Document analysis assistant system, method, electronic device, and medium
By integrating a generative pre-trained language model into an academic website, a literature analysis assistant system has been developed, addressing the shortcomings of traditional academic websites in terms of intelligence. This system enables semantic retrieval and multi-turn natural language interaction, thereby improving user experience and analytical accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TONGFANG KNOWLEDGE DIGITAL PUBLISHING TECH CO LTD
- Filing Date
- 2023-09-27
- Publication Date
- 2026-08-04
AI Technical Summary
Traditional academic websites have low levels of intelligence, lack semantic retrieval, inductive analysis capabilities, and interactive design, and cannot effectively utilize the complex relationships and semantic information between documents, resulting in a poor user experience.
This paper provides a literature analysis assistant system that integrates with academic websites through generative pre-trained language models. The system includes a summary analysis module, a full-text analysis module, and a source tracing module, which improves the intelligence level of academic websites and supports semantic retrieval and multi-turn natural language interaction.
It has improved the intelligence level of academic websites, enhanced semantic retrieval and interactive capabilities, improved user experience, and increased the accuracy and readability of literature analysis.
Smart Images
Figure CN117407577B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a document analysis assistant system, method, electronic device, and medium. Background Technology
[0002] Most academic websites are currently built on traditional information retrieval technologies, resulting in a generally low level of intelligence and a poor user experience. This is mainly reflected in the following aspects: (1) Traditional academic websites do not support semantic search and natural language questioning. Traditional academic websites contain a vast amount of academic resources, and traditional rule-based and feature engineering methods often struggle to capture the complex relationships and semantic information between these resources. Traditional academic websites typically employ keyword and keyword combination searches, making it difficult to effectively utilize semantic information.
[0003] (2) Traditional academic websites lack the ability to summarize and analyze. Traditional academic websites can only provide resource lists for references and citations of existing documents, without further categorization or summarization. Furthermore, readers are required to summarize the content of the documents themselves, hindering rapid reading.
[0004] (3) Traditional academic websites lack interactive design Traditional academic websites only provide services such as searching and reading, and their interaction with users mainly relies on information input. They cannot engage in multi-turn dialogues with users using natural language.
[0005] Given the problems faced by traditional academic websites, upgrading and transforming them intelligently using artificial intelligence technologies, such as large language models, has become an inevitable requirement of the current wave of AI applications. However, how to link traditional academic websites with large language models has become a pressing technical problem that needs to be solved. Summary of the Invention
[0006] In view of the above situation, embodiments of this application provide a literature analysis assistant system, method, electronic device and medium, which aim to solve the above problems or at least partially solve the above problems.
[0007] The first aspect provides a literature analysis assistant system, which acts as a browser plugin to integrate a generative pre-trained language model into a literature retrieval page. The system includes: Summary analysis module, full-text analysis module, and source tracing module; The summary analysis module is used to obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain literature summary analysis information; The full-text analysis module is used to obtain the full-text information of the document on the current page and provide the full-text information of the document to the generative pre-trained language model to obtain the full-text analysis information of the document. The source tracing module is used to obtain the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model, and to determine whether the literature summary analysis information or the literature full-text analysis information is a model illusion based on the literature knowledge base.
[0008] Preferably, the summary analysis module includes a question recommendation submodule, a classification summary submodule, and a free-response question and answer submodule; The question recommendation submodule is used to obtain the search terms and literature search results entered by the user on the current page, and to provide the search terms and literature search results to the generative pre-trained language model to obtain multiple questions related to the search results; The classification and summary submodule is used to obtain the literature topics of the search results on the current page and provide the literature topics to the generative pre-trained language model to obtain the research direction categories and summary analysis results of the literature search results. The free-response question-and-answer submodule is used to obtain the literature topics and user questions of the search results on the current page, and to provide the literature topics and user questions to the generative pre-trained language model to obtain analysis results related to the literature search results.
[0009] Preferably, the full-text analysis module includes a slicing module, a vector module, a storage module, and an analysis module; The slicing module is used to obtain the full text of the document on the current page and cut the full text of the document into multiple text segments based on the pre-set paragraph segmentation rules; The vector module is used to convert the text fragment into a first vector based on a pre-trained model; The storage module is used to store the first vector; The analysis module receives user questions, segments the user questions into multiple question fragments based on the slicing module, converts the question fragments into second vectors based on the vector module, queries a first vector related to the second vector in the storage module, returns a text fragment corresponding to the first vector, provides the returned text fragment and the user questions to the generative pre-trained language model to obtain analysis results, and displays the analysis results to the user.
