Literature review generation method and related device
By generating a literature review outline and classifying literature based on the outline title, combined with large language model analysis, the problems of low efficiency and credibility in the generation of large-scale literature reviews were solved, and efficient and high-quality literature review generation was achieved.
Patent Information
- Application Number
- CN202510455063.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-11
- Publication Date
- 2025-07-18
AI Technical Summary
The existing literature review generation methods are inefficient in processing large-scale literature, incomplete content coverage, and lack reliable traceability, resulting in insufficient review quality and credibility.
By obtaining the titles of multiple documents, we generate a literature review outline, including the general title and outline title, classify the literature based on the outline title, and generate a review text, ensuring that the review text corresponding to each outline title contains the citation marks of the original literature, and use a large language model for in-depth analysis and summary.
It realizes efficient and high-quality review generation based on large-scale literature, ensures comprehensive and reliable traceability of content, and improves the readability and logical consistency of the review.
Smart Images
Figure CN120336520A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of natural language processing, and in particular to a method for generating a literature review and related devices. Background Art
[0002] A literature review is a style of writing that is different from research literature. It is written by a researcher after reading the literature on a certain topic in advance, understanding, organizing, integrating, comprehensively analyzing and evaluating, and summarizing the basic facts, basic viewpoints and methods of several articles on the same topic.
[0003] Existing literature review generation methods are generally divided into literature review generation methods based on manual writing and literature review generation methods based on partial automation generation. Literature review generation methods based on manual writing consume a lot of manpower and time, and can usually only process a small number of documents, and the quality of the review depends on the experience and ability of experts. Literature review generation methods based on partial automation generation mainly rely on the summary generation of document content, and generate simple reviews by extracting keywords and selecting sentences from the document content. However, when the number of documents is large, existing literature review generation methods based on partial automation generation also have the problems of low generation efficiency and incomplete content coverage, which cannot meet the needs of modern information processing. In addition, the above two existing literature review generation methods lack reliable traceability. The generated review content is difficult to verify, and users cannot quickly locate the original documents cited in the review, thereby reducing the credibility of the review.
[0004] Therefore, how to provide a literature review generation method to achieve literature review generation based on large-scale literature while ensuring generation efficiency, generation quality and reliable traceability has become a technical problem that needs to be urgently solved by technical personnel in this field. Summary of the invention
[0005] In view of the above problems, this application provides a method and related device for generating a literature review, so as to realize the generation of a literature review based on large-scale literature and ensure the generation efficiency, generation quality and reliable traceability. The specific scheme is as follows:
[0006] The first aspect of the present application provides a method for generating a literature review, comprising:
[0007] Get multiple documents;
[0008] Using the titles of the various documents, a literature review outline is generated, wherein the literature review outline includes a general title and an outline title;
[0009] Classifying each document based on the outline title, each outline title corresponds to a document set, and each document set includes at least one document;
[0010] For each outline title, generate a review text corresponding to the outline title based on the collection of documents corresponding to the outline title;
[0011] Generate a final literature review based on the general title, the outline titles, and the review texts corresponding to each of the outline titles. In the final literature review, the review texts corresponding to each outline title include citation identifiers for the corresponding original documents.
[0012] In one possible implementation, the generating a literature review outline using the titles of each document includes:
[0013] Obtain an outline generation prompt template, which includes outline generation task description information and a document title filling slot;
[0014] Fill the titles of the documents into the document title filling slot to obtain an outline generation prompt;
[0015] Input the outline generation prompt into an outline generation model to obtain the literature review outline, where the outline generation model is a large language model trained to have the ability to generate a literature review outline.
[0016] In one possible implementation, the classifying each document based on the outline title includes:
[0017] For each document, determine the matching judgment result between the document and each of the outline titles by performing a matching judgment between the document and each of the outline titles;
[0018] Classify each document based on the matching judgment results between each document and each outline title.
[0019] In one possible implementation, before the determining the matching judgment result between the document and each of the outline titles by performing a matching judgment between the document and each of the outline titles, the method further includes:
[0020] Summarize each document to obtain a summary of each document;
[0021] Then the determining the matching judgment result between the document and each of the outline titles by performing a matching judgment between the document and each of the outline titles includes:
[0022] For each document, determine the matching judgment result between the summary of the document and each of the outline titles by performing a matching judgment between the summary of the document and each of the outline titles.
[0023] In one possible implementation, the summarizing each document to obtain a summary of each document includes:
[0024] Asynchronously call the single-document summary model to summarize each document, and obtain the summaries of each document. The single-document summary model is a large language model trained to have the ability to summarize a single document.
