Literature review generation method and device, electronic equipment and readable storage medium

Abstract

The invention discloses a literature review generation method and device, electronic equipment and a readable storage medium. The method comprises the following steps: acquiring a review theme and a plurality of literatures; clustering processing is conducted on the multiple literatures, multiple literature clusters are generated, and each literature cluster comprises at least two similar literatures; generating a chapter text corresponding to each literature cluster through a large language model based on the overview theme and the plurality of literature clusters; and generating a literature review based on the plurality of chapter texts. Through the combination of the clustering technology and the large language model, the problem that the input length is limited when the large language model generates the review is solved, and the efficiency and quality of literature extraction are effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and in particular to a method, device, electronic device and readable storage medium for generating a literature review. Background Art

[0002] With the rapid development of the information age, the demand for document content processing and analysis in various fields is growing. In order to improve the efficiency and quality of document processing, with the development and popularization of large language model applications, large language models are being used to automatically extract documents. However, due to the input length limit of large language models, the number of tokens input by large language models is difficult to meet the huge amount of document input. This limitation makes it impossible for large language models to summarize a large number of documents into a review at one time. Summary of the invention

[0003] In view of this, the present application provides a method, device, electronic device and readable storage medium for generating a literature review, the main purpose of which is to solve the technical problem in the prior art that due to the limitation on the input length of a large language model, the number of tokens input by the large language model is difficult to meet the huge amount of literature input, making it impossible for the large language model to summarize a large amount of literature into a review at one time.

[0004] According to the first aspect of the present application, a method for generating a literature review is provided, the method comprising:

[0005] Obtain review topics and multiple papers;

[0006] Clustering multiple documents to generate multiple document clusters, wherein each document cluster includes at least two similar documents;

[0007] Based on the review topic and multiple document clusters, a large language model is used to generate the chapter text corresponding to each document cluster;

[0008] Generate a literature review based on multiple chapter texts.

[0009] Optionally, the step of clustering multiple documents to generate multiple document clusters specifically includes:

[0010] Perform word segmentation preprocessing on each document to obtain multiple words corresponding to each document;

[0011] Convert multiple words in each document into embedding vectors and perform dimensionality reduction on the embedding vectors;

[0012] Based on the reduced-dimensional embedding vector corresponding to each document, multiple documents are clustered to generate multiple document clusters.

[0013] Optionally, based on the reduced-dimensional embedding vector corresponding to each document, a step of clustering multiple documents to generate multiple document clusters specifically includes:

[0014] Based on the reduced-dimensional embedding vector of each document, the similarity between any two documents is calculated using the distance measurement method;

[0015] Based on the similarity value between any two documents, multiple documents are clustered through a preset clustering algorithm to generate multiple document clusters.

[0016] Optionally, based on the review topic and multiple document clusters, the step of generating chapter text corresponding to each document cluster by using a large language model specifically includes:

[0017] For any document cluster, obtain multiple document titles and multiple document abstracts of multiple similar documents included in the document cluster;

[0018] Based on the review topic, multiple document titles and multiple document abstracts, a chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text are generated through a large language model;

[0019] Based on the review topic, chapter title, multiple paragraph titles and at least two similar documents contained in the document cluster, the chapter text corresponding to the document cluster is generated through a large language model.

[0020] Optionally, based on the review topic, multiple document titles and multiple document abstracts, the step of generating the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text by using a large language model specifically includes:

[0021] For any document cluster, the review topic, multiple document titles and multiple document abstracts are filled into the first preset question template to generate the first question;

[0022] A question is asked to the large language model based on the first question, so that the large language model outputs the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text based on the review topic, multiple document titles and multiple document abstracts.

[0023] Optionally, based on the review topic, chapter titles, multiple paragraph titles and at least two similar documents included in the document cluster, the step of generating the chapter text corresponding to the document cluster through the large language model specifically includes:

[0024] For any document cluster, the review topic, chapter title and multiple paragraph titles are filled into the second preset question template to generate multiple second questions corresponding to multiple paragraphs contained in the chapter text;

[0025] Asking questions to the large language model based on the multiple second questions enables the large language model to output paragraph content corresponding to each paragraph based on the document content of at least two similar documents in the document cluster;

[0026] The chapter title, multiple paragraph titles, and multiple paragraph contents of multiple paragraphs are aggregated to generate the chapter text corresponding to the document cluster.