[0010] Preferably, the full-text analysis module includes a key point sub-module, a related work sub-module, a research method sub-module, and a research conclusion sub-module; The key point submodule is used to provide the analysis module with a first user question so that the key points of the literature research on the current page can be obtained through the generative pre-trained language model. The relevant work submodule is used to provide the analysis module with a second user question, so as to obtain the relevant work of the literature research on the current page through the generative pre-trained language model; The research method submodule is used to provide the analysis module with a third user question so as to obtain the research method of the literature research on the current page through the generative pre-trained language model; The research conclusions submodule is used to provide the analysis module with a fourth user question, so as to obtain the research conclusions of the literature research on the current page through the generative pre-trained language model.
[0011] Preferably, the traceability module includes a traceability acquisition module, a traceability query module, and a traceability judgment module; The source tracing module is used to obtain the literature summary analysis information or the full-text analysis information of the literature generated by the generative pre-trained language model; The source tracing and query module is used to search for documents related to the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model in the pre-established document knowledge base. The source tracing and judgment module is used to determine, based on the search results, whether the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion.
[0012] Preferably, the source tracing judgment module is used to determine that the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model is a model illusion when no documents related to the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model are found in the document knowledge base. The source tracing judgment module is also used to determine that the document summary analysis information or document full-text analysis information generated by the generative pre-trained language model is not a model illusion when a document related to the document summary analysis information or document full-text analysis information generated by the generative pre-trained language model is found in the document knowledge base.
[0013] Preferably, the system further includes a triggering module; The triggering module is used to obtain the type of the current page, and based on the pre-established mapping relationship between the page type and the summary analysis module and the full-text analysis module, triggers the summary analysis module or the full-text analysis module corresponding to the current page type.
[0014] Secondly, a method for searching using a literature analysis assistant is provided, including: Obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain literature summary and analysis information; Obtain the full-text information of the document on the current page and provide the full-text information of the document to the generative pre-trained language model to obtain full-text analysis information of the document; Obtain the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model, and determine whether the literature summary analysis information or the literature full-text analysis information is a model illusion based on the literature knowledge base.
[0015] Thirdly, an electronic device is provided, comprising: Processor; and A memory is configured to store computer-executable instructions, which, when executed, cause the processor to perform the retrieval method described above using a document analysis assistant.
[0016] Fourthly, a computer-readable storage medium is provided that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the method of using a document analysis assistant for retrieval as described above.
[0017] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: The literature analysis assistant provided in this application can function as a browser plugin, integrating a generative pre-trained language model into a literature retrieval page. It obtains the page information of the current page and provides this information to the generative pre-trained language model to generate literature analysis information. This application links traditional academic websites with generative pre-trained language models, improving the intelligence level of traditional academic websites and enhancing the user experience. Attached Figure Description
[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of an application environment of a document analysis assistant system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the summary analysis module interface in one embodiment of the present invention; Figure 3 This is a schematic diagram of the full-text analysis module interface in one embodiment of the present invention; Figure 4This is a schematic diagram of the search page interface in one embodiment of the present invention; Figure 5 This is a flowchart illustrating a retrieval method using a Chinese literature analysis assistant according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0020] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.
[0021] The literature analysis assistant system provided in this embodiment of the invention can be applied to, for example... Figure 1 In this application environment, the device communicates with the server via a network. The literature analysis assistant system is installed on the device as a browser plugin, connecting the server's generative pre-trained language model to the device's literature search page.
[0022] Specifically, the system includes: a summary analysis module, a full-text analysis module, and a source tracing module; the summary analysis module is used to obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain document summary analysis information; the full-text analysis module is used to obtain the full-text information of the documents on the current page and provide the full-text information to the generative pre-trained language model to obtain document full-text analysis information; the source tracing module is used to obtain the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model, and determine whether the document summary analysis information or the document full-text analysis information is a model illusion based on the document knowledge base.
[0023] It should be noted that generative pre-trained language models have the following characteristics: First, they are large in scale, with network parameters reaching tens of billions, hundreds of billions, or even more; second, they are general, meaning they are not limited to specific problems or domains; and third, they are emergent, meaning they generate unexpected new capabilities.
