Abstract generation method driven by multi-perspective fusion based on LLM

Abstract

The present invention provides a multi-perspective fusion-driven summary generation method based on LLM, belonging to the field of data processing technology. The method specifically includes the following steps: Step 1: After obtaining a target document uploaded by a user, multiple potential analysis perspectives are generated based on the document content; Step 2: Using a clustering algorithm to group and cluster the potential analysis perspectives to obtain a final analysis perspective; Step 3: Generating an emotional perspective based on user needs; Step 4: Inputting the final analysis perspective and the emotional perspective into a fuzzy inference system for fuzzification, calculating the membership of the fuzzified perspective combination, and selecting the optimal perspective combination accordingly; Step 5: Designing a fusion perspective prompt template based on the optimal perspective combination; Step 6: Generating a summary corresponding to the target document based on the fusion perspective prompt template and a pre-trained language model. The present invention improves the efficiency, accuracy, and adaptability of summary generation.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data processing technology, and in particular to a summary generation method driven by multi-perspective fusion based on LLM. Background Art

[0002] With the rapid development of artificial intelligence (AI), particularly natural language processing (NLP), text generation techniques based on large language models (LLMs) have been widely applied in various fields. For example, in the biomedical field, the volume of clinical research literature has rapidly increased, including extensive clinical trial data, drug development information, disease treatment methods, and more. Traditional summary generation methods often rely on simple text extraction or compression techniques. While these methods can extract key information from the literature and generate short summaries, they often overlook the personalized needs of users and fail to provide summaries tailored to their specific needs. For example, doctors may be more concerned with efficacy and safety data from clinical trials, while researchers may require more comprehensive research methods, data analysis, and potential research directions. However, traditional methods struggle to meet these diverse needs.

[0003] In recent years, the rise of large language models has significantly improved text generation technology, enabling the generation of more fluent and contextually coherent text. Using large language models to generate customized summaries that suit the needs of different types of users remains a challenge that needs to be addressed. Traditional perspective selection methods often rely on fixed rules or simple clustering algorithms, which are unable to flexibly handle complex and changing user needs and the diverse content of literature. Therefore, there is an urgent need for a customized clinical research summary generation method driven by multi-perspective fusion based on large language models to provide efficient, accurate, and personalized summary generation services. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides a summary generation method driven by multi-perspective fusion based on LLM, which at least partially solves the problems of poor generation efficiency, accuracy and adaptability in the existing technology.

[0005] The embodiment of the present invention provides a method for generating a summary driven by multi-perspective fusion based on LLM, including:

[0006] Step 1: After obtaining the target document uploaded by the user, multiple potential analysis perspectives are generated based on the document content;

[0007] Step 2: Use clustering algorithms to group and cluster potential analysis perspectives to obtain the final analysis perspectives;

[0008] Step 3: Generate emotional perspective based on user needs;

[0009] Step 4: Input the final analysis perspective and emotional perspective into the fuzzy inference system for fuzzification, calculate the membership degree of the fuzzified perspective combination, and select the best perspective combination accordingly;

[0010] Step 5: Design a fusion perspective prompt template based on the optimal perspective combination;

[0011] Step 6: Generate a summary corresponding to the target document based on the fused perspective prompt template and the pre-trained language model.

[0012] According to a specific implementation of the embodiment of the present invention, step 1 specifically includes:

[0013] Step 1.1: Use the named entity recognition model to identify the target document Extract different potential analysis perspectives, wherein the potential analysis perspectives include occupation ,interest and knowledge areas ;

[0014] Step 1.2: Convert potential analysis perspectives into perspective descriptions :

[0015] ;

[0016] in, Indicates different perspectives.

[0017] According to a specific implementation of the embodiment of the present invention, step 2 specifically includes:

[0018] Step 2.1: Extract perspective description using the Sentence-BERT model Embedding vector of ;

[0019] Step 2.2, use k-means clustering algorithm to embed the vector To group:

[0020] ;

[0021] in, Indicates the cluster groups;

[0022] Step 2.3: Select the view closest to the center in each cluster group As the final analysis perspective

[0023] ;

[0024] ;

[0025] in, Represents the embedding vector of the jth analysis perspective, which is used to calculate the distance between the perspective and the cluster center, Indicates the The centers of the cluster groups, It is The number of vectors in a cluster group, yes The samples in .