[0027] Optionally, the method further comprises:

[0028] Generate review titles and review summaries based on the content of the literature review through a large language model.

[0029] According to a second aspect of the present application, a device for generating a literature review is provided, the device comprising:

[0030] The acquisition module is used to obtain review topics and multiple documents;

[0031] A first generation module is used to perform clustering processing on a plurality of documents to generate a plurality of document clusters, wherein each document cluster includes at least two similar documents;

[0032] The second generation module is used to generate the chapter text corresponding to each document cluster through a large language model based on the review topic and multiple document clusters;

[0033] The third generation module is used to generate a literature review based on multiple chapter texts.

[0034] According to a third aspect of the present application, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the methods described in the first aspect when executing the computer program.

[0035] According to a fourth aspect of the present application, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0036] By means of the above technical solution, the present application provides a method, device, electronic device and readable storage medium for generating a literature review. First, a cluster analysis is performed on multiple documents, and documents with similar contents are divided into the same set to form multiple document clusters, so that relatively small document sets are focused on in the subsequent processing process, solving the technical problem of limited input length of the large language model. Afterwards, the large language model is used to generate the corresponding chapter text for each document cluster, and the corresponding chapter text is generated based on the document cluster composed of similar documents, and finally the literature review is summarized to ensure the relevance and consistency of the chapter text content, so that the chapter paragraphs are more logical, avoid information isolation, and ensure that the literature review comprehensively covers related research under the review topic. This application solves the problem of limited input length of the large language model when generating a review by combining clustering technology and a large language model, and effectively improves the efficiency and quality of document extraction.

[0037] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0039] Figure 1 A schematic diagram of a method for generating a literature review provided in an embodiment of the present application is shown;

[0040] Figure 2 A schematic block diagram of another method for generating a literature review provided in an embodiment of the present application is shown;

[0041] Figure 3 A schematic diagram of the structure of a literature review generating device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0043] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0044] The following, in conjunction with the accompanying drawings, describes in detail the literature review generation method, device, electronic device and readable storage medium provided in the embodiments of the present application through specific embodiments and their application scenarios.

[0045] This application embodiment provides a method for generating a literature review, such as Figure 1 As shown, the method includes:

[0046] S10. Obtain review topics and multiple papers.

[0047] In this step, the review topic input by the user and multiple documents related to the review topic are received, where the review topic is a pre-selected research topic. After the topic is clear, the researcher focuses on relevant literature citations to ensure the pertinence of the review.

[0048] It should be noted that the number of multiple literature citations here is not just a few or dozens of articles, but can be tens of thousands. By collecting and preparing a large number of literature citations, it is helpful to form a comprehensive understanding of the topic.

[0049] S20, clustering multiple documents to generate multiple document clusters;

[0050] Each document cluster includes at least two similar documents.

[0051] In this step, a large number of literature citations are clustered and analyzed to classify literature with similar contents into the same set to form several literature clusters, wherein each literature cluster contains at least two similar literatures with high similarity.

[0052] Through the above method, multiple documents are clustered and analyzed to form document clusters, ensuring that the documents contained in each cluster are consistent in subject matter.

[0053] In one embodiment of the present application, step S20, clustering multiple documents to generate multiple document clusters, includes:

[0054] S21, performing word segmentation preprocessing on each document to obtain multiple words corresponding to each document;

[0055] S22, converting multiple words in each document into embedding vectors, and performing dimensionality reduction processing on the embedding vectors;

[0056] S23. Based on the reduced-dimensional embedding vector corresponding to each document, multiple documents are clustered to generate multiple document clusters.