[0024] It's important to clarify that "model illusion" refers to content generated by a generative pre-trained language model that isn't based on any real-world data, but rather is a product of the model's own imagination. For example, when faced with a user's question, a generative pre-trained language model might fabricate seemingly authoritative and accurate false information, such as creating non-existent books and research reports, fake academic papers, or fake legal citations. This false information can exist in the form of text, images, audio, or video. The literature analysis assistant provided in this application can function as a browser plugin, integrating a generative pre-trained language model into a literature retrieval page. It obtains the page information of the current page and provides this information to the generative pre-trained language model to generate literature analysis information. This application links traditional academic websites with generative pre-trained language models, improving the intelligence level of traditional academic websites and enhancing the user experience.
[0025] Specifically, the system also includes a triggering module; the triggering module is used to obtain the type of the current page, and based on the pre-established mapping relationship between the page type and the summary analysis module and the full-text analysis module, trigger the summary analysis module (e.g., ...) corresponding to the current page type. Figure 2 (as shown) or the full-text analysis module (such as Figure 3 (As shown).
[0026] Understandably, academic websites typically display three types of pages when conducting searches: a list page of search results, an overview page of search results, and a full-text page of search results. A list page of search results (such as...) Figure 4 The search results (as shown) are displayed in a list format to users, including topic items, list items, and related search terms. The topic items display all topics covered by the search results and categorize them. The list items display all search results, which can be paginated when multiple results are available. Related search terms display commonly used search terms related to the search term. The overview page displays the abstract, keywords, references, cited documents, and citation network of the search result. The full-text page displays the full text of the search result. The search process generally includes: entering search terms, performing a search, the academic website displaying a list of search results, selecting a search result in the overview page, and then selecting "HTML Reading" in the overview page to access the full-text page.
[0027] In this embodiment, when the triggering module obtains that the current page type is a search results list page or a search results overview page, it triggers the summary analysis module to start and displays the results to the user, such as... Figure 2The interface shown is as follows. When the triggering module determines that the current page type is a full-text page of the search results, it triggers the full-text analysis module to start and displays the results to the user. Figure 3 The interface shown.
[0028] Specifically, such as Figure 2 As shown, the summary analysis module includes a question recommendation submodule, a classification summary submodule, and a free question and answer submodule.
[0029] It should be noted that the summary analysis module is used to retrieve the text on the current page.
[0030] The question recommendation submodule is used to obtain the user-inputted search terms and document search results on the current page, and provide these to the generative pre-trained language model to generate multiple questions related to the search results. Specifically, such as... Figure 2 The interface shown includes a question recommendation control. Clicking "Question Recommendation" allows users to submit a question to the generative pre-trained language model, prompting them to "recommend a few related questions." Simultaneously, the question recommendation submodule retrieves information such as... Figure 4 The current page displays the user's input search terms, the text in the topic items of the search results page, and the text in the related search terms. These are simultaneously provided to a generative pre-trained language model. The generative pre-trained language model generates multiple questions related to the search results based on these texts and displays them to the user. The user can click on any of these questions to ask the generative pre-trained language model a question. In this embodiment, the number of related questions displayed to the user can be preset.
[0031] The classification and summarization submodule is used to obtain the literature topics of the search results on the current page and provide these literature topics to the generative pre-trained language model to obtain the research direction categories and summary analysis results of the literature search results. Specifically, for example... Figure 2 The interface shown includes a classification and summary control. Clicking "Classify and Summarize" provides the generative pre-trained language model with the question "Please classify and summarize the search results on this page." Simultaneously, the classification and summary submodule retrieves data such as... Figure 4 The text of the literature topics in the list items of the search results page shown on the current page is provided to the generative pre-trained language model. The corresponding questions and the text of the literature topics in the list items are also provided to the generative pre-trained language model. The generative pre-trained language model generates a classification summary of the literature topics in the list items based on the corresponding questions and the text of the literature topics in the list items, and displays it to the user.
