Intelligent multi-modal document fusion and personalized retrieval system and method

Abstract

The invention discloses an intelligent multi-modal document fusion and personalized retrieval system and method, and relates to the technical field of document fusion and personalized retrieval, the intelligent multi-modal document fusion and personalized retrieval system comprises a document fusion module, a personalized retrieval module and a personalized recommendation module, and the personalized recommendation module analyzes basic keywords, style keywords and emotion keywords of each document to obtain a personalized recommendation result; therefore, fusion of the documents is realized, uniformity of contents, styles and emotions of the fused documents is guaranteed, and description portraits are performed on the user based on historical reading records of the user, so that document recommendation is performed on the user according to the description portraits and search keywords of the user during search of the user, and user experience is improved. According to the method, content relevance, expression suitability and emotional resonance of the documents and user retrieval are guaranteed, meanwhile, the reading preference of the user is analyzed, typesetting of all recommended documents is set according to the reading preference, and personalized recommendation for the user is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of document fusion and personalized retrieval, in particular to an intelligent multi-modal document fusion and personalized retrieval system and method. BACKGROUND

[0002] Traditional information retrieval returns a large number of fragmented search results, and users need to spend a lot of time screening. It is beneficial to improve the efficiency of users obtaining information to fuse the same type of information and provide customized search results according to the reading habits of users. Therefore, the present application proposes an intelligent multi-modal document fusion and personalized retrieval system and method.

[0003] Prior art such as the invention patent application for document-level knowledge extraction and fusion method and system based on large language model with publication number CN119358546A belongs to the field of industrial robots, which includes: determining the required key information range and establishing a keyword dictionary; dividing the document-level unstructured data according to the keyword dictionary by paragraph to obtain the divided sub-documents; using the producer-consumer mode to integrate the asynchronous architecture of the large model to build a software system, and using the software system to sequentially perform knowledge extraction tasks on the divided sub-documents to extract key information from the unstructured data of the sub-documents; integrating and classifying all the key information extracted from the same sub-document to obtain regular data, and then performing knowledge fusion processing on the regular data; the correlation degree between paragraphs is matched with the keyword dictionary to divide the document, the content of the divided sub-document is highly aggregated, the difficulty of processing complex documents by the large model is reduced, the producer-consumer mode is integrated in the large model to avoid system blocking and improve the concurrent processing capability of the system.

[0004] Prior art such as the invention patent application for a personalized retrieval method with publication number CN116821965B discloses a high-efficiency and intelligent personalized search scheme based on user interest. The data user only needs to input the query keyword. In the specific retrieval process, the query keyword given by the data user is combined with the user interest model to intelligently perform multi-keyword accurate retrieval and fuzzy semantic retrieval, and the first k accurate search results are returned. In addition, in the retrieval process, the query keyword, the extended keyword, and the keyword matching obtain the keyword candidate index, and the candidate ciphertext document set is obtained, which improves the retrieval efficiency. In order to protect the privacy of data, the inverted index is encrypted using homomorphic encryption, and a hybrid cloud server is used. The user interest model is stored in the private cloud server, and the search results are sorted on the private cloud server. This not only ensures the security of user data and computing operations, but also reduces the burden on the user end.

[0005] For the above scheme, there are the following technical problems: 1. The current technology mainly extracts keywords or key information to realize document fusion, but the current technology does not consider analyzing the document style and the overall sentiment of the document. When the main key subjects described by each document are the same, but the document style and sentiment are completely opposite, the fusion of the documents will lead to content conflict and logical rupture of the fused documents.

[0006] 2. The current technology mainly performs multi-keyword accurate search based on the user's query keywords and the user's interest model during the retrieval process, but the current technology does not consider analyzing the user's historical reading style, and then displays different document content and document types to the user according to the user's historical reading style. SUMMARY

[0007] The purpose of the present application is to provide an intelligent multi-modal document fusion and personalized retrieval system and method, which solves the problems existing in the background art.

[0008] To solve the above technical problems, the present application adopts the following technical scheme: In a first aspect, the present application provides an intelligent multi-modal document fusion and personalized retrieval system, comprising: a document fusion module: for extracting each basic keyword and each description keyword of each document, and then fusing each document according to each basic keyword and each description keyword to obtain each fused document.

[0009] A personalized retrieval module: for analyzing the description type of a target user according to the historical reading records of the target user, and then describing the portrait of the target user according to the description type of the target user, and when the target user performs retrieval, analyzing each retrieval recommendation document of the target user according to the description portrait of the target user and each retrieval content, and analyzing the reading preference of the target user based on the historical reading records of the target user, thereby recommending and typesetting each retrieval recommendation document based on the reading preference.

[0010] A personalized recommendation module: for analyzing the reading association of the target user with other users based on the historical reading records of the target user, and then recommending the homepage display document of the target user.