[0026] According to a specific implementation of the embodiment of the present invention, step 3 specifically includes:

[0027] Step 3.1: Send a preset question to the user and obtain the user's answer, and extract the emotional keywords in the user's answer;

[0028] Step 3.2, construct emotional perspective based on emotional keywords.

[0029] According to a specific implementation of the embodiment of the present invention, step 4 specifically includes:

[0030] Step 4.1: Use the sentiment perspective as the user feature U and the final analysis perspective as the document feature D;

[0031] Step 4.2: Fuzzify the user feature U and document feature D to obtain the fuzzy membership corresponding to the sentiment perspective. Fuzzy membership corresponding to the final analysis perspective :

[0032] ;

[0033] ;

[0034] Step 4.3, define fuzzy rules and use the input fuzzy membership and fuzzy membership Calculate the fuzzy membership of each perspective combination :

[0035] ;

[0036] Step 4.4, the fuzzy membership Converted into specific numerical membership:

[0037]

[0038] in, is the membership function obtained by fuzzy reasoning, Indicates the perspective combination sequence number, the value range of x is [0,1];

[0039] In step 4.5, the perspective combination with the highest specific numerical membership is selected as the optimal perspective combination.

[0040] According to a specific implementation of an embodiment of the present invention, the expression of the fusion perspective prompt template is:

[0041]

[0042] in, Represents the analytical perspective, Expressing emotional perspective, Represents an undefined conjunction or placeholder.

[0043] According to a specific implementation of an embodiment of the present invention, the expression for generating the summary corresponding to the target document is:

[0044] ;

[0045] in, Represents a pre-trained language model.

[0046] According to a specific implementation of the embodiment of the present invention, after step 6, the method further includes:

[0047] Collect user pairs to generate summaries Feedback And analyze it to generate analysis results, where the expression of the analysis results is

[0048] ;

[0049] in, Represents a preset analysis function;

[0050] Adjust the fusion perspective prompt template based on the analysis results .

[0051] The LLM-based multi-perspective fusion-driven summary generation scheme in the embodiment of the present invention includes: step 1, after obtaining the target document uploaded by the user, generating multiple potential analysis perspectives according to the document content; step 2, using a clustering algorithm to group and cluster the potential analysis perspectives to obtain the final analysis perspective; step 3, generating an emotional perspective according to user needs; step 4, inputting the final analysis perspective and the emotional perspective into the fuzzy reasoning system for fuzzification, calculating the membership of the fuzzified perspective combination, and selecting the best perspective combination accordingly; step 5, designing a fusion perspective prompt template according to the best perspective combination; step 6, generating a summary corresponding to the target document according to the fusion perspective prompt template and the pre-trained language model.

[0052] The beneficial effects of the embodiments of the present invention are as follows: through the solution of the present invention, the concept of membership is introduced to quantify the applicability of each potential perspective in clinical research literature or user preferences, thereby making perspective selection more accurate and dynamic, and improving the efficiency, accuracy and adaptability of summary generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 A schematic diagram of a flow chart of a summary generation method driven by multi-perspective fusion based on LLM provided in an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of the specific implementation process of a summary generation method driven by multi-perspective fusion based on LLM provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0057] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present invention, those skilled in the art will appreciate that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement an apparatus and / or practice a method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this apparatus and / or practice this method.

[0059] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention. The illustrations only show components related to the present invention and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be changed at will, and the component layout form may also be more complicated.

[0060] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0061] An embodiment of the present invention provides a summary generation method driven by multi-perspective fusion based on LLM, which can be applied to text recognition and understanding processes in artificial intelligence scenarios.

[0062] See also Figure 1 , is a flowchart of a summary generation method driven by multi-view fusion based on LLM provided by an embodiment of the present invention. Figure 1 and Figure 2 As shown, the method mainly includes the following steps:

[0063] Step 1: After obtaining the target document uploaded by the user, multiple potential analysis perspectives are generated based on the document content;

[0064] Furthermore, the step 1 specifically includes:

[0065] Step 1.1: Use the named entity recognition model to identify the target document Extract different potential analysis perspectives, wherein the potential analysis perspectives include occupation ,interest and knowledge areas ;

[0066] Step 1.2: Convert potential analysis perspectives into perspective descriptions :

[0067]

[0068] in, Indicates different perspectives.