[0032] The free-response question-answering submodule is used to obtain the document topics and user questions from the search results of the current page, and provides these topics and user questions to the generative pre-trained language model to obtain analysis results related to the document search results. Specifically, such as... Figure 2 The interface shown includes a dialog box where users can enter their questions and click "send". Figure 2 The paper airplane icon (referring to a Chinese character) provides user questions to a generative pre-trained language model, while simultaneously allowing the free question-and-answer submodule to obtain information such as... Figure 4 The text of the literature topics in the list items of the search results page shown on the current page is provided to the generative pre-trained language model along with the user's question and the text of the literature topics in the list items. The generative pre-trained language model generates analysis results related to the literature search results based on the user's question and the text of the literature topics in the list items as the answer to the question, and displays it to the user.
[0033] Preferably, the summary analysis module also includes a topic acquisition submodule, which is used to acquire the literature topics of the search results on the current page and assign the acquired literature topics to a dialog box. Users can edit the literature topics and provide user questions in the dialog box, which are then provided to the generative pre-trained language model. The generative pre-trained language model generates analysis results related to the literature search results based on the user questions and the edited text of the literature topics as the answers to the questions, and displays them to the user.
[0034] The summary analysis module also includes a reference submodule and a cited literature submodule. When the page type is a search results overview page, the trigger module is used to trigger the reference submodule and the cited literature submodule, and in... Figure 2 The interface shown displays the "References" and "Cited References" controls, while hiding the "Recommended Questions" and "Category Summary" controls.
[0035] The References submodule is used to retrieve the topic text of the references from the search results on the current page and provide this topic text to the generative pre-trained language model to obtain the classification and summary analysis results of the references. Specifically, clicking "References" provides the generative pre-trained language model with the corresponding question, "Please classify and summarize the references in this article." The generative pre-trained language model generates classification and summary analysis results based on the corresponding question and displays them to the user.
[0036] The Citations submodule is used to obtain the topic text of the citations in the search results of the current page, and provide the topic text of the citations to the generative pre-trained language model to obtain the classification and summary analysis results of the citations. Specifically, by clicking "Citations," users can provide the corresponding question "Please classify and summarize the citations in this article" to the generative pre-trained language model. The generative pre-trained language model generates classification and summary analysis results based on the corresponding question and displays them to the user.
[0037] The summary analysis module also includes a literature recommendation module. This module retrieves the text of search terms entered by the user on academic websites and the text of the topics of documents read, and provides these texts to the generative pre-trained language model to generate relevant recommended literature, which is then displayed to the user. The user can click on any of these recommended documents to ask questions to the generative pre-trained language model.
[0038] Preferably, the full-text analysis module includes a slicing module, a vector module, a storage module, and an analysis module. The slicing module is used to obtain the full-text text of the document on the current page and, based on pre-set paragraph segmentation rules, cut the full-text text into multiple text segments. The vector module is used to convert the text segments into first vectors based on a pre-trained model. The storage module is used to store the first vectors. The analysis module is used to receive user questions, cut the user questions into multiple question segments based on the slicing module, convert the question segments into second vectors based on the vector module, query the storage module for first vectors related to the second vectors, return the text segments corresponding to the first vectors, provide the returned text segments and user questions to the generative pre-trained language model to obtain analysis results, and display the analysis results to the user.
[0039] Specifically, in this embodiment, the slicing module obtains the full text of the document on the current page and, according to the set size or paragraph segmentation rules, divides the full text of the document into text segments of the same size or different sizes (e.g., divided by paragraphs).
[0040] Specifically, pre-trained models refer to Microsoft E5 models, BERT models, or text2vec models, which convert each text segment into a first vector with a fixed dimension (e.g., 768 dimensions).
[0041] Specifically, the generated first vectors are stored in the storage module.
[0042] When the analysis module receives a user question, it segments the user question into multiple question fragments based on the slicing module according to a set size or paragraph segmentation rule; it converts each question fragment into a second vector based on the vector module, the second vector having the same dimension as the first vector; it queries the storage module for the first vector related to the second vector and returns the text fragment corresponding to the first vector; it provides the returned text fragment and the user question to the generative pre-trained language model to obtain the analysis results, and displays the analysis results to the user.
[0043] Specifically, the full-text analysis module includes a key points sub-module, a related work sub-module, a research methods sub-module, and a research conclusions sub-module. For example... Figure 3 The interface shown includes controls for "Key Points", "Related Work", "Research Methods", and "Research Conclusions".
