlines corresponding to each information summary group to obtain an information compilation outline.

[0267] Taking the structural integration of the information outlines corresponding to the news summary groups as an example, the following prompt can be input into the large language model, so as to call the large language model to perform structural integration on the information outlines corresponding to each news summary group and obtain an integrated news compilation outline:

[0268] Task description: Generate a complete structured news compilation outline based on several sets of input news outlines and the user's compilation instructions. Ensure that the integration process follows a logical order and reflects the core points and importance of the news. The outline should contain a clear hierarchical structure, and the level should be marked before each title, such as [Title 1], [Title 2], etc., to facilitate subsequent analysis.

[0269] News outline input:

[0270] [List of news outlines corresponding to each news summary group]

[0271] Generate a complete structured news outline covering all the key points of the news outline.

[0272] The outline should be in a logical order and highlight the core information.

[0273] Example of outline levels: [Heading 1], [Heading 2], and so on.

[0274] The outline should contain the following structures or parts (adaptable according to instructions):

[0275] The following is an example of the output:

[0276] [Title 1] Introduction

[0277] [Title 1] Overview of core events

[0278] [Title 1] Economic Event Analysis

[0279] [Title 2] Economic growth slows down

[0280] [Title 3] Analysis of the reasons for the slowdown

[0281] [Title 3] Analysis of the consequences of emergency slowdown ...

[0283] In the above prompt, "a list of news outlines corresponding to each news summary group" is the news outline corresponding to each news summary group obtained in step C3. By inputting the above prompt into the large language model, the large language model can perform structured integration processing on these news outlines to generate a news compilation outline.

[0284] Through the above method, a large language model can be used to adaptively generate an information compilation outline that is compatible with the retrieved information summary.

[0285] In some embodiments, in the process of calling the large language model to generate an information compilation outline that is adapted to the retrieved information summary, the large language model is also called to perform topic classification on the retrieved information summary, and based on the topic classification results, a preset information compilation outline corresponding to the topic classification results is obtained from the outline library.

[0286] Among them, the above-mentioned outline library is a database for storing information compilation outlines preset by users for information on different topics. In an embodiment of the present application, the user can pre-design the information compilation outlines corresponding to the information on different topics according to the characteristics of the information on different topics, wherein the information compilation outlines corresponding to the information on different topics may have the same parts or different parts. For example, under the economic topic, the user can set the "Similar Cases" outline title in the corresponding information compilation outline, while under other types of topics, the user can not set the "Similar Cases" outline title in the corresponding information compilation outline. Therefore, in the above-mentioned outline library, the preset information compilation outlines corresponding to different information topics are stored.

[0287] Based on the outline library, after the large language model retrieves the information summary from the information summary library, it further performs subject classification on the retrieved information summary to determine the subject of the retrieved information summary.

[0288] Exemplarily, a prompt for instructing the large language model to classify the information abstract into topics is input into the large language model, so that the large language model can classify the retrieved information abstract into topics, thereby determining the topic classification result. The large language model classifies the information abstract into topics, which can be to classify the topic of the information abstract into a topic among preset topics.

[0289] For example, you can input the following prompt into the large language model to enable the large language model to perform the information summary topic classification task:

[0290] Based on the information abstract you entered, identify the main topics covered in the information abstract and categorize them into one of the following predefined topics. If the information abstract covers multiple topics, please sort them in order of primary and secondary and label the secondary topics. The following are possible topic categories: {type}

[0291] The information summary you need to analyze is: {input}. Please directly input the corresponding topic. Please note that you only need to give the most appropriate topic.

[0292] In the above prompt, "type" is a plurality of pre-set topics, and "input" is the information summary to be classified. The above prompt is input into the large language model, so that the large language model can classify the topic of the information summary in the prompt into one of the preset topics, and obtain the topic classification result.

[0293] The above-mentioned processing of subject classification of the retrieved information summaries can be performed at any stage after the information summaries are retrieved. For example, the subject classification processing can be performed after the retrieved information summaries are clustered, or it can be performed after the large language model generates an information compilation outline. The embodiments of the present application are not limited to this.

[0294] After obtaining the subject classification result through the subject classification processing described above, the preset information compilation outline corresponding to the subject classification result is obtained from the outline library according to the subject classification result, that is, the preset information compilation outline corresponding to the subject classification result is retrieved from the outline library. The retrieval of the preset information compilation outline can be performed by a large language model or by an information processing system.