[0025] In a possible implementation, for each document, call the single-document summary model to summarize the document, and obtain the summary of the document, including:
[0026] Obtain the single-document summary prompt template, which contains single-document summary task description information and a document filling slot;
[0027] Fill the document into the document filling slot to obtain a single-document summary prompt;
[0028] Input the single-document summary prompt into the single-document summary model to obtain the summary of the document generated by the single-document summary model.
[0029] In a possible implementation, each of the outline titles corresponds to a document set. Each document set includes information of at least one document. The information of each document includes the title and summary of the document. Then, generating the review text corresponding to the outline title based on the document set corresponding to the outline title includes:
[0030] Obtain the multi-document summary prompt template, which contains multi-document summary task description information, multi-document summary examples, and a multi-document information filling slot;
[0031] Fill the information of each document in the document set corresponding to the outline title into the multi-document information filling slot to obtain a multi-document summary prompt;
[0032] Input the multi-document summary prompt into the multi-document summary model to obtain the review text corresponding to the outline title generated by the multi-document summary model.
[0033] The second aspect of this application provides a device for generating a literature review, including:
[0034] An acquisition unit, configured to acquire multiple documents;
[0035] A literature review outline generation unit, configured to generate a literature review outline using the titles of each document. The literature review outline includes a general title and outline titles;
[0036] A literature classification unit for classifying each piece of literature based on the outline titles, where each outline title corresponds to a literature collection, and each literature collection includes at least one piece of literature;
[0037] A review text generation unit for generating a review text corresponding to each outline title based on the literature collection corresponding to the outline title;
[0038] A literature review generation unit for generating a final literature review based on the general title, the outline titles, and the review texts corresponding to each outline title. In the final literature review, the review texts corresponding to each outline title include citation identifiers for the corresponding original pieces of literature.
[0039] In a possible implementation, the literature review outline generation unit is specifically configured to:
[0040] Obtain an outline generation prompt template, where the outline generation prompt template includes outline generation task description information and a literature title filling slot;
[0041] Fill the titles of the pieces of literature into the literature title filling slot to obtain an outline generation prompt;
[0042] Input the outline generation prompt into an outline generation model to obtain the literature review outline, where the outline generation model is a large language model trained to have the ability to generate a literature review outline.
[0043] In a possible implementation, the literature classification unit includes:
[0044] A matching judgment unit for, for each piece of literature, judging the matching of the literature with each of the outline titles to determine the matching judgment results of the literature with each of the outline titles;
[0045] A classification unit for classifying each piece of literature based on the matching judgment results of each piece of literature with each outline title.
[0046] In a possible implementation, the device further includes:
[0047] A literature summary unit for summarizing each piece of literature to obtain a summary of each piece of literature before judging the matching of the literature with each of the outline titles to determine the matching judgment results of the literature with each of the outline titles;
[0048] Then the matching judgment unit is specifically configured to:
[0049] For each document, match the summary of the document with each of the outline headings to determine the matching judgment results between the document and each of the outline headings.
[0050] In a possible implementation, the document summary unit is specifically configured to:
[0051] Asynchronously call a single-document summary model to summarize each document, obtaining the summaries of each document. The single-document summary model is a large language model trained to have the ability to summarize a single document.
[0052] In a possible implementation, the document summary unit is specifically configured to:
[0053] Obtain a single-document summary prompt template, where the single-document summary prompt template includes single-document summary task description information and a document filling slot;
[0054] Fill the document into the document filling slot to obtain a single-document summary prompt;
[0055] Input the single-document summary prompt into the single-document summary model to obtain the summary of the document generated by the single-document summary model.
[0056] In a possible implementation, each of the outline headings corresponds to a document set, each document set includes information of at least one document, and the information of each document includes the title and summary of the document. Then, the review text generation unit is specifically configured to:
[0057] Obtain a multi-document summary prompt template, where the multi-document summary prompt template includes multi-document summary task description information, multi-document summary examples, and a multi-document information filling slot;
[0058] Fill the information of each document in the document set corresponding to the outline heading into the multi-document information filling slot to obtain a multi-document summary prompt;
[0059] Input the multi-document summary prompt into the multi-document summary model to obtain the review text corresponding to the outline heading generated by the multi-document summary model.
[0060] The third aspect of this application provides a computer program product, including computer-readable instructions. When the computer-readable instructions run on an electronic device, the electronic device is enabled to implement the literature review generation method in the above first aspect or any implementation manner of the first aspect.