[0057] For steps S21-S23, the document corpus cannot be used as the initial data for vectorized modeling. Only standardized data formats can ensure the smooth progress of subsequent steps. Therefore, before clustering the documents, it is necessary to pre-process the document content by word segmentation, and divide the continuous character strings in the document into independent words with specific semantics according to certain rules.

[0058] Afterwards, a text encoder model (such as sentence-transformers / all-MiniLM-L6-v2 model) is used to convert multiple words in each document into word vectors and embed them to obtain embedding vectors. Subsequently, the high-dimensional embedding vectors are mapped to a lower-dimensional space to simplify the representation while retaining the original information.

[0059] Furthermore, a clustering algorithm is used to perform cluster analysis on the embedded vectors after dimensionality reduction. According to the distance similarity, the documents with similar contents are divided into the same cluster, so that each cluster finally formed can summarize a specific topic and provide a basis for the subsequent summary of the literature review.

[0060] Through the above method, it is only necessary to focus on a relatively small text set in the subsequent large language model processing process. While solving the problem of limited input length of the large language model, it helps to generate more focused and relevant chapter content, so that the chapter articles formed by each cluster can be developed around a specific topic, avoiding the problem of scattered or inconsistent topics in the generated chapter articles, and ensuring the overall coherence and logic of the final review.

[0061] In one embodiment of the present application, step S23, clustering multiple documents based on the reduced-dimensional embedding vector corresponding to each document to generate the multiple document clusters, includes:

[0062] S231, based on the reduced dimension embedding vector of each document, calculating the similarity value between any two documents by using a distance measurement method;

[0063] S232. Based on the similarity values ​​between any two documents, multiple documents are clustered using a preset clustering algorithm to generate multiple document clusters.

[0064] For steps S231-232, select a suitable distance measurement method (such as cosine similarity, Euclidean distance, etc.) to measure the numerical proximity of two vectors, and then calculate the similarity between any two documents. Afterwards, select a suitable clustering algorithm to group similar documents into the same cluster, and each document cluster contains a group of documents that are similar in some characteristics (such as research questions, methods or conclusions). Among them, each document cluster contains at least two similar documents, indicating that these documents have commonalities in a certain research topic.

[0065] Optionally, the preset clustering algorithm may be Agglomerative Clustering, Kmeans, DBSCAN, etc.

[0066] In one embodiment of the present application, step S30, based on the review topic and multiple document clusters, generates chapter text corresponding to each document cluster through a large language model, including:

[0067] S31. For any document cluster, obtain multiple document titles and multiple document abstracts of multiple similar documents included in the document cluster;

[0068] S32, based on the review topic, multiple document titles and multiple document abstracts, generating, through a large language model, a chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text;

[0069] S33. Based on the review topic, chapter title, multiple paragraph titles and at least two similar documents contained in the document cluster, generate the chapter text corresponding to the document cluster through a large language model.

[0070] For steps S31-33, for any cluster, the selected review topic, the document titles and document abstracts of similar documents in the document cluster are used as input items, and the chapter titles of the chapter articles corresponding to the document cluster and the multiple paragraph titles of the multiple paragraphs in the chapter articles are automatically generated through the large language model. Among them, the chapter title generates a summary title for each chapter of the review, and this title can summarize the main content or theme of the chapter; multiple paragraph titles are the paragraph titles of the paragraphs contained in each chapter, and these paragraph titles will guide readers to understand the specific content of each paragraph, making the review structure clearer.

[0071] Furthermore, the large language model is used to select the specific content of the appropriate supplementary paragraphs in the document cluster according to the chapter title and each paragraph title of each chapter. The paragraph content of each paragraph is generated in turn in the above manner, and finally all the paragraph contents are summarized to form the chapter article corresponding to the document cluster.

[0072] It should be noted that the large language model (such as GPT-3 or GPT-4, etc.) in this application refers to a natural language processing model that can complete basic conversations, and does not require the size of parameters, adaptive language, deployment platform, or implementation plan.