[0044] The key point submodule is used to provide the analysis module with a first user question, so that the generative pre-trained language model can be used to obtain the key points of the literature research on the current page. Specifically, by clicking "Key Points," the user can provide the analysis module with the corresponding first user question, "What are the key points of this paper?" When the analysis module receives the first user question, it uses the slicing module to cut the first user question into multiple question fragments according to the set size or paragraph segmentation rules; it uses the vector module to convert each question fragment into a second vector, which has the same dimension as the first vector; it uses the storage module to query the first vector related to the second vector and returns the text fragment corresponding to the first vector; it provides the returned text fragment and the first user question to the generative pre-trained language model to obtain the analysis results, and then displays the analysis results to the user.
[0045] The related work submodule is used to provide the analysis module with a second user question, so that the generative pre-trained language model can be used to obtain the related work of the literature research on the current page. Specifically, by clicking "Related Work", the user can provide the analysis module with the corresponding second user question, "What is the related work of this paper?" When the analysis module receives the second user question, it uses the slicing module to cut the second user question into multiple question fragments according to the set size or paragraph segmentation rules; it uses the vector module to convert each question fragment into a second vector, which has the same dimension as the first vector; it uses the storage module to query the first vector related to the second vector and returns the text fragment corresponding to the first vector; it provides the returned text fragment and the second user question to the generative pre-trained language model to obtain the analysis results, and then displays the analysis results to the user.
[0046] The research method submodule provides a third user question to the analysis module, enabling the generative pre-trained language model to derive the research methods of the literature on the current page. Specifically, clicking "Research Method" provides the analysis module with the corresponding third user question, "What is the research method of this paper?" Upon receiving the third user question, the analysis module, based on the slicing module, divides the third user question into multiple question fragments according to a set size or paragraph segmentation rule; based on the vector module, each question fragment is converted into a second vector, which has the same dimension as the first vector; the storage module queries for a first vector related to the second vector and returns the text fragment corresponding to the first vector; the returned text fragment and the third user question are provided to the generative pre-trained language model to obtain the analysis results, which are then displayed to the user.
[0047] The research conclusion submodule is used to provide the analysis module with a fourth user question, so that the generative pre-trained language model can be used to obtain the research conclusions of the literature research on the current page. Specifically, by clicking "Research Conclusions," the user can provide the analysis module with the corresponding fourth user question, "What are the research conclusions of this paper?" When the analysis module receives the fourth user question, it uses the slicing module to cut the fourth user question into multiple question fragments according to the set size or paragraph segmentation rules; it uses the vector module to convert each question fragment into a second vector, which has the same dimension as the first vector; it uses the storage module to query the first vector related to the second vector and returns the text fragment corresponding to the first vector; it provides the returned text fragment and the fourth user question to the generative pre-trained language model to obtain the analysis results, and then displays the analysis results to the user.
[0048] In this embodiment, by providing the user's question and related literature text to the generative pre-trained language model, the accuracy and readability of the analysis results generated by the generative pre-trained language model can be improved, making it easier for users to understand and enhancing the user experience.
[0049] like Figure 2 and Figure 3 As shown, a "magnifying glass" control is provided below the analysis results generated by the generative pre-trained language model. This control allows users to activate the source tracing module. The source tracing module includes a source acquisition module, a source query module, and a source judgment module.
[0050] The source tracing module is used to acquire the literature summary analysis information or the full-text analysis information of the literature generated by the generative pre-trained language model. Each analysis result generated by the generative pre-trained language model has an interface below it that connects to the source tracing module, which in turn connects the source tracing module and a "magnifying glass" control. Each analysis result generated by the generative pre-trained language model has a "magnifying glass" control below it, which allows for source tracing analysis of the current analysis result.
[0051] The source tracing and query module is used to search for documents related to the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model, based on a pre-established document knowledge base.
[0052] It should be noted that the literature knowledge base is a collection of known entity content, such as various laws promulgated by the state, various terms in the field of engineering and technology, conceptual knowledge elements extracted from academic databases, knowledge points in various reference books included in CNKI, etc.
[0053] The source tracing and judgment module is used to determine, based on the search results, whether the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion.
[0054] Specifically, the source tracing judgment module is used to determine that the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion when no literature related to the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is found in the literature knowledge base.
[0055] Specifically, the source tracing judgment module is also used to determine that the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model is not a model illusion when a document related to the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model is found in the document knowledge base.