[0069] In specific implementation, when a user uploads a document that needs to be summarized After that, the system automatically analyzes the document content and generates multiple potential analysis perspectives. , these perspectives can include:

[0070] Occupation: ;

[0071] Interest: ;

[0072] Knowledge Domain: ;

[0073] The analysis perspective generation process is as follows:

[0074] Named Entity Recognition: Use named entity recognition (NER) models to identify Extract occupation ,interest and knowledge areas ;

[0075] .

[0076] View template generation: convert the extracted view information into a view template ,in Representing different perspectives

[0077] .

[0078] Step 2: Use clustering algorithms to group and cluster potential analysis perspectives to obtain the final analysis perspectives;

[0079] Based on the above embodiment, step 2 specifically includes:

[0080] Step 2.1: Extract perspective description using the Sentence-BERT model Embedding vector of ;

[0081] Step 2.2, use k-means clustering algorithm to embed the vector To group:

[0082]

[0083] in, Indicates the cluster groups;

[0084] Step 2.3: Select the view closest to the center in each cluster group As the final analysis perspective

[0085] ;

[0086] ;

[0087] in, Represents the embedding vector of the jth analysis perspective, which is used to calculate the distance between the perspective and the cluster center, Indicates the The centers of the cluster groups, It is The number of vectors in a cluster group, R is The samples in .

[0088] In the specific implementation, in order to ensure the diversity and representativeness of perspectives, a clustering algorithm is used to group and cluster the generated analysis perspectives. The specific process is as follows:

[0089] 1) Embedding vector extraction: Let the perspective be described as , use Sentence-BERT (SBERT) to extract embedding vectors :

[0090] ;

[0091] 2) Perspective clustering: Use k-means clustering algorithm to cluster the embedding vectors To group:

[0092] ;

[0093] in, Indicates the Cluster groups.

[0094] 3) Cluster center perspective selection: select the n perspectives closest to the center in each cluster group As a final analytical perspective:

[0095] ;

[0096] ;

[0097] in, Indicates the The center of the cluster group.

[0098] Step 3: Generate emotional perspective based on user needs;

[0099] Furthermore, the step 3 specifically includes:

[0100] Step 3.1: Send a preset question to the user and obtain the user's answer, and extract the emotional keywords in the user's answer;

[0101] Step 3.2, construct emotional perspective based on emotional keywords.

[0102] For example, users can customize their emotional perspective by answering a question like, "What aspects of clinical research literature are you most interested in?" The system extracts emotional keywords from the user's answer and constructs an emotional perspective template. The system asks users questions like, "What aspects of clinical research literature are you most interested in?" to guide them in expressing their emotional concerns about clinical research. For example, if a user is more interested in efficacy and safety data, they might answer, "I want to know the impact of this clinical study on efficacy and safety."

[0103] Emotional keyword extraction:

[0104] Let the user answer , use sentiment analysis model to extract sentiment keywords :

[0105] For example, if a user answers "I want to know the impact of this clinical study on efficacy and safety", the emotional keywords extracted by the system may be "efficacy" and "safety".

[0106] ;

[0107] Then build the emotional perspective template based on this:

[0108] Emotional Perspective Template Expressed as .

[0109] Step 4: Input the final analysis perspective and emotional perspective into the fuzzy inference system for fuzzification, calculate the membership degree of the fuzzified perspective combination, and select the best perspective combination accordingly;

[0110] Based on the above embodiment, step 4 specifically includes:

[0111] Step 4.1: Use the sentiment perspective as the user feature U and the final analysis perspective as the document feature D;

[0112] Step 4.2: Fuzzify the user feature U and document feature D to obtain the fuzzy membership corresponding to the sentiment perspective. Fuzzy membership corresponding to the final analysis perspective :

[0113] ;

[0114] ;

[0115] Step 4.3, define fuzzy rules and use the input fuzzy membership and fuzzy membership Calculate the fuzzy membership of each perspective combination :

[0116] ;

[0117] Step 4.4, the fuzzy membership Converted into specific numerical membership:

[0118] ;

[0119] in, is the membership function obtained by fuzzy reasoning, Indicates the perspective combination sequence number, the value range of x is [0,1];

[0120] In step 4.5, the perspective combination with the highest specific numerical membership is selected as the optimal perspective combination.