[0295] After retrieving the preset information compilation outline corresponding to the above-mentioned subject classification result from the outline library, the large language model is called to compare the generated information compilation outline with the above-mentioned preset information compilation outline, and the generated information compilation outline is supplemented with missing content.

[0296] The information compilation outline generated by the large language model is compared with the above-mentioned preset information compilation outline, mainly to compare whether the outline content in the above-mentioned preset information compilation outline is recorded in the information compilation outline generated by the large language model, that is, to compare and confirm whether the information compilation outline generated by the large language model omits the outline content in the above-mentioned preset information compilation outline.

[0297] If the information compilation outline generated by the large language model omits the outline content in the above-mentioned preset information compilation outline, the generated information compilation outline is supplemented with the missing content, so that the information compilation outline generated by the large language model can meet the user's preset information compilation outline requirements, and thus the final information compilation result meets the user's needs.

[0298] Exemplarily, by inputting into the large language model a prompt for instructing the large language model to compare the generated information compilation outline with the above-mentioned preset information compilation outline and to supplement the missing content of the generated information compilation outline, the large language model can perform the task of supplementing the generated information compilation outline.

[0299] For example, the above prompt can take the following form:

[0300] I will provide a generated outline and an artificial expected outline content. Please compare the two outlines and identify the missing content. The specific requirements are as follows:

[0301] 1. **Generated outline**:

[0302] -[insert generated outline here]

[0303] 2. **Outline of artificial expectations**:

[0304] -[insert outline of artificial expectations here]

[0305] 3. **Comparative Analysis**:

[0306] -Please compare the generated outline with the manually expected outline item by item and list the missing points or content.

[0307] 4. **Supplementary omissions**:

[0308] -Add the identified omissions to the generated outline to ensure that the new outline covers all the expected points.

[0309] 5. **Generate a new outline**:

[0310] -Regenerate a new outline that is logically clear and includes all the missing content.

[0311] Please start a comparative analysis and generate a new supplementary outline.

[0312] By adding the generated information compilation outline and the preset information compilation outline retrieved from the outline library to the above prompt, the large language model can supplement the generated information compilation outline based on the retrieved preset information compilation outline to obtain an information compilation outline that meets user requirements.

[0313] Finally, the information summary is compiled based on the information compilation outline after the missing content is supplemented, and the information compilation result that meets the user's expectations can be obtained.

[0314] Based on the above-mentioned embodiments, it can be seen that the information compilation method proposed in this application realizes automatic information compilation by calling a large language model through an information processing system, and the method can adaptively generate an information compilation outline through a large language model. Figure 2 A specific processing process of the information compilation method embodied in the above embodiments is shown. In the information compilation method proposed in this application, the entire information compilation process is more intelligent and accurate, and an information compilation result that meets the user's expectations can be obtained.

[0315] In other embodiments, after the large language model retrieves information documents corresponding to the information compilation requirements from the information document library and retrieves information summaries related to the information documents from the information summary library, the retrieved information summaries are first clustered to obtain at least one information summary group. For example, the retrieved information summaries can be clustered as described in step C1 of the above embodiment.

[0316] Then, the large language model integrates the information summaries in each information summary group obtained by clustering, such as summarizing, deduplicating, and correcting the information summaries.

[0317] Due to the diversity of information summaries, different information summaries may involve different topics, viewpoints or perspectives. Directly processing these information summaries can easily lead to information fragmentation and incompleteness. By clustering and integrating information summaries, similar or related information summaries can be classified into the same group, and the information summaries in the same group can be integrated, so that the information summaries under the same information summary group can be more complete and accurate.

[0318] Finally, the large language model is called to determine the information compilation outline, and the information in each of the information summary groups is compiled according to the information compilation outline to obtain an information compilation result. In which, calling the large language model to determine the information compilation outline can be calling the large language model to generate an information compilation outline that is compatible with the retrieved information summary, or calling the large language model to parse a preset compilation report template to obtain the information compilation outline. For the specific processing process, refer to the corresponding processing process in the above embodiment.

[0319] After determining the information compilation outline, the large language model compiles the information summaries in each information summary group after information integration according to the information compilation outline to obtain the information compilation result. The specific compilation process can refer to the compilation process introduced in the above embodiment.