[0061] A fourth aspect of the present application provides an electronic device, including at least one processor and a memory connected to the processor, where:
[0062] The memory is used to store a computer program;
[0063] The processor is used to execute the computer program so that the electronic device can implement the literature review generation method according to the first aspect or any implementation manner of the first aspect above.
[0064] A fifth aspect of the present application provides a computer-readable storage medium carrying one or more computer programs, which can enable the electronic device to implement the literature review generation method according to the first aspect or any implementation manner of the first aspect above when the one or more computer programs are executed by the electronic device.
[0065] By means of the above technical solutions, for the literature review generation method and related devices provided by the present application, after obtaining multiple documents, first use the titles of each document to generate a literature review outline, and the literature review outline includes a general title and outline titles; then classify each document based on the outline titles, each outline title corresponds to a document set, and each document set includes at least one document; and for each outline title, generate a review text corresponding to the outline title based on the document set corresponding to the outline title; finally, generate the final literature review based on the general title, outline titles, and review texts corresponding to each outline title. In the final literature review, the review texts corresponding to each outline title include citation identifiers corresponding to each original document. This solution can realize the generation of a literature review based on a large number of documents, and ensure the generation efficiency, generation quality, and reliable traceability. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] In combination with the drawings and with reference to the following specific embodiments, the above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more obvious. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the original elements and elements are not necessarily drawn to scale.
[0067] Figure 1 It is a schematic flowchart of a literature review generation method provided by an embodiment of the present application;
[0068] Figure 2 It is a schematic diagram of an example of a final literature review provided by an embodiment of the present application;
[0069] Figure 3 It is a schematic structural diagram of a literature review generation device provided by an embodiment of the present application;
[0070] Figure 4 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners
[0071] The embodiments of the present application will be described below with reference to the accompanying drawings in the embodiments of the present application. The terms used in the implementation manners part of the present application are only used to explain the specific embodiments of the present application, rather than to limit the present application.
[0072] The embodiments of the present application will be described below with reference to the accompanying drawings. As can be known to those of ordinary skill in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.
[0073] The terms "first", "second", etc. in the description and claims of the present application and the above accompanying drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such terms can be interchanged under appropriate circumstances, which is only a way of distinguishing objects with the same attributes when describing the embodiments of the present application. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion, so that a process, method, system, product or device comprising a series of units does not have to be limited to those units, but may include other units not clearly listed or inherent to these processes, methods, products or devices.
[0074] A literature review is a type of writing different from research literature that researchers form after reading the literature on a certain topic in advance, through understanding, organizing, integrating, comprehensively analyzing and evaluating, and summarizing the basic facts, basic viewpoints and methods of several articles on the same topic.
[0075] Existing literature review generation methods are generally divided into literature review generation methods based on manual writing and literature review generation methods based on partial automation. The literature review generation method based on manual writing consumes a large amount of manpower and time, usually can only process a small number of literatures, and the quality of the review depends on the experience and ability of experts. The literature review generation method based on partial automation mainly relies on the abstract generation of the literature content, and generates a simple review by methods such as keyword extraction and sentence selection of the literature content. However, when the number of literatures is large, the existing literature review generation methods based on partial automation also have problems of low generation efficiency and incomplete content coverage, and cannot meet the modern information processing requirements. In addition, both of the above two existing literature review generation methods lack reliable traceability, and the generated review content is difficult to verify. Users cannot quickly locate the original literature cited in the review, thereby reducing the credibility of the review.
[0076] To solve the above problems, the embodiments of the present application provide a literature review generation method. The literature review generation method of the embodiments of the present application will be introduced in detail below with reference to the accompanying drawings.
[0077] Refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for generating a literature review provided by an embodiment of the present application. As Figure 1 shown, a method for generating a literature review provided by an embodiment of the present application may include the following steps, and these steps will be described in detail below.
[0078] S101: Obtain multiple pieces of literature;
[0079] In the present application, the number of multiple pieces of literature is not limited. In one possible implementation, the number of multiple pieces of literature may be more than a hundred.