[0011] In a second aspect, the present application provides an intelligent multi-modal document fusion and personalized retrieval method, comprising: step one, extracting each basic keyword and each description keyword of each document, and then fusing each document according to each basic keyword and each description keyword to obtain each fused document.

[0012] Step two, analyze the description type of the target user according to the historical reading records of the target user, and then describe the portrait of the target user according to the description type of the target user, and when the target user searches, analyze the target user's each search recommended document according to the description portrait of the target user and each search content, and analyze the reading preferences of the target user based on the historical reading records of the target user, so as to recommend and arrange each search recommended document based on the reading preferences.

[0013] Step three, based on the historical reading records of the target user, analyze the reading association between the target user and other users, and then recommend the homepage display document for the target user.

[0014] The beneficial effects of the present application are: 1. The intelligent multi-modal document fusion and personalized search system and method provided by the present application analyzes the basic keywords, style keywords and emotional keywords of each document, and then realizes the fusion of each document, ensures the unity of the content, style and emotion of the fused document, and describes the portrait of the user based on the historical reading records of the user, so as to recommend the document to the user according to the description portrait of the user and the search keywords when the user searches, ensure the content relevance, expression adaptability and emotional resonance of the document and the user search, and analyze the reading preferences of the user, and set the layout of each recommended document accordingly, realize the personalized recommendation for the user.

[0015] 2. The present application extracts and analyzes the basic keywords, style keywords and emotional keywords of each document, and fuses each document accordingly, not only retains the core elements of the document at the content level, but also ensures the unity of the style and emotion of the fused document, and lays a foundation for subsequent personalized search of the user.

[0016] 3. The present application analyzes the reading style and reading emotional type preference of the target user through the historical reading records of the target user, and when the target user searches, not only matches the document content based on the search keywords, but also matches the document style and emotion based on the reading style and reading emotional type preference of the target user, thereby greatly improving the information acquisition efficiency, and the user's favorite document style is also beneficial to the user's quick understanding of the document content. At the same time, the document type preference of the target user is analyzed, so as to display the document obtained by searching, which is beneficial to enhancing the reading experience of the user and improving the content relevance, expression adaptability and emotional resonance of the search content and the document obtained by searching of the target user. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0018] Figure 1 The system structure connection diagram of the present application.

[0019] Figure 2 The method implementation step flow diagram of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0021] Referring to Figure 1 As shown in the figure, the present application provides an intelligent multi-modal document fusion and personalized retrieval system in the first aspect, which includes the following modules: a document fusion module: used for extracting each basic keyword and each description keyword of each document, and then fusing each document according to each basic keyword and each description keyword to obtain each fusion document.

[0022] It should be noted that the each basic keyword includes the word related to the core theme, key concept and important information in the document; the each description keyword includes the keyword describing the style and emotion of the document; for example, in the sentence "Through popular interpretation and interesting analogy, we will unveil the mysterious veil behind the extinction of dinosaurs", "popular interpretation" and "interesting analogy" are style keywords, and "extinction of dinosaurs" is a basic keyword.

[0023] In one specific example, the each document includes each text document, each video document, each audio document and each picture document.

[0024] It should be noted that for each text document, the TextRank algorithm is used to extract the basic keywords, and the style keyword and the sentiment keyword are extracted through the style library and the sentiment library respectively; for each video document, first, the subtitle text of the video is extracted, and then the analysis method of the text document is used to analyze the basic keywords, the style keywords and the sentiment keywords of the video document, and the key frame images in the video are identified through the image recognition technology, and then the basic keywords are obtained; for each audio document, first, the speech recognition technology is used to convert the audio document into a text document, and then the analysis method of the text document is used to analyze the basic keywords, the style keywords and the sentiment keywords of the audio document, and the style keywords and the sentiment keywords of the audio document are analyzed through the speech emotion analysis model; for each picture document, the image recognition technology is used to analyze the objects, scenes, characters, colors, compositions, lights and performances in the picture, and then the basic keywords, the style keywords and the sentiment keywords of the picture document are obtained, wherein the TextRank algorithm and the image recognition technology are prior art, and thus will not be described again.

[0025] In one specific example, the basic keywords and the description keywords of each document are matched to obtain each fusion document, and the specific process is as follows: according to the basic keywords of each document, the synonyms of the basic keywords stored in the web database are combined to construct the basic keyword set of each document.

[0026] According to the description keywords of each document, the style keywords and the sentiment keywords of each document are extracted, the synonyms of the style keywords and the sentiment keywords stored in the web database are combined to construct the style keyword set and the sentiment keyword set of each document, and the description keyword set of each document is comprehensively recorded.

[0027] The basic keyword set and the description keyword set of each document are matched to obtain each fusion document.