[0320] In this embodiment, before the information summaries are compiled, they are first clustered and information integrated, which can improve the compilation efficiency and ensure that the compilation results are clearer, more complete and more accurate.

[0321] Corresponding to the above-mentioned information compilation method, the embodiment of the present application also provides an information compilation device, see Figure 3 As shown, the device comprises:

[0322] The information retrieval unit 100 is used to call the pre-trained large language model, retrieve the information document corresponding to the information compilation requirement from the information document library, and retrieve the information summary related to the information document from the information summary library, wherein the information summary library is used to store the information summary extracted from the information document;

[0323] The information compilation unit 110 is used to call the large language model, determine an information compilation outline, and compile the retrieved information summaries according to the information compilation outline to obtain an information compilation result.

[0324] In some implementations, the information compilation unit 110 calls the large language model to determine the information compilation outline, including:

[0325] The large language model is called to generate an information compilation outline that is adapted to the retrieved information summary, or the large language model is called to parse a preset compilation report template to obtain the information compilation outline.

[0326] In some implementations, the information compilation unit 110 calls the large language model to generate an information compilation outline adapted to the retrieved information summary, including:

[0327] The large language model is called to perform the following processing:

[0328] Clustering the retrieved information summaries to obtain at least one information summary group;

[0329] Extract key concepts and analyze the relevance of key concepts from the obtained information summary groups, and integrate the content of the information summary based on the relevance between the key concepts;

[0330] Sort the content display order of the information summary after content integration to generate an information outline;

[0331] The information outline is structured and integrated to obtain an information compilation outline.

[0332] In some implementations, the information compilation unit 110 calls the large language model to generate an information compilation outline adapted to the retrieved information summary, and further includes:

[0333] The large language model is called to subject-classify the retrieved information abstracts, and according to the subject-classification results, a preset information compilation outline corresponding to the subject-classification results is obtained from an outline library;

[0334] Furthermore, the large language model is called to compare the generated information compilation outline with the preset information compilation outline, and the generated information compilation outline is supplemented with missing content; the missing content includes outline content recorded in the preset information compilation outline and not recorded in the generated information compilation outline.

[0335] In some implementations, the information compilation unit 110 calls the large language model and compiles the retrieved information summaries according to the information compilation outline to obtain information compilation results, including:

[0336] The large language model is called to perform the following processing:

[0337] Generate an information compilation title according to the information compilation requirement;

[0338] According to the outline title in the information compilation outline, extract information from the retrieved information summary to obtain information content corresponding to the outline title;

[0339] According to the structure of the information compilation outline, the information compilation titles and the information contents corresponding to each outline title are integrated to obtain an information compilation result.

[0340] In some implementations, the information retrieval unit 100 calls a pre-trained large language model, retrieves information documents corresponding to the information compilation requirements from an information document library, and retrieves information summaries related to the information documents from an information summary library, including:

[0341] Call the pre-trained large language model to perform the following processing:

[0342] Retrieve information documents corresponding to the information compilation requirements from the information document library;

[0343] Analyze the information compilation requirements and determine information compilation restriction conditions;

[0344] Based on the information compilation restriction condition, selecting target information documents that meet the information compilation restriction condition from information documents corresponding to the information compilation requirement;

[0345] An information summary related to the target information document is retrieved from an information summary library.

[0346] In some implementations, the apparatus further includes:

[0347] An information processing unit, configured to cluster the retrieved information summaries to obtain at least one information summary group; and for each information summary group, perform information integration on the information summaries therein;

[0348] The information compilation unit 110 compiles the retrieved information summaries according to the information compilation outline to obtain information compilation results, including compiling the information in each of the information summary groups according to the information compilation outline to obtain information compilation results.

[0349] The information compilation device provided in this embodiment belongs to the same application concept as the information compilation method provided in the above embodiments of this application, and can execute the information compilation method provided in any of the above embodiments of this application, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not fully described in this embodiment, please refer to the specific processing content of the information compilation method provided in the above embodiments of this application, and will not be repeated here.

[0350] The functions implemented by the above information retrieval unit 100 and information compilation unit 110 may be implemented by the same or different processors, respectively, and the embodiments of the present application are not limited thereto.