[0073] Compared with the related art, which divides a large amount of text into multiple batches for separate processing and then merges the results, this application first performs clustering and then uses a large language model to generate chapters. This can provide a more systematic and structured framework for the review. Each cluster corresponds to a chapter, making the structure of the review clearer and the logic more rigorous. While solving the problem of input length limitation of the large language model, it avoids the possible severance of information connections between texts caused by direct batch generation, and the generated review text is prone to problems such as incoherent structure, repeated content or logical disorder, thereby improving the quality of the final generated literature review.

[0074] In one embodiment of the present application, step S32, based on the review topic, multiple document titles and multiple document abstracts, generates the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text through a large language model, including:

[0075] S321, for any document cluster, fill the review topic, multiple document titles and multiple document abstracts into a first preset question template to generate a first question;

[0076] S322. Ask the large language model a question based on the first question, so that the large language model outputs the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text based on the review topic, multiple document titles and multiple document abstracts.

[0077] For steps S321-322, the first preset question template is a preset question framework for asking questions to the large language model to generate chapter titles and paragraph titles. In the process of generating chapter articles of any document cluster, the review topic, the document titles and document abstracts corresponding to the similar documents included in the document cluster are filled into the first preset question template. After the first question is generated, the large language model is asked questions based on the first question, so that the large language model generates chapter titles and multiple paragraph titles of the chapter articles corresponding to the document cluster based on the review topic, multiple document titles and multiple document abstracts.

[0078] For example, the format of the first preset question template is:

[0079] Review topic: XXX;

[0080] Documents included in the document cluster:

[0081] Document 1: Document title 1: XXX; Document abstract 1: XXX;

[0082] Document 2: Document title 2: XXX; Document abstract 2: XXX; ......

[0084] Document n: Document title n: XXX; Document abstract n: XXX;

[0085] Please generate a suitable chapter title and multiple paragraph titles based on the above information.

[0086] In one embodiment of the present application, step S33 generates the chapter text corresponding to the document cluster through a large language model based on the review topic, chapter title, multiple paragraph titles and at least two similar documents included in the document cluster, including:

[0087] S331, for any document cluster, fill the review topic, chapter title and multiple paragraph titles into the second preset question template to generate multiple second questions corresponding to multiple paragraphs included in the chapter text;

[0088] S332, asking questions to the large language model based on the multiple second questions, so that the large language model outputs the paragraph content corresponding to each paragraph based on the document content of at least two similar documents in the document cluster;

[0089] S333: Summarize the chapter title, multiple paragraph titles, and multiple paragraph contents of multiple paragraphs to generate chapter text corresponding to the document cluster.

[0090] For steps S331-333, for any document cluster, after generating the chapter titles and multiple paragraph titles of the chapter articles corresponding to the document cluster, the review topic, the chapter titles of the chapter text and the paragraph titles of each paragraph are filled into the second preset question template to generate the second question, wherein the second preset question template is used to ask questions to the large language model, so that the large language model fills in the specific content of each paragraph based on the question. Thereafter, based on the generated second question, the large language model is asked questions, so that the large language model outputs the paragraph content corresponding to each paragraph based on the document content of at least two similar documents in the document cluster.

[0091] Afterwards, the multiple chapter texts output by the large language model are summarized and compiled into a complete review article, so that the final literature review can systematically cover the multiple topics and research directions involved in multiple literature citations, and show the connections and differences between different literature.

[0092] In one embodiment of the present application, after S40, generating a literature review based on multiple chapter texts, the method further includes:

[0093] Generate review titles and review summaries based on the content of the literature review through a large language model.

[0094] In this embodiment, in order to further improve the quality of the literature review, after generating the literature review, a large language model is used to generate a review summary to ensure that the summary can accurately summarize the core content of the entire article, and the model is used to automatically generate a title based on the review text content.

[0095] Furthermore, the generated literature review can be manually reviewed and optimized, and the order of chapters and paragraphs can be adjusted to make the literature review structure more reasonable and the logic more rigorous, ensuring that the transition between chapters is natural and smooth, and the overall content is consistent to avoid duplication or omissions.