[0121] In practice, a fuzzy inference system (FIS) can be used to adjust the membership of each perspective and dynamically select the most suitable perspective combination based on document characteristics and user preferences. Using predefined fuzzy rules, the system can generate personalized summaries for different users.

[0122] 1) Input variables

[0123] Define the input variables of user features U and document features D:

[0124] ;

[0125] ;

[0126] Where U and D are text embedding vectors.

[0127] 2) Fuzzification

[0128] The user feature U and document feature D are fuzzy processed and converted into membership values:

[0129] ;

[0130] ;

[0131] Fuzzification of user features:

[0132] The user feature U is fuzzified and converted into a fuzzy set (such as "high", "medium", and "low").

[0133] For example, a user's emotional perspective on efficacy may be blurred to "high".

[0134] Fuzzification of document features:

[0135] The document feature D is fuzzified and converted into a fuzzy set (such as "high correlation", "medium correlation", "low correlation").

[0136] For example, the relevance of the efficacy analysis perspective in the document may be blurred as "medium".

[0137] 3) Application of fuzzy rules

[0138] Define fuzzy rules and calculate the membership of each perspective combination based on the input fuzzy membership:

[0139] Example fuzzy rule:

[0140] Rule 1: If the sentiment perspective is "high" and the analysis perspective is "high correlation", then the perspective combination membership is "very high".

[0141] Rule 2: If the sentiment perspective is "medium" and the analysis perspective is "medium correlation", then the perspective combination membership is "medium".

[0142] Rule 3: If the sentiment perspective is "low" and the analysis perspective is "high correlation", then the perspective combination membership is "low".

[0143] 4) Fuzzy reasoning

[0144] Calculate the fuzzy membership of the perspective combination:

[0145] ;

[0146] For each perspective combination, fuzzy rules are applied for reasoning based on the input fuzzified membership values.

[0147] The fuzzy membership (e.g., "very high," "medium," etc.) of each viewpoint combination is calculated.

[0148] 5) Defuzzification

[0149] Fuzzy membership Converted into specific numerical membership:

[0150] ;

[0151] in, is the membership function obtained by fuzzy reasoning, and the value range of x is [0,1]

[0152] 6) Optimal viewing angle combination selection

[0153] Select the perspective combination with the highest membership:

[0154] ;

[0155] in, Represents the final membership of each perspective combination.

[0156] Step 5: Design a fusion perspective prompt template based on the optimal perspective combination;

[0157] Furthermore, the expression of the fusion perspective prompt template is:

[0158] ;

[0159] in, Represents the analytical perspective, Expressing emotional perspective, Represents an undefined conjunction or placeholder.

[0160] In specific implementation, when designing the analysis perspective prompt template, the membership degree will be taken into consideration, and the perspective with a higher membership degree will be given priority.

[0161] For example: Tanalysis = "Please summarize this clinical research paper on [Ki] from the perspective of [Pi], focusing on the efficacy data and trial results related to [Ii]. When incorporating the emotional perspective, the system also considers the degree of emotional perspective, generating a fusion perspective prompt template that better meets the personalized needs of users (such as medical staff or researchers).

[0162] Different prompt templates are designed according to the perspective information to guide the subsequent summary generation process.

[0163] Analysis perspective prompt template design:

[0164] Assuming there is analytical perspective information ,in:

[0165] : Occupation

[0166] : Interest

[0167] : Knowledge Domain

[0168] Analysis Perspective Prompt Template It can be expressed as:

[0169] = "Please from" + [ ]+" perspective, summarize this article about"+[ ]+"Clinical research literature, focus on"+[ ] + “Relevant efficacy data and trial results.”

[0170] In this template, occupation Determines the professionalism and interests of the perspective It emphasizes the experimental details or research directions of interest, knowledge areas It specifies the clinical research scope of the literature.

[0171] Emotional perspective prompt template design:

[0172] Assuming there is emotional perspective information ,in:

[0173] : Emotional keywords extracted from user answers

[0174] Emotional Perspective Prompt Template It can be expressed as:

[0175] ="Please summarize this clinical research paper from your personal perspective, focusing on its impact on"+[ ] + “The impact.”