[0056] In this embodiment, the accuracy of the analysis results generated by the generative pre-trained language model is improved by setting up a source tracing module.
[0057] This embodiment's full-text analysis module also includes a free-response question-and-answer module. Users can drag and drop text that needs to be translated, rewritten, or have keywords extracted into the module. Figure 3In the dialog box shown, a user is given a corresponding question. The free-response question-and-answer module will retrieve the text dragged into the dialog box and the user's question, and provide them to the generative pre-trained language model to obtain the analysis results. For example, if a user drags a section of text from the full text into the text box and gives the question "Translate the above text into Chinese," the free-response question-and-answer module will retrieve the text dragged into the dialog box and the user's question "Translate the above text into Chinese," and provide them to the generative pre-trained language model. The generative pre-trained language model will then generate a Chinese translation of the text dragged into the dialog box and display it to the user. In this embodiment, the literature analysis assistant overcomes the language limitations of researchers, achieving the goal of quickly reading and understanding foreign language literature.
[0058] In this embodiment, multiple generative pre-trained language models can be configured. This embodiment includes a first generative pre-trained language model and a second generative pre-trained language model. The first generative pre-trained language model is communicatively connected to the summary analysis module, and the second generative pre-trained language model is communicatively connected to the full-text analysis module. Using a multi-model fusion strategy, specific pre-trained large language models can be selected to support different specific tasks, avoiding the limitations of a single model. Furthermore, the integrated generative pre-trained language models can be switched between self-developed models and open-source commercially available models as needed.
[0059] Furthermore, in this embodiment, the literature analysis assistant system includes a prompting engineering template. When the literature analysis assistant system receives a user question, it provides the user question to the prompting engineering template, and the prompting engineering template provides the question to the generative pre-trained language model.
[0060] It's important to note that the "Prompt" in the prompting template is a piece of text or a question provided to the generative pre-trained language model to guide it in generating specific types of text or answers. The Prompt can be a sentence, a question, an article, or a topic, and it can guide the generative pre-trained language model to generate information or answers related to the Prompt.
[0061] For example, in this embodiment, the prompt in the prompt engineering template is "As a practitioner in this field", when the question is "What are the research conclusions of this paper?", the question provided to the generative pre-trained language model is "As a practitioner in this field, what are the research conclusions of this paper?"
[0062] Prompts can improve the generation quality and efficiency of generative pre-trained language models, reducing the probability of generating meaningless or irrelevant content. Simultaneously, prompts can also be used for comparative experiments, testing the model's generation capabilities and performance with different prompts.
[0063] In this embodiment, the generative pre-trained language model can be guided to generate the desired output by designing, optimizing, and evaluating the input prompt, that is, finding the best way to ask questions to the generative model in order to obtain the most useful and accurate answers.
[0064] The literature analysis assistant system provided in this application, without disrupting the original academic website's element layout, functions as a browser plugin to integrate generative pre-trained language models into the existing academic website, thereby enhancing the website's intelligence level. Furthermore, it emphasizes interactivity, expanding the previously relatively one-way research activity into two-way communication, thus improving the user experience.
[0065] This embodiment also provides a method for searching using a literature analysis assistant, including: Obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain literature summary and analysis information; Obtain the full-text information of the document on the current page and provide the full-text information of the document to the generative pre-trained language model to obtain full-text analysis information of the document; Obtain the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model, and determine whether the literature summary analysis information or the literature full-text analysis information is a model illusion based on the literature knowledge base.
[0066] Specifically, the method includes: S01: Get the type of the current page; S02: Based on the pre-established mapping relationship between page types and the summary analysis module and the full-text analysis module, trigger the summary analysis module or the full-text analysis module corresponding to the current page type; S03: Obtain the search results information of the current page or the full text information of the document on the current page, and provide the search results information of the current page or the full text information of the document to the generative pre-trained language model to obtain the document summary analysis information or the full text analysis information of the document.
[0067] In one embodiment, step S03 specifically includes: The system obtains the user-inputted search terms and document search results on the current page, and provides these search terms and results to the generative pre-trained language model to generate multiple questions related to the search results.
[0068] In one embodiment, step S03 further includes: The document topics of the search results on the current page are obtained and provided to the generative pre-trained language model to obtain the research direction categories and summary analysis results of the document search results.