[0351] It should be understood that the units in the above devices can be implemented in the form of a processor calling software. For example, the device includes a processor, the processor is connected to a memory, and instructions are stored in the memory. The processor calls the instructions stored in the memory to implement any of the above methods or realize the functions of each unit of the device, wherein the processor can be a general-purpose processor, such as a CPU or a microprocessor, etc., and the memory can be a memory in the device or a memory outside the device. Alternatively, the units in the device can be implemented in the form of hardware circuits, and the functions of some or all units can be realized by designing the hardware circuits. The hardware circuit can be understood as one or more processors; for example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above units are realized by designing the logical relationship of the components in the circuit; for another example, in another implementation, the hardware circuit can be implemented by PLD, taking FPGA as an example, which can include a large number of logic gate circuits, and the connection relationship between the logic gate circuits is configured by the configuration file, so as to realize the functions of some or all of the above units. All units of the above devices can be implemented in the form of a processor calling software, or in the form of hardware circuits, or in part by a processor calling software, and the remaining part is implemented in the form of hardware circuits.

[0352] In an embodiment of the present application, a processor is a circuit with the ability to process signals. In one implementation, the processor may be a circuit with the ability to read and run instructions, such as a CPU, a microprocessor, a GPU, or a DSP; in another implementation, the processor may implement certain functions through the logical relationship of a hardware circuit, and the logical relationship of the hardware circuit is fixed or reconfigurable, such as a hardware circuit implemented by an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document to implement the hardware circuit configuration can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, DPU, etc.

[0353] It can be seen that each unit in the above device can be one or more processors (or processing circuits) configured to implement the above method, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0354] In addition, all or part of the units in the above device can be integrated together, or can be implemented independently. In one implementation, these units are integrated together and implemented in the form of a SOC. The SOC may include at least one processor for implementing any of the above methods or implementing the functions of each unit of the device. The type of the at least one processor may be different, for example, including a CPU and an FPGA, a CPU and an artificial intelligence processor, a CPU and a GPU, etc.

[0355] Another embodiment of the present application also provides an information compilation device, see Figure 4 As shown, the device includes:

[0356] Memory 200 and processor 210;

[0357] The memory 200 is connected to the processor 210 and is used to store programs;

[0358] The processor 210 is used to implement the information compilation method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0359] Specifically, the above-mentioned information reorganization device may further include: a bus, a communication interface 220 , an input device 230 and an output device 240 .

[0360] The processor 210, the memory 200, the communication interface 220, the input device 230 and the output device 240 are connected to each other via a bus.

[0361] A bus may include a pathway that transfers information between components of a computer system.

[0362] The processor 210 may be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of the program of the scheme of the present invention. It may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0363] The processor 210 may include a main processor, and may also include a baseband chip, a modem, and the like.

[0364] The memory 200 stores a program for executing the technical solution of the present invention, and may also store an operating system and other key services. Specifically, the program may include a program code, and the program code includes a computer operation instruction. More specifically, the memory 200 may include a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM), other types of dynamic storage devices that can store information and instructions, a disk storage, a flash, and the like.

[0365] The input device 230 may include a device for receiving data and information input by a user, such as a keyboard, a mouse, a camera, a scanner, a light pen, a voice input device, a touch screen, a pedometer, or a gravity sensor.

[0366] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0367] The communication interface 220 may include any transceiver or the like to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0368] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement each step of any information compilation method provided in the above embodiments of the present application.

[0369] An embodiment of the present application also proposes a chip, which includes a processor and a data interface. The processor reads and runs a program stored in a memory through the data interface to execute the information compilation method introduced in any of the above embodiments. The specific processing process and its beneficial effects can be found in the introduction to the embodiment of the above information compilation method.

[0370] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the information compilation method described in any of the above embodiments of this specification.

[0371] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0372] In addition, an embodiment of the present application may also be a storage medium on which a computer program is stored, and the computer program is executed by a processor to execute the steps of the information compilation method described in any of the above embodiments of this specification.

[0373] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0374] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0375] The steps in the methods of each embodiment of the present application can be adjusted in sequence, combined and deleted according to actual needs, and the technical features recorded in each embodiment can be replaced or combined.

[0376] The modules and sub-modules in the devices and terminals of the various embodiments of the present application can be combined, divided and deleted according to actual needs.