[0176] Fusion perspective prompt template design:

[0177] Formula for the fusion perspective prompt template:

[0178] ;

[0179] in, Indicates the analysis perspective, including information such as profession, interest, and knowledge domain. Indicates emotional perspective, including the user's emotional keywords or focus points (such as emotional tendencies, areas of concern), Indicates an undefined connective or placeholder, which may appear anywhere in the prompt template and can be multiple connectives.

[0180] Fusion perspective prompt template Prompt templates by combining analytical perspectives and emotional perspective prompt templates Generate, expressed as:

[0181] = +"At the same time, please pay attention to the content of the literature" +[ ] + “The impact.”

[0182] Step 6: Generate a summary corresponding to the target document based on the fused perspective prompt template and the pre-trained language model.

[0183] Furthermore, the expression for generating the summary corresponding to the target document is

[0184] ;

[0185] in, Represents a pre-trained language model.

[0186] Furthermore, after step 6, the method further includes:

[0187] Collect user pairs to generate summaries Feedback And analyze it to generate analysis results, where the expression of the analysis results is

[0188] ;

[0189] in, Represents a preset analysis function;

[0190] Adjust the fusion perspective prompt template based on the analysis results .

[0191] In specific implementation, let the pre-trained language model be LLM and the prompt template be , the original document is , generating a summary that fits the perspective :

[0192] .

[0193] At the same time, based on user feedback, the system can adjust the membership of each perspective to continuously improve the accuracy and personalization of summary generation. For example, if users (such as medical staff or researchers) feedback that certain perspectives contain too much or too little content, the system will adjust their membership to generate clinical research summaries that better meet user expectations.

[0194] Feedback analysis: Collect user feedback on generated summaries feedback and make it , such as user ratings, comments, etc., analyze user feedback and identify deficiencies in summary generation:

[0195] .

[0196] Template optimization: Adjust the prompt template based on feedback analysis results or information extraction strategies to continuously improve the quality and personalization of summary generation.

[0197] .

[0198] The LLM-based multi-perspective fusion-driven summary generation method provided in this embodiment introduces the concept of membership to quantify the applicability of each potential perspective in clinical research literature or user preferences, thereby making perspective selection more accurate and dynamic, and improving the efficiency, accuracy, and adaptability of summary generation.

[0199] The following describes the method of the present invention in detail using a specific example. For example, let's assume the system detects that user A has a membership of 0.8 for drug efficacy and 0.4 for research methodology. The generated summary will primarily focus on efficacy, while also taking into account information related to the research methodology.

[0200] Example of an analytical perspective:

[0201] Occupation: Doctor

[0202] Interest: Clinical trial efficacy

[0203] Knowledge Domain: Drug Development

[0204] Membership degree: 0.8

[0205] The generated abstract will focus on describing the efficacy data from the clinical trial, while briefly mentioning the trial details and research methods.

[0206] Suppose there is a clinical research paper with the following content:

[0207] "New drugs have demonstrated significant improvements in efficacy in clinical trials. A multicenter randomized controlled trial in patients with chronic diseases showed that the drug reduced primary efficacy indicators (such as the disease activity index) by 45% over 12 weeks, significantly better than the 15% reduction in the placebo group. Furthermore, the drug performed well in terms of safety, with only 5% of patients reporting mild adverse reactions (such as headache and nausea), and no serious adverse events were observed. Researchers pointed out that the drug's mechanism of action involves inhibiting specific inflammatory factors, which may provide theoretical support for its high efficacy and low side effects. The trial also explored the performance of the drug at different doses, and the results showed that a dose of 10 mg / day achieved the best balance between efficacy and safety. The study was conducted in collaboration with multiple medical institutions, and advanced statistical models were used in data analysis to ensure the reliability of the results."

[0208] Example of an analytical perspective (objective perspective):

[0209] Occupation: Doctor Interest: Clinical trial efficacy

[0210] Knowledge Domain: Drug Development

[0211] Analysis Perspective prompt template: "Please summarize this clinical research paper on a new drug from a technical perspective, focusing on its efficacy data and trial results."

[0212] Generated abstract: "A new drug has demonstrated significant improvements in efficacy in clinical trials. A multicenter randomized controlled trial showed that the drug reduced the disease activity index by 45% over 12 weeks, significantly better than the 15% reduction in the placebo group. This drug is expected to provide a more effective treatment option for patients with chronic diseases."