[0080] S102: Generate a literature review outline using the titles of each piece of literature, where the literature review outline includes a general title and outline titles;
[0081] In the present application, the literature review outline may include first-level titles and second-level titles. Among them, the first-level title corresponds to the general title, and the second-level title corresponds to the outline title. Generally, there is only one general title, and there may be multiple outline titles. For ease of understanding, in the present application, an example of a literature review outline is provided as follows:
[0082] Research Review of Large Language Models
[0083] 1. Introduction
[0084] 2. Overview
[0085] 3. Resources Related to Large Language Models
[0086] 4. Pre-training Technologies of Large Language Models
[0087] 5. Adaptation Technologies of Large Language Models
[0088] 6. Usage Technologies of Large Language Models
[0089] 7. Capability Evaluation of Large Language Models
[0090] 8. Guide for Using Prompt Design of Large Language Models
[0091] Among them, "Research Review of Large Language Models" is the general title, and the remaining 8 titles are "outline titles".
[0092] In the present application, generating a literature review outline including a general title and outline titles can facilitate the structured organization of the content of the literature review.
[0093] S103: Classify each piece of literature based on the outline titles. Each outline title corresponds to a literature set, and each literature set includes at least one piece of literature;
[0094] In this application, each document in the document set corresponding to each outline title is a document whose content matches the outline title. Classifying each document based on the outline title realizes the matching of the document content with the outline title.
[0095] S104: For each outline title, generate a review text corresponding to the outline title based on the document set corresponding to the outline title;
[0096] In this application, for each outline title, a review text corresponding to the outline title is generated based on the document set corresponding to the outline title, so that the essence content of each document can be reasonably distributed in different parts of the review outline, thereby ensuring the comprehensiveness of the content of the generated literature review and the rationality of the logical structure.
[0097] S105: Generate a final literature review based on the general title, the outline titles, and the review texts corresponding to each of the outline titles. In the final literature review, the review texts corresponding to each outline title include citation identifiers corresponding to the respective original documents.
[0098] In this application, the general title, the outline titles, and the review texts corresponding to each of the outline titles can be post-processed and converted into HTML format to generate a final literature review. In this application, in the final literature review, the review texts corresponding to each outline title including citation identifiers corresponding to the respective original documents can ensure the reliable traceability of the final literature review.
[0099] For ease of understanding, refer to Figure 2 , Figure 2 which is a schematic diagram of an example of a final literature review provided by an embodiment of this application. As Figure 2 shown, the review text corresponding to the outline title "1 Introduction" includes a citation identifier and can be clicked to jump to the corresponding original document.
[0100] The literature review generation method provided in this embodiment, after obtaining multiple documents, first generates a literature review outline using the titles of each document. The literature review outline includes a general title and outline titles; then classifies each document based on the outline titles, each outline title corresponding to a document set, and each document set including at least one document; and for each outline title, generates a review text corresponding to the outline title based on the document set corresponding to the outline title; finally generates a final literature review based on the general title, the outline titles, and the review texts corresponding to each of the outline titles. In the final literature review, the review texts corresponding to each outline title include citation identifiers corresponding to the respective original documents. This solution can realize the generation of a literature review based on a large number of documents and ensure the generation efficiency, generation quality, and reliable traceability.
[0101] In another embodiment of the present application, a specific implementation method for generating a literature review outline by using the titles of each piece of literature is described. This method may include the following steps:
[0102] S201: Obtain an outline generation prompt template, which contains outline generation task description information and a literature title filling slot;
[0103] For ease of understanding, an example of an outline generation prompt template is provided in the present application, as follows: This is a literature clustering task. Please cluster according to the literature titles. Each line represents the title of a piece of literature
[0104] [××××××]
[0105] [××××××]
[0106] [××××××]
[0107] ……
[0108] Your answer should follow the following instructions:
[0109] (1) Use all the literature and keep the literature titles unchanged;
[0110] (2) Cluster these literatures according to their research directions, with a maximum of no more than 6 categories;
[0111] (3) Each cluster has no more than 10 pieces of literature;
[0112] (4) According to the clustering results, each category is used as a chapter, and write a title for each chapter;
[0113] (5) According to the written chapters, give a general title (ending with "Review") and an outline for your review;
[0114] (6) Only give an answer in JSON format, referring to the following format:
[0115] {
[0116] "General Title":"____",
[0117] "Outline":{
[0118] "____":["",""],
[0119] "____":["",""], ......
[0121] }
[0122] }
[0123] Among them, "each line represents the title of a document
[0124] [××××××]
[0125] [××××××]
[0126] [××××××]
[0127] ……" is the document title filling slot.
[0128] S202: Fill the titles of the respective documents into the document title filling slot to obtain an outline generation prompt;
[0129] S203: Input the outline generation prompt into an outline generation model to obtain the literature review outline, and the outline generation model is a large language model trained with the ability to generate literature review outlines.