[0377] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the terminal embodiments described above are only schematic, for example, the division of modules or submodules is only a logical function division, and there may be other division methods in actual implementation, for example, multiple submodules or modules can be combined or integrated into another module, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0378] The modules or submodules described as separate components may or may not be physically separated, and the components of the modules or submodules may or may not be physical modules or submodules, that is, they may be located in one place, or they may be distributed on multiple network modules or submodules. Some or all of the modules or submodules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0379] In addition, each functional module or submodule in each embodiment of the present application may be integrated into one processing module, or each module or submodule may exist physically separately, or two or more modules or submodules may be integrated into one module. The above-mentioned integrated modules or submodules may be implemented in the form of hardware or in the form of software functional modules or submodules.

[0380] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0381] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly by hardware, software units executed by a processor, or a combination of the two. The software units may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0382] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0383] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information compilation method, characterized in that: include: Calling a pre-trained large language model, retrieving an information document corresponding to the information compilation requirement from an information document library, and retrieving an information summary related to the information document from an information summary library, wherein the information summary library is used to store summaries extracted from the information document; The large language model is called to determine an information compilation outline, and the retrieved information abstracts are compiled according to the information compilation outline to obtain an information compilation result.

2. The method according to claim 1, characterized in that The calling of the large language model to determine the information compilation outline includes: The large language model is called to generate an information compilation outline adapted to the information summary, or the large language model is called to parse a preset compilation report template to obtain the information compilation outline.

3. The method according to claim 2, characterized in that The calling of the large language model to generate an information compilation outline adapted to the information summary includes: The large language model is called to perform the following processing: Clustering the retrieved information summaries to obtain at least one information summary group; Extract key concepts and analyze the relevance of key concepts from the obtained information summary groups, and integrate the content of the information summary based on the relevance between the key concepts; Sort the content display order of the information summary after content integration to generate an information outline; The information outline is structured and integrated to obtain an information compilation outline.

4. The method according to claim 3, characterized in that The calling of the large language model to generate an information compilation outline adapted to the information summary also includes: Calling the large language model to subject-classify the retrieved information abstracts, and acquiring a preset information compilation outline corresponding to the subject-classification result from an outline library according to the subject-classification result; Furthermore, the large language model is called to compare the generated information compilation outline with the preset information compilation outline, and the generated information compilation outline is supplemented with missing content; the missing content includes outline content recorded in the preset information compilation outline and not recorded in the generated information compilation outline.

5. The method according to any one of claims 1 to 4, characterized in that: The large language model is called to organize the retrieved information summary according to the information organization outline to obtain an information organization result, including: The large language model is called to perform the following processing: Generate an information compilation title according to the information compilation requirement; According to the outline title in the information compilation outline, extract information from the retrieved information summary to obtain information content corresponding to the outline title; According to the structure of the information compilation outline, the information compilation titles and the information contents corresponding to the outline titles are integrated to obtain an information compilation result.

6. The method according to any one of claims 1 to 4, characterized in that The calling of the pre-trained large language model, retrieving an information document corresponding to the information compilation requirement from the information document library, and retrieving an information summary related to the information document from the information summary library, includes: Call the pre-trained large language model to perform the following processing: Retrieve information documents corresponding to the information compilation requirements from the information document library; Analyze the information compilation requirements and determine information compilation restriction conditions; Based on the information compilation restriction condition, selecting target information documents that meet the information compilation restriction condition from information documents corresponding to the information compilation requirement; An information summary related to the target information document is retrieved from an information summary library.

7. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Clustering the retrieved information summaries to obtain at least one information summary group; For each of the information summary groups, integrating the information summaries therein; The step of organizing the retrieved information summaries according to the information organization outline to obtain information organization results includes organizing the information in each information summary group according to the information organization outline to obtain information organization results.

8. An information compilation device, characterized in that: include: An information retrieval unit, used to call a pre-trained large language model, retrieve information documents corresponding to information compilation requirements from an information document library, and retrieve information summaries related to the information documents from an information summary library, wherein the information summary library is used to store summaries extracted from information documents; The information compilation unit is used to call the large language model, determine the information compilation outline, and compile the retrieved information abstracts according to the information compilation outline to obtain the information compilation result.

9. An information compilation device, characterized in that: include: Memory and processor; The memory is connected to the processor and is used to store programs; The processor is used to implement the information compilation method according to any one of claims 1 to 7 by running the program in the memory.

10. A computer program product, characterized in that The method comprises computer program instructions, which, when executed by a processor, enable the processor to execute the information compilation method according to any one of claims 1 to 7.

Images