[0096] The method for generating a literature review proposed in this application first performs cluster analysis on multiple documents, and divides documents with similar contents into the same set to form multiple document clusters, so that a relatively small document set is focused on in the subsequent processing process, solving the technical problem of limited input length of the large language model. Afterwards, the large language model is used to generate the corresponding chapter text for each document cluster, and the corresponding chapter text is generated based on the document cluster composed of similar documents, and finally the literature review is summarized to ensure the relevance and consistency of the chapter text content, make the chapter paragraphs more logical, avoid information isolation, and ensure that the literature review comprehensively covers related research under the review topic. This application solves the problem of limited input length of the large language model when generating a review by combining clustering technology and a large language model, effectively improving the efficiency and quality of document extraction.

[0097] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, the embodiment of this application provides another method for generating a literature review, such as Figure 2 As shown in the figure, it is a schematic diagram of the literature review generation method, which includes the following four steps:

[0098] Step 1: Clustering of documents.

[0099] Specifically, all input documents are classified according to a specific clustering algorithm to classify documents with similar or related contents into the same group to form several clusters. Each cluster represents a group of documents with similar contents, and a corresponding chapter will eventually be generated in the literature review. This grouping method can ensure that the content of the literature review has a good organizational structure, and each chapter revolves around a specific topic, thereby improving the overall coherence and logic of the review.

[0100] Step 2: Generation of chapter text.

[0101] Specifically, for each clustered text cluster, the large language model will be used to generate the specific content of the chapter. Technologies such as RAG (Retrieval-Augmented Generation) can be used. During the generation process, the large language model first determines the internal structure of each chapter, that is, divides it into several sections. Next, the model will automatically write the specific content of each chapter based on the title of each chapter and the content of the document. This process aims to extract key ideas from multiple sources to ensure that the generated content is both comprehensive and accurate.

[0102] Optionally, the generation of the chapter text includes the following steps:

[0103] Step 1: Chapter and paragraph division: First, based on the review topic, the titles and abstracts of similar documents contained in each text cluster, the internal structure of each chapter is generated and divided into several paragraphs. This process will reasonably divide the paragraphs according to the title and abstract of the document. For example, the topics covered after document clustering may involve multi-angle research perspectives or technical solutions. The model will reasonably group them according to these different perspectives, so that the paragraph division logic is clear and the theme is focused.

[0104] Step 2: Question generation and chapter text writing: Generate questions for each paragraph to ask the large language model to guide the generation of chapter content. During the writing process, the large language model will combine the content of multiple similar documents in the document cluster and gradually write the paragraph content. This process ensures that the generated paragraph content is closely related to the paragraph title and has a basis.

[0105] Step 3: Integration of multiple literature information: For each paragraph, the literature content of multiple documents is integrated. The title and abstract of the document are used as context input, and questions are generated through the question template to extract key information. During the generation process, the appropriate literature content is automatically screened and cited based on the relevance of the paragraph title. In addition, when the model generates a paragraph, it will attach the corresponding reference number after each sentence or point of view to ensure the authority and accuracy of the content.

[0106] Step 4: Output: Output paragraphs, with a length of about 300 words. Each paragraph will strictly follow the references of the literature to ensure that the content is coherent and meets scientific standards, avoiding any information that is inconsistent with the topic or causes confusion. In this process, the model will automatically remove irrelevant or repeated references to ensure that the information in each paragraph is accurate and complete.

[0107] Step 3: Generation of literature review.

[0108] Specifically, the chapter texts generated in steps one and two are compiled into a complete review article, so that the final literature review can systematically cover the multiple topics or research directions involved and show the connections and differences between different documents.

[0109] Step 4: Optimization of literature review.

[0110] Specifically, in order to further improve the quality of the review article, the literature review is refined and modified. This includes: Abstract generation: Generate a review summary using a large language model to ensure that the summary can accurately summarize the core content of the entire review. Title generation: Automatically generate an accurate title based on the content of the literature review. Structural optimization and content coherence check: Manually check and correct the generated literature review, further adjust the order and content of chapters and paragraphs, make the review text structure more reasonable, and make the logic more rigorous, ensure that the transition between chapters is natural and smooth, and the overall content is consistent to avoid duplication or omissions.