[0069] In one embodiment, step S03 further includes: The document topics and user questions of the search results on the current page are obtained, and the document topics and user questions are provided to the generative pre-trained language model to obtain analysis results related to the document search results.
[0070] In one embodiment, step S03 further includes: Get the full text of the document on the current page, and cut the full text into multiple text segments based on the pre-set paragraph segmentation rules; Based on the pre-trained model, the text fragment is converted into a first vector; Store the first vector to the storage module; The system receives user questions, segments them into multiple question fragments based on pre-set paragraph segmentation rules, converts the question fragments into second vectors based on a pre-trained model, queries a first vector related to the second vector in the storage module, returns the text fragment corresponding to the first vector, provides the returned text fragment and the user question to the generative pre-trained language model to obtain analysis results, and displays the analysis results to the user.
[0071] In one embodiment, step S03 further includes: Obtain the literature summary analysis information or the full-text analysis information of the literature generated by the generative pre-trained language model; Based on a pre-established literature knowledge base, search the literature knowledge base for literature related to the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model. Based on the search results, determine whether the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion.
[0072] In one embodiment, step S03 further includes: When no literature related to the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model is found in the literature knowledge base, it is determined that the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model is a model illusion. When a document related to the summary analysis information or full-text analysis information of the document generated by the generative pre-trained language model is found in the document knowledge base, it is determined that the summary analysis information or full-text analysis information of the document generated by the generative pre-trained language model is not a model illusion.
[0073] The literature analysis assistant system provided in this application, without disrupting the original academic website's element layout, functions as a browser plugin to integrate generative pre-trained language models into the existing academic website, thereby enhancing the website's intelligence level. Furthermore, it emphasizes interactivity, expanding the previously relatively one-way research activity into two-way communication, thus improving the user experience.
[0074] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to perform the method of using a document analysis assistant for retrieval in any of the above embodiments.
[0075] This embodiment also provides an electronic device, such as... Figure 6 As shown, Figure 6 The illustrated electronic device 400 includes a processor 401 and a memory 403. The processor 401 and the memory 403 are connected, for example, via a bus 402. Optionally, the electronic device 400 may also include a transceiver 404. It should be noted that in practical applications, the transceiver 404 is not limited to one type, and the structure of this electronic device 400 does not constitute a limitation on the embodiments of this application.
[0076] The processor 401 can be a GPU (Graphics Processing Unit) or other types of processors, and this embodiment does not limit this.
[0077] Bus 402 may include a pathway for transmitting information between the aforementioned components. Bus 402 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 402 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.
[0078] The memory 403 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.
[0079] The memory 403 stores computer program code that executes the scheme of this application, and its execution is controlled by the processor 401. The processor 401 executes the computer program code stored in the memory 403 to implement the content shown in the foregoing method embodiments.
[0080] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0081] Based on the same inventive concept, this application also provides a storage medium storing a computer program, wherein the computer program is configured to execute the method of using the document analysis assistant for retrieval in any of the above embodiments when running.
[0082] Those skilled in the art will clearly understand that the specific working process of the systems, devices, and modules described above can be referred to the corresponding process in the foregoing method embodiments. For the sake of brevity, it will not be repeated here.
[0083] Those skilled in the art will understand that the technical solution of this application, or all or part of it, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several program instructions to cause an electronic device (e.g., a personal computer, server, or network device) to execute all or part of the steps of the methods described in the embodiments of this application when running the program instructions. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.
[0084] Alternatively, all or part of the steps of the foregoing method embodiments can be implemented by hardware (such as electronic devices like personal computers, servers, or network devices) associated with program instructions. The program instructions can be stored in a computer-readable storage medium. When the program instructions are executed by the processor of the electronic device, the electronic device executes all or part of the steps of the methods described in the embodiments of this application.
[0085] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that within the spirit and principles of this application, modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the corresponding technical solutions to leave the protection scope of this application.