[0130] In this application, the powerful training and understanding capabilities of the large language model can be fully utilized to deeply analyze and summarize the titles of the input respective documents. Through semantic understanding and knowledge expression capabilities, the large language model extracts the core themes from the titles of each document to ensure the accuracy and generality of each outline title. Then, based on the semantic similarity and logical relationship between different research directions, the general title and outline titles are generated, making the review structure hierarchical and the content well-organized. In this way, the readability and logical consistency of the overall literature review can be effectively improved, facilitating users to more quickly grasp the core information of each research field and deeply understand the key development trends of each research direction. This also further reflects the advantages of the large language model in multi-dimensional understanding and induction, ensuring that the generated literature review outline accurately reflects the overall theme and details of multiple documents.
[0131] In another embodiment of this application, the specific implementation manner of classifying each document based on the outline title is described, and this manner may include the following steps:
[0132] S301: For each document, match and judge the document with each of the outline titles to determine the match and judgment results of the document with each of the outline titles;
[0133] In this application, for each document, various technical means can be used to match and judge it with each of the outline headings, so as to improve the matching and judging results of the document and each of the outline headings, and further enable the content of each document to be reasonably classified under the appropriate outline heading. The various technical means include, but are not limited to, the TF-IDF (Term Frequency-Inverse Document Frequency) method, the embedding model, and the string matching technology. Regarding this, this application does not make any limitations.
[0134] Among them, the TF-IDF (Term Frequency-Inverse Document Frequency) method is used to extract the key features of each document, so as to quantify the text relevance between each document and the outline heading, ensuring that documents that are relatively well-matched with the content of a certain outline heading can be quickly screened out. At the same time, in combination with the embedding model, deep learning technology is used to represent the text as a vector to capture the semantic similarity between the document and the outline heading. By calculating the similarity between the vectors, documents with relevant content can be more accurately matched with the corresponding outline heading. In addition, the string matching technology is adopted. By analyzing the specific keywords and phrases in the document and the outline heading, the matching accuracy is further improved to ensure that the document can be classified under the most suitable review outline heading.
[0135] In this application, in a possible implementation, for each document, the original text of the document can be used to match and judge with each of the outline headings, and then the classification of each document based on the outline headings can be realized. Considering the large number of documents and the problem of the excessive length of the original text of the documents, directly using the original text of each document to match and judge with each of the outline headings is inefficient.
[0136] Therefore, in another possible implementation, before the document is matched and judged with each of the outline headings to determine the matching and judging results of the document and each of the outline headings, the method further includes: summarizing each document to obtain a summary of each document.
[0137] In this application, by summarizing each document to obtain a summary of each document, the text length and information volume of the document can be reduced, ensuring the efficiency of subsequent processing.
[0138] Then, for each document, matching and judging the document with each of the outline headings to determine the matching and judging results of the document and each of the outline headings includes:
[0139] For each document, matching and judging the summary of the document with each of the outline headings to determine the matching and judging results of the document and each of the outline headings.
[0140] S302: Classify each document based on the matching judgment results between each document and each outline title.
[0141] In this application, for each outline title, the set of documents that match the outline title is the document set corresponding to the outline title.
[0142] In this application, the summarizing each document to obtain the summary of each document includes:
[0143] Asynchronously call the single-document summarization model to summarize each document to obtain the summary of each document. The single-document summarization model is a large language model trained to have the ability to summarize a single document.
[0144] Considering the large number of documents, in this application, the asynchronous call of the single-document summarization model can be used to achieve efficient document summarization. Through experiments, it can complete the summarization of a hundred documents within half a minute by using the asynchronous call technology.
[0145] Among them, for each document, calling the single-document summarization model to summarize the document to obtain the summary of the document includes: obtaining the single-document summarization prompt template, which contains the single-document summarization task description information and the document filling slot; filling the document into the document filling slot to obtain the single-document summarization prompt; inputting the single-document summarization prompt into the single-document summarization model to obtain the summary of the document generated by the single-document summarization model.
[0146] For ease of understanding, in this application, a single-document summarization prompt template example is provided as follows: This is a text summarization task, and the text that needs to be summarized is as follows:
[0147] [××××××]
[0148] Please summarize according to the following requirements:
[0149] (1) If the paragraph is in English, generate an English summary with no more than 200 English words;
[0150] (2) If the paragraph is in Chinese, generate a Chinese summary with no more than 300 Chinese characters;
[0151] (3) If the paragraph involves important information such as the author, time, research method, experimental results, etc., it must be retained.