[0111] Further, as Figure 1 In a specific implementation of the method, the present application embodiment provides a literature review generating device 300, such as Figure 3 As shown, the device includes: an acquisition module 301, a first generation module 302, a second generation module 303 and a third generation module 304.

[0112] An acquisition module 301 is used to acquire review topics and multiple documents;

[0113] A first generating module 302 is used to generate multiple document clusters based on multiple documents, wherein each document cluster includes at least two similar documents whose similarity is greater than or equal to a preset similarity threshold;

[0114] The second generation module 303 is used to generate chapter text corresponding to each document cluster through a large language model based on the review topic and multiple document clusters;

[0115] The third generating module 304 is used to generate a literature review based on multiple chapter texts.

[0116] In a specific application scenario, the first generating module 302 is specifically used to:

[0117] Perform word segmentation preprocessing on each document to obtain multiple words corresponding to each document;

[0118] Convert multiple words in each document into embedding vectors and perform dimensionality reduction on the embedding vectors;

[0119] Based on the reduced-dimensional embedding vector corresponding to each document, multiple documents are clustered to generate multiple document clusters.

[0120] In a specific application scenario, the first generating module 302 is further used to:

[0121] Based on the reduced-dimensional embedding vector of each document, the similarity between any two documents is calculated using the distance measurement method;

[0122] Based on the similarity value between any two documents, multiple documents are clustered through a preset clustering algorithm to generate multiple document clusters.

[0123] In a specific application scenario, the second generation module 303 is specifically used to:

[0124] For any document cluster, obtain multiple document titles and multiple document abstracts of multiple similar documents included in the document cluster;

[0125] Based on the review topic, multiple document titles and multiple document abstracts, a chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text are generated through a large language model;

[0126] Based on the review topic, chapter title, multiple paragraph titles and at least two similar documents contained in the document cluster, the chapter text corresponding to the document cluster is generated through a large language model.

[0127] In a specific application scenario, the second generating module 303 is further used to:

[0128] For any document cluster, the review topic, multiple document titles and multiple document abstracts are filled into the first preset question template to generate the first question;

[0129] A question is asked to the large language model based on the first question, so that the large language model outputs the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text based on the review topic, multiple document titles and multiple document abstracts.

[0130] In a specific application scenario, the second generating module 303 is further used to:

[0131] For any document cluster, the review topic, chapter title and multiple paragraph titles are filled into the second preset question template to generate multiple second questions corresponding to multiple paragraphs contained in the chapter text;

[0132] Asking questions to the large language model based on the multiple second questions enables the large language model to output paragraph content corresponding to each paragraph based on the document content of at least two similar documents in the document cluster;

[0133] The chapter title, multiple paragraph titles, and multiple paragraph contents of multiple paragraphs are aggregated to generate the chapter text corresponding to the document cluster.

[0134] In a specific application scenario, the device also includes:

[0135] The fourth generation module is used to generate a review title and a review summary based on the content of the literature review through a large language model.

[0136] The document review generation device 300 provided in the embodiment of the present application first performs cluster analysis on multiple documents, and divides documents with similar contents into the same set to form multiple document clusters, so that a relatively small document set is focused on in the subsequent processing process, solving the technical problem of limited input length of the large language model. Afterwards, the large language model is used to generate the corresponding chapter text for each document cluster, and the corresponding chapter text is generated based on the document cluster composed of similar documents, and finally the document review is summarized to ensure the relevance and consistency of the chapter text content, so that the chapter paragraphs are more logical, avoid information isolation, and ensure that the document review comprehensively covers related research under the review topic. The present application solves the problem of limited input length of the large language model when generating a review by combining clustering technology and a large language model, and effectively improves the efficiency and quality of document extraction.

[0137] In an exemplary embodiment, the present application further provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor is used to execute the program stored in the memory and execute the method for generating a literature review in the above embodiment.

[0138] In an exemplary embodiment, the present application also provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the literature review generating method are implemented.

[0139] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present application can be implemented by hardware, or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile readable storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for enabling an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present application.