Claims
1. A document analysis assistant system, characterized in that, The literature analysis assistant system, as a browser plugin, integrates a generative pre-trained language model into the literature retrieval page. The system includes: Summary analysis module, full-text analysis module, and source tracing module; The summary analysis module is used to obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain literature summary analysis information; The full-text analysis module is used to obtain the full-text information of the document on the current page and provide the full-text information of the document to the generative pre-trained language model to obtain the full-text analysis information of the document. The source tracing module is used to obtain the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model, and to determine whether the literature summary analysis information or the literature full-text analysis information is a model illusion based on the literature knowledge base. The source tracing module includes a source acquisition module, a source query module, and a source judgment module; The source tracing module is used to obtain the literature summary analysis information or the full-text analysis information of the literature generated by the generative pre-trained language model; The source tracing and query module is used to search for documents related to the document summary analysis information or the document full-text analysis information generated by the generative pre-trained language model in the pre-established document knowledge base. The source tracing and judgment module is used to determine, based on the search results, whether the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion.
2. The system according to claim 1, characterized in that, The summary and analysis module includes a question recommendation submodule, a classification summary submodule, and a free-response question and answer submodule; The question recommendation submodule is used to obtain the search terms and literature search results entered by the user on the current page, and to provide the search terms and literature search results to the generative pre-trained language model to obtain multiple questions related to the search results; The classification and summary submodule is used to obtain the literature topics of the search results on the current page and provide the literature topics to the generative pre-trained language model to obtain the research direction categories and summary analysis results of the literature search results. The free-response question-and-answer submodule is used to obtain the literature topics and user questions of the search results on the current page, and to provide the literature topics and user questions to the generative pre-trained language model to obtain analysis results related to the literature search results.
3. The system according to claim 1, characterized in that, The full-text analysis module includes a slicing module, a vector module, a storage module, and an analysis module; The slicing module is used to obtain the full text of the document on the current page and cut the full text of the document into multiple text segments based on the pre-set paragraph segmentation rules; The vector module is used to convert the text fragment into a first vector based on a pre-trained model; The storage module is used to store the first vector; The analysis module receives user questions, segments the user questions into multiple question fragments based on the slicing module, converts the question fragments into second vectors based on the vector module, queries a first vector related to the second vector in the storage module, returns a text fragment corresponding to the first vector, provides the returned text fragment and the user questions to the generative pre-trained language model to obtain analysis results, and displays the analysis results to the user.
4. The system according to claim 3, characterized in that, The full-text analysis module includes a key points sub-module, a related work sub-module, a research methods sub-module, and a research conclusions sub-module. The key point submodule is used to provide the analysis module with a first user question so that the key points of the literature research on the current page can be obtained through the generative pre-trained language model. The relevant work submodule is used to provide the analysis module with a second user question, so as to obtain the relevant work of the literature research on the current page through the generative pre-trained language model; The research method submodule is used to provide the analysis module with a third user question so as to obtain the research method of the literature research on the current page through the generative pre-trained language model; The research conclusions submodule is used to provide the analysis module with a fourth user question, so as to obtain the research conclusions of the literature research on the current page through the generative pre-trained language model.
5. The system according to claim 1, characterized in that, The source tracing judgment module is used to determine that the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion when no literature related to the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is found in the literature knowledge base. The source tracing judgment module is also used to determine that the document summary analysis information or document full-text analysis information generated by the generative pre-trained language model is not a model illusion when a document related to the document summary analysis information or document full-text analysis information generated by the generative pre-trained language model is found in the document knowledge base.
6. The system according to claim 1, characterized in that, The system also includes a triggering module; The triggering module is used to obtain the type of the current page, and based on the pre-established mapping relationship between the page type and the summary analysis module and the full-text analysis module, triggers the summary analysis module or the full-text analysis module corresponding to the current page type.
7. A method for searching using a literature analysis assistant, characterized in that, include: Obtain the search results information of the current page and provide the search results information to the generative pre-trained language model to obtain literature summary and analysis information; Obtain the full-text information of the document on the current page and provide the full-text information of the document to the generative pre-trained language model to obtain full-text analysis information of the document; Obtain the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model, and determine whether the literature summary analysis information or the literature full-text analysis information is a model illusion based on the literature knowledge base; based on the pre-established literature knowledge base, search in the literature knowledge base for literature related to the literature summary analysis information or the literature full-text analysis information generated by the generative pre-trained language model; Based on the search results, determine whether the literature summary analysis information or the full-text analysis information generated by the generative pre-trained language model is a model illusion.
8. An electronic device, comprising: processor; as well as A memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the method of using a document analysis assistant as described in claim 7.
9. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of applications, cause the electronic device to perform the method of using a document analysis assistant as described in claim 7.