[0152] Among them, "[The text that needs to be summarized is as follows: [××××××]]" is the document filling slot.
[0153] As mentioned in the above embodiments, before matching and determining the literature with each of the outline headings to obtain the matching judgment results of the literature with each of the outline headings, the method further includes: summarizing each piece of literature to obtain a summary of each piece of literature. In this application, by summarizing each piece of literature to obtain a summary of each piece of literature, the text length and information volume of the literature can be reduced, ensuring the efficiency of subsequent processing. Therefore, in a possible implementation, each of the outline headings corresponds to a literature set, each literature set includes information of at least one piece of literature, and the information of each piece of literature includes the title and summary of the literature. Then, generating the review text corresponding to the outline heading based on the literature set corresponding to the outline heading includes: obtaining a multi-literature summary prompt template, where the multi-literature summary prompt template contains multi-literature summary task description information, multi-literature summary examples, and multi-literature information filling slots; filling the information of each piece of literature in the literature set corresponding to the outline heading into the multi-literature information filling slots to obtain a multi-literature summary prompt; inputting the multi-literature summary prompt into the multi-literature summary model to obtain the review text corresponding to the outline heading generated by the multi-literature summary model.
[0154] For ease of understanding, a multi-literature summary prompt template example is provided in this application, which is as follows: This is a literature summary task, and the examples are provided as follows:
[0155] "Now summarize the following literature on the optimization directions of reconfigurable intelligent surfaces
[0156] Title: ××××××
[0157] Summary: ××××××
[0158] Title: ××××××
[0159] Summary: ××××××
[0160] Output: Reconfigurable intelligent surface (RIS) is a rapidly developing wireless communication technology recently, aiming to significantly enhance communication performance by intelligently modulating the propagation environment of wireless signals. RIS consists of a large number of low-cost, low-complexity, and low-power reflection units, which can independently and actively adjust the amplitude and phase of the electromagnetic waves reflected from the surface.
[0161] Intelligent reflecting surface (IRS)-enhanced full-duplex multiple-input multiple-output (MIMO) two-way communication systems have been studied. By jointly optimizing the source precoder and the IRS phase shift matrix, the total rate of the system is maximized. Using the idea of the Arimoto-Blahut algorithm, they split the non-convex optimization problem into three sub-problems and solved them separately, and all sub-problems can be effectively solved by closed-form solutions. In addition, they also considered practical IRS assumptions, such as discrete phase shift levels, and numerical results verified the convergence and performance of the proposed scheme. On the other hand, for the case of line-of-sight paths being ignored in indoor millimeter-wave environments, using reconfigurable intelligent surfaces (RIS) for channel capacity optimization has been studied. Two optimization schemes are proposed to customize the characteristics of RIS reflection elements to maximize the channel capacity. The first optimization scheme only utilizes the adjustability of RIS reflection elements and derives an approximate expression explaining the relationship between the channel capacity gain and system parameters. The second optimization scheme jointly optimizes the RIS reflection elements and the transmit phase precoder, and for this scheme, a low-complexity technique called global co-phasing is proposed to determine the phase shift values used on the RIS. Simulation results show that optimizing the RIS reflection elements results in significant channel capacity gains, which increase with the increase in the number of RIS elements. All these studies open up new possibilities for the practical application of RIS in wireless communication.”
[0162] “Now summarize the literature on the application directions of the following complex adaptive systems
[0163] Title: [××××××]
[0164] Summary: [××××××]
[0165] Title: [××××××]
[0166] Summary: [××××××]
[0167] Title: [××××××]
[0168] Summary: [××××××]
[0169] Your answer should follow the following requirements:
[0170] (1) First introduce the direction and then analyze the literature;
[0171] (2) Do not include specific literature names and do not summarize
[0172] (3) The literature is presented in the same paragraph without separation
[0173] (4) There should be no more than 3 paragraphs.”
[0174] Among them, “Now summarize the literature on the application directions of the following complex adaptive systems
[0175] Title: [××××××]
[0176] Summary: [××××××]
[0177] Title: [××××××]
[0178] Summary: [××××××]
[0179] Title: [××××××]
[0180] Summary: [××××××]” are slots for filling in information from multiple documents.
[0181] The above introduced a method for generating a literature review provided by an embodiment of the present application. Next, a device for executing the above-mentioned method for generating a literature review will be introduced.