[0028] In one specific example, the basic keyword set and the description keyword set of each document are matched to obtain each fusion document, and the specific process is as follows: the basic keywords and the description keywords of each document are substituted into the document association evaluation model, and the association evaluation coefficient between each document is output through the document evaluation model expression.

[0029] It should be noted that any document is recorded as a target document, and then the basic keyword set and the description keyword set of the target document are recorded as U and W respectively, wherein W={E,T}, E and T are the style keyword set and the sentiment keyword set of the target document respectively, and the basic keyword set and the description keyword set of each document except the target document are recorded as U k and W kwherein W k = {E k , T k}, wherein k is the number of each document other than the target document, k is a positive integer, E k and T k are the style keyword set and the sentiment keyword set of the document numbered k, respectively, and are calculated according to the calculation formula:

[0030] The correlation evaluation coefficient χ k of the target document and each document is calculated, wherein a1 and a2 are the weight factors corresponding to the basic keywords and the description keywords of the document, respectively, and z1 and z2 are the weight factors corresponding to the style keywords and the sentiment keywords of the document, respectively, and the correlation evaluation coefficient of each document is obtained according to the analysis.

[0031] wherein a1 and a2 are both greater than 0 and less than 1, a1 and a2 are obtained by the analytic hierarchy process, which is a prior art and thus will not be described in detail, and z1 and z2 are set in the same way as a1 and a2.

[0032] The correlation evaluation coefficient of each document is compared with the set document correlation evaluation coefficient threshold value, when the correlation evaluation coefficient is greater than or equal to the set document correlation evaluation coefficient, the corresponding each document is recorded as each fusible document, and vice versa, the corresponding each document is recorded as each non-fusible document, and each fusible document and each non-fusible document are obtained, and each fusible document is fused to obtain each fused document.

[0033] It should be noted that the document correlation evaluation coefficient threshold value is set by relevant staff and is not specifically limited herein.

[0034] The personalized retrieval module is used to analyze the description type of the target user according to the historical reading record of the target user, further describe the portrait of the target user according to the description type of the target user, and analyze the each retrieval recommended document of the target user according to the description portrait of the target user and each retrieval content when the target user performs retrieval, and recommend the each retrieval recommended document based on the reading preference of the target user analyzed according to the historical reading record of the target user.

[0035] In a specific example, the description type of the target user includes a style description type and a sentiment description type.

[0036] It should be noted that when the target user tends to read straightforward, concise and rigorous documents, the style description type of the target user is concise and rigorous, and the emotional description type is neutral emotion, etc.; when the target user tends to read humorous and fast-paced video documents, the style description type of the target user is humorous and fast-paced, and the emotional description type is positive emotion, etc.

[0037] In one specific example, the description type of the target user is analyzed according to the historical reading records of the target user, and then the description portrait of the target user is described according to the description type of the target user, and the specific process is as follows: obtaining the historical reading records of the target user from the data center, and then analyzing to obtain the description keyword set of each document read by the target user in the historical reading records, extracting the style description keywords and the emotional description keywords of each historical document based on the description keyword set of each historical document, and then statistically obtaining the appearance frequency of each style description keyword and each emotional description keyword, and recording each style keyword and each emotional keyword with an appearance frequency greater than or equal to a preset frequency as the description portrait label of the target user.

[0038] In one specific example, the description portrait of the target user and each search keyword are analyzed to obtain each search recommended document of the target user, and the specific process is as follows: first, the search content of the target user is preprocessed to obtain each search keyword, and then each search keyword and the description portrait label of the target user are respectively converted into index word vectors V n , style description word vectors Q ξ and emotional description word vectors V σ , wherein n, ξ and σ respectively represent the number of index word vectors, the number of style description word vectors and the number of emotional description word vectors, n, ξ and σ are all positive integers, and each basic keyword, each style keyword and each emotional keyword of each fusion document are respectively converted into document basic vectors R jm , document style description vectors P jω and document emotional description vectors , wherein j represents the number of each fusion document, m, ω and respectively represent the number of document basic vectors, the number of document style vectors and the number of temperature emotional vectors, j, m, ω and are all positive integers, the index word vectors and the description word vectors of the target user and the document basic vectors and the document description vectors of each fusion document are substituted into the vector correlation degree matching model, and the expression of the vector correlation degree matching model is:

[0039] The search correlation evaluation coefficient L j of the target user and the jth fusion document is calculated, wherein M and N respectively represent the total number of index word vectors and the total number of document basic vectors, and respectively represent the total number of style description word vectors of the target user and the total number of document style description vectors, and and respectively represent the total number of sentiment description word vectors of the target user and the total number of document sentiment description vectors.

[0040] It should be noted that the pre-processing of the target user's search content includes word segmentation, stop word removal and part-of-speech filtering, and then the search keywords are obtained, for example, the search content is "artificial intelligence development trend", and the pre-processed search keywords are "artificial intelligence", "development" and "trend".