[0213] Example of emotional perspective (subjective perspective):

[0214] Occupation: General Reader

[0215] Interest: Disease treatment effects

[0216] Knowledge Domain: Health Management

[0217] Emotional perspective prompt template: "Please summarize this clinical research article from a personal perspective, focusing on your overall impression and emotional response."

[0218] Generated abstract: "The breakthrough progress of a new drug in clinical trials is encouraging. The drug significantly reduced disease activity within 12 weeks, with only 5% of patients reporting mild adverse reactions. Developers say it will provide more effective treatment for patients with chronic diseases, which is good news for improving health management."

[0219] Comparative analysis:

[0220] Analytical perspective (objective perspective): The focus is on describing the trial data and efficacy results, such as a 45% reduction in the disease activity index compared to the control group. The text should be objective, primarily presenting facts and trial results.

[0221] Emotional perspective (subjective perspective): focuses more on personal feelings and emotional reactions, such as excitement and hope. The text content is subjective and includes emotional evaluations of drug efficacy and expectations for health management.

[0222] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware or a combination thereof.

[0223] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A summary generation method driven by multi-perspective fusion based on LLM, characterized by: include: Step 1: After obtaining the target document uploaded by the user, multiple potential analysis perspectives are generated based on the document content; Step 2: Use clustering algorithms to group and cluster potential analysis perspectives to obtain the final analysis perspectives; The step 2 specifically includes: Step 2.1: Extract perspective description using the Sentence-BERT model Embedding vector of ; Step 2.2, use k-means clustering algorithm to embed the vector To group: ； in, Indicates the cluster groups; Step 2.3: Select the view closest to the center in each cluster group As the final analysis perspective ； ； in, Representative The embedding vector of the analysis perspective is used to calculate the distance between the perspective and the cluster center. Indicates the The centers of the cluster groups, It is The number of vectors in a cluster group, R is Samples in Step 3: Generate emotional perspective based on user needs; Step 4: Input the final analysis perspective and emotional perspective into the fuzzy inference system for fuzzification, calculate the membership degree of the fuzzified perspective combination, and select the best perspective combination accordingly; The step 4 specifically includes: Step 4.1: Use emotional perspective as user characteristics , taking the final analysis perspective as the document feature ; Step 4.2: User features and document features Perform fuzzy processing to obtain the fuzzy membership corresponding to the emotional perspective Fuzzy membership corresponding to the final analysis perspective : ； ； Step 4.3, define fuzzy rules and use the input fuzzy membership and fuzzy membership Calculate the fuzzy membership of each perspective combination : ； Step 4.4, the fuzzy membership Converted into specific numerical membership: ； in, is the membership function obtained by fuzzy reasoning, Indicates the perspective combination number, The value range of is [0,1]; Step 4.5, selecting the perspective combination with the highest specific numerical membership as the best perspective combination; Step 5: Design a fusion perspective prompt template based on the optimal perspective combination; The expression of the fusion perspective prompt template is: ； Among them, LLM represents the pre-trained language model, Represents the analytical perspective, Expressing emotional perspective, Indicates an undefined conjunction or placeholder; Step 6: Generate a summary corresponding to the target document based on the fused perspective prompt template and the pre-trained language model.

2. The method according to claim 1, characterized in that , the step 1 specifically includes: Step 1.1: Use the named entity recognition model to identify the target document Extract different potential analysis perspectives, wherein the potential analysis perspectives include occupation ,interest and knowledge areas ; Step 1.2: Convert potential analysis perspectives into perspective descriptions : ； in, Indicates different perspectives.

3. The method according to claim 2, characterized in that , the step 3 specifically includes: Step 3.1: Send a preset question to the user and obtain the user's answer, and extract the emotional keywords in the user's answer; Step 3.2, construct emotional perspective based on emotional keywords.

4. The method according to claim 3, characterized in that , the expression for generating the summary corresponding to the target document is 。 5. The method according to claim 4, characterized in that After step 6, the method further includes: Collect user pairs to generate summaries Feedback And analyze it to generate analysis results, where the expression of the analysis results is ； in, Represents a preset analysis function; Adjust the fusion perspective prompt template based on the analysis results .

Images