[0140] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present application.

[0141] Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple units.

[0142] The above application serial numbers are for description only and do not represent the advantages or disadvantages of the implementation scenarios.

[0143] The above disclosure only discloses several specific implementation scenarios of the present application. However, the present application is not limited thereto, and any changes that can be conceived by technicians in this field should fall within the scope of protection of the present application.

Claims

1. A method for generating a literature review, characterized in that: include: Obtain review topics and multiple papers; Performing clustering processing on the plurality of documents to generate a plurality of document clusters, wherein each document cluster includes at least two similar documents; Based on the review topic and the multiple document clusters, generating chapter text corresponding to each document cluster through a large language model; Generate a literature review based on multiple chapter texts.

2. The method according to claim 1, characterized in that The step of clustering the multiple documents to generate multiple document clusters specifically includes: Perform word segmentation preprocessing on each document to obtain multiple words corresponding to each document; Convert the multiple words in each document into embedding vectors, and perform dimensionality reduction processing on the embedding vectors; Based on the reduced-dimensional embedding vector corresponding to each document, the multiple documents are clustered to generate the multiple document clusters.

3. The method according to claim 2, characterized in that The step of clustering the multiple documents based on the reduced-dimensional embedding vector corresponding to each document to generate the multiple document clusters specifically includes: Based on the reduced-dimensional embedding vector of each document, the similarity between any two documents is calculated using the distance measurement method; Based on the similarity value between any two documents, a plurality of documents are clustered by a preset clustering algorithm to generate the plurality of document clusters.

4. The method according to claim 1, characterized in that: The step of generating the chapter text corresponding to each document cluster by using a large language model based on the review topic and the multiple document clusters specifically includes: For any document cluster, obtain multiple document titles and multiple document abstracts of multiple similar documents included in the document cluster; Based on the review topic, multiple document titles and multiple document abstracts, generating, by means of a large language model, a chapter title of a chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text; Based on the review topic, chapter title, multiple paragraph titles and at least two similar documents included in the document cluster, a chapter text corresponding to the document cluster is generated through a large language model.

5. The method according to claim 4, characterized in that The step of generating the chapter title of the chapter text corresponding to the document cluster and the multiple paragraph titles of the multiple paragraphs contained in the chapter text based on the review topic, the multiple document titles and the multiple document abstracts through the large language model specifically includes: For any document cluster, the review topic, multiple document titles and multiple document abstracts are filled into the first preset question template to generate the first question; A question is asked to the large language model based on the first question, so that the large language model outputs the chapter title of the chapter text corresponding to the document cluster and multiple paragraph titles of multiple paragraphs contained in the chapter text based on the review topic, multiple document titles and multiple document abstracts.

6. The method according to claim 4, characterized in that The step of generating the chapter text corresponding to the document cluster through a large language model based on the review topic, chapter title, multiple paragraph titles and at least two similar documents included in the document cluster specifically includes: For any document cluster, the review topic, chapter title and multiple paragraph titles are filled into the second preset question template to generate multiple second questions corresponding to multiple paragraphs contained in the chapter text; Asking questions to the large language model based on the multiple second questions enables the large language model to output paragraph content corresponding to each paragraph based on the document content of at least two similar documents in the document cluster; The chapter title, multiple paragraph titles, and multiple paragraph contents of multiple paragraphs are aggregated to generate the chapter text corresponding to the document cluster.

7. The method according to any one of claims 1 to 6, characterized in that After generating the literature review based on multiple chapter texts, the following is also included: Generate review titles and review summaries based on the content of the literature review through a large language model.

8. A device for generating a literature review, characterized in that: include: The acquisition module is used to obtain review topics and multiple documents; A first generating module is used to perform clustering processing on the plurality of documents to generate a plurality of document clusters, wherein each document cluster includes at least two similar documents; A second generation module is used to generate a chapter text corresponding to each document cluster through a large language model based on the review topic and the multiple document clusters; The third generation module is used to generate a literature review based on multiple chapter texts.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When a processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images