[0182] Please refer to Figure 3 , Figure 3 , which is a schematic structural diagram of a device for generating a literature review provided by an embodiment of the present application. As Figure 3 shown, the device for generating a literature review includes:
[0183] An acquisition unit 11, configured to acquire multiple documents;
[0184] A literature review outline generation unit 12, configured to generate a literature review outline by using the titles of each document, where the literature review outline includes a general title and outline titles;
[0185] A literature classification unit 13, configured to classify each document based on the outline titles, where each outline title corresponds to a document set, and each document set includes at least one document;
[0186] A review text generation unit 14, configured to generate a review text corresponding to each outline title based on the document set corresponding to the outline title;
[0187] A literature review generation unit 15, configured to generate a final literature review based on the general title, the outline titles, and the review texts corresponding to each of the outline titles, where in the final literature review, the review texts corresponding to each outline title include citation identifiers of the corresponding original documents.
[0188] In a possible implementation, the literature review outline generation unit is specifically configured to:
[0189] Acquire an outline generation prompt template, where the outline generation prompt template includes outline generation task description information and a document title filling slot;
[0190] Fill the titles of the respective documents into the document title filling slots to obtain an outline generation prompt;
[0191] Input the outline generation prompt into an outline generation model to obtain the literature review outline, where the outline generation model is a large language model trained to have the ability to generate a literature review outline.
[0192] In a possible implementation, the document classification unit includes:
[0193] A matching judgment unit for, for each document, matching and judging the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles;
[0194] A classification unit for classifying each document based on the matching judgment results of each document with each outline title.
[0195] In a possible implementation, the device further includes:
[0196] A document summarization unit for summarizing each document to obtain a summary of each document before matching and judging the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles;
[0197] Then the matching judgment unit is specifically used for:
[0198] For each document, matching and judging the summary of the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles.
[0199] In a possible implementation, the document summarization unit is specifically used for:
[0200] Asynchronously call a single-document summarization model to summarize each document to obtain summaries of each document, where the single-document summarization model is a large language model trained to have the ability to summarize a single document.
[0201] In a possible implementation, the document summarization unit is specifically used for:
[0202] Obtain a single-document summarization prompt template, where the single-document summarization prompt template contains single-document summarization task description information and a document filling slot;
[0203] Fill the document into the document filling slot to obtain a single-document summarization prompt;
[0204] Input the single-document summarization prompt into the single-document summarization model to obtain the summary of the document generated by the single-document summarization model.
[0205] In a possible implementation, each of the outline headings corresponds to a document set, each document set includes information of at least one document, and the information of each document includes the title and summary of the document. Then, the review text generation unit is specifically configured to:
[0206] Obtain a plurality of document summary prompt templates, where the plurality of document summary prompt templates include information on a plurality of document summary tasks, examples of a plurality of document summaries, and slots for filling in information of a plurality of documents;
[0207] Fill the information of each document in the document set corresponding to the outline heading into the slots for filling in information of a plurality of documents to obtain a plurality of document summary prompts;
[0208] Input the plurality of document summary prompts into the plurality of document summary models to obtain a review text corresponding to the outline heading generated by the plurality of document summary models.
[0209] An electronic device is further provided in an embodiment of the present application. Refer to Figure 4 As shown, it shows a schematic structural diagram of an electronic device suitable for implementing the electronic device in the embodiment of the present application. The electronic device in the embodiment of the present application may include, but is not limited to, fixed terminals such as mobile phones, laptop computers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), desktop computers, and the like. Figure 4 The electronic device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiment of the present application.
[0210] As Figure 4 shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 601, which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0211] Typically, the following devices can be connected to the I / O interface 605: input devices 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 can allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 4 an electronic device with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices can be implemented or had.
[0212] An embodiment of the present application also provides a computer program product including computer-readable instructions, which, when running on an electronic device, enable the electronic device to implement any one of the literature review generation methods provided by the embodiments of the present application.
[0213] An embodiment of the present application also provides a computer-readable storage medium carrying one or more computer programs, which, when executed by an electronic device, can enable the electronic device to implement any one of the literature review generation methods provided by the embodiments of the present application.
[0214] In addition, it should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, in the drawings of the device embodiments provided in the present application, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines.
[0215] Through the description of the above embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software plus necessary general hardware. Of course, it can also be implemented by dedicated hardware including application-specific integrated circuits, dedicated CPUs, dedicated memories, dedicated components, etc. Generally, functions accomplished by computer programs can be easily implemented by corresponding hardware, and the specific hardware structures for implementing the same function can also be diverse, such as analog circuits, digital circuits, or dedicated circuits, etc. However, for the present application, software program implementation is a better embodiment in more cases. Based on such an understanding, the technical solution of the present application, in essence, or the part that makes contributions to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disc of a computer, etc., and includes several instructions for causing a computer device (which can be a personal computer, training device, or network device, etc.) to execute the methods described in various embodiments of the present application.
[0216] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product.
[0217] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a dedicated computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, training device, or data center to another website, computer, training device, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can store, or a data storage device such as a training device or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state drive (SSD)).
Claims
1. A method for generating a literature review, characterized in that, Including: Obtain multiple documents; Generate a literature review outline using the titles of each document, where the literature review outline includes a general title and outline titles; Classify each document based on the outline titles, with each outline title corresponding to a document set, and each document set including at least one document; For each outline title, generate a review text corresponding to the outline title based on the document set corresponding to the outline title; Generate a final literature review based on the general title, the outline titles, and the review texts corresponding to each outline title. In the final literature review, the review texts corresponding to each outline title include citation identifiers for the corresponding original documents.
2. The method according to claim 1, wherein The generating a literature review outline using the titles of each document includes: Obtain an outline generation prompt template, which contains outline generation task description information and a document title filling slot; Fill the titles of each document into the document title filling slot to obtain an outline generation prompt; Input the outline generation prompt into an outline generation model to obtain the literature review outline, where the outline generation model is a large language model trained to have the ability to generate literature review outlines.
3. The method according to claim 1, wherein The classifying each document based on the outline titles includes: For each document, match the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles; Classify each document based on the matching judgment results of each document with each outline title.
4. The method according to claim 3, wherein Before the matching judgment of the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles, the method further includes: Summarize each document to obtain a summary of each document; Then, for each document, the matching judgment of the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles includes: For each document, match the summary of the document with each of the outline titles to determine the matching judgment results of the document with each of the outline titles.
5. The method according to claim 4, wherein The summarizing each document to obtain a summary of each document includes: Asynchronously call a single-document summary model to summarize each document to obtain a summary of each document, where the single-document summary model is a large language model trained to have the ability to summarize a single document.
6. The method according to claim 5, wherein For each document, calling the single-document summary model to summarize the document to obtain a summary of the document includes: Obtain a single-document summary prompt template, which contains single-document summary task description information and a document filling slot; Fill the document into the document filling slot to obtain a single-document summary prompt; Input the single-document summary prompt into the single-document summary model to obtain the summary of the document generated by the single-document summary model.
7. The method according to claim 4, wherein Each of the outline headings corresponds to a literature collection, each literature collection includes information of at least one literature, and the information of each literature includes the title and summary of the literature. Then, generating the review text corresponding to the outline heading based on the literature collection corresponding to the outline heading includes: Obtain a multi-literature summary prompt template, which contains multi-literature summary task description information, multi-literature summary examples, and multi-literature information filling slots; Fill the information of each literature in the literature collection corresponding to the outline heading into the multi-literature information filling slots to obtain a multi-literature summary prompt; Input the multi-literature summary prompt into the multi-literature summary model to obtain the review text corresponding to the outline heading generated by the multi-literature summary model.
8. A literature review generation device, characterized in that Includes: An acquisition unit for acquiring multiple literatures; A literature review outline generation unit for generating a literature review outline using the titles of each literature, where the literature review outline includes a general title and outline headings; A literature classification unit for classifying each literature based on the outline heading, each outline heading corresponding to a literature collection, and each literature collection including at least one literature; A review text generation unit for generating the review text corresponding to each outline heading based on the literature collection corresponding to the outline heading; A literature review generation unit for generating a final literature review based on the general title, the outline headings, and the review texts corresponding to each of the outline headings. In the final literature review, the review texts corresponding to each outline heading include citation identifiers corresponding to each original literature.
9. A computer program product, characterized in that, Includes computer-readable instructions that, when run on an electronic device, cause the electronic device to implement the literature review generation method according to any one of claims 1 to 7.
10. An electronic device, characterized in that, Includes at least one processor and a memory connected to the processor, where: The memory is used to store a computer program; The processor is used to execute the computer program so that the electronic device can implement the literature review generation method according to any one of claims 1 to 7.
11. A computer-readable storage medium, characterized in that, The storage medium carries one or more computer programs that, when executed by an electronic device, can cause the electronic device to implement the literature review generation method according to any one of claims 1 to 7.