[0041] It should be noted that the word vector method includes vectorizing each search keyword according to a word vector model such as Word2Vec, GloVe or BERT, wherein Word2Vec, GloVe and BERT are all prior art, and thus will not be described again.

[0042] The target user and the search association evaluation coefficient of each fusion document are sorted from large to small, and the display order of each fusion document for the target user's current search is obtained accordingly.

[0043] In one specific example, the historical reading records of the target user are analyzed based on the reading preferences of the target user, and each search recommended document is recommended and arranged based on the reading preferences, and the specific process is as follows: the historical reading records of the target user are obtained from the data center, the reading times and reading durations of the target user for each type of document are obtained based on the historical reading records of the target user, the reading preference evaluation model is substituted, the reading preference evaluation coefficient of the target user for each document is output through the expression of the reading preference evaluation model, and then the ratio of the reading preference evaluation coefficient of the target user for each document is obtained, and the display form of each fusion document on the target user's current search page is set according to the ratio of the reading preference evaluation coefficient of the target user for each document.

[0044] It should be noted that the expression of the reading preference evaluation model is: wherein β u represents the reading preference index of the target user for the type μ document, μ is the document type number, μ is a positive integer, the document types include picture type, video type, audio type and text type, and χ u and η u are the reading times and reading durations of the target user for each type of document, and represent the average reading times and average reading durations of the target user, U represents the total number of document types, and φ represent the total reading times and total reading durations of the target user.

[0045] When the ratio of the reading preference evaluation coefficients of the target user for the picture type, video type, audio type and text type documents is 2:1:3:1, the documents of the picture type, video type, audio type and text type are displayed in the ratio of 2:1:3:1 on the current search page of the target user.

[0046] The personalized recommendation module is configured to analyze the reading association of the target user with other users based on the historical reading records of the target user, and then recommend the homepage display documents for the target user.

[0047] In a specific example, the reading association of the target user with other users is analyzed based on the historical reading records of the target user, and then the homepage display documents for the target user are recommended, and the specific analysis process is as follows: the description word vectors of each user are analyzed in the manner of analyzing the description word vector of the target user, the reading description similarity evaluation coefficient of the target user with each user is analyzed based on the description word vector of the target user and the description word vector of each user, each user whose reading description similarity coefficient is higher than a set reading description similarity coefficient threshold is recorded as each reference user, each fusion document read by each reference user in a preset time period is obtained, and the homepage display document for the target user is randomly recommended.

[0048] It should be noted that the reading description similarity evaluation coefficient of the target user with each user is analyzed by the vector correlation matching model expression, and the vector correlation matching model expression has been described in analyzing the search association evaluation coefficient of the target user with each fusion document, and thus will not be described again.

[0049] It should be noted that the reading description similarity coefficient threshold and the preset time period are both set by relevant staff, and will not be specifically limited here.

[0050] It should be noted that the reading association of the target user with other users is analyzed based on the historical reading records of the target user, and then the homepage display documents for the target user are recommended, which ensures the richness of the user's daily browsing and breaks the information cocoon phenomenon.

[0051] Referring to Figure 2 The present application provides an intelligent multi-modal document fusion and personalized search method in a second aspect, which includes the following steps: step one, extracting each basic keyword and each description keyword of each document, and then fusing each document according to each basic keyword and each description keyword to obtain each fusion document.

[0052] Step two, analyzing the description type of the target user according to the historical reading record of the target user, and then describing the portrait of the target user according to the description type of the target user, and when the target user searches, analyzing the target user's each search recommended document according to the description portrait of the target user and each search content, and analyzing the reading preference of the target user based on the historical reading record of the target user, so as to recommend and arrange each search recommended document based on the reading preference.

[0053] Step three, analyzing the reading association between the target user and other users based on the historical reading record of the target user, and then recommending the homepage display document of the target user.

[0054] The intelligent multi-modal document fusion and personalized search system and method provided by the application analyzes the basic keywords, style keywords and emotional keywords of each document, and then realizes the fusion of each document, ensures the unity of the content, style and emotion of the fused document, describes the user based on the historical reading record of the user, and then recommends the document to the user according to the description portrait of the user and the search keywords when the user searches, ensures the content relevance, expression adaptability and emotional resonance of the document and the user search, and analyzes the reading preference of the user, and sets the layout of each recommended document accordingly, realizes the personalized recommendation to the user.

[0055] The above content is only an example and description of the concept of the application. Those skilled in the art can make various modifications, supplements or substitutions of similar ways to the described specific embodiments without departing from the concept of the application or exceeding the scope defined by the application.

Claims

1. An intelligent multi-modal document fusion and personalized retrieval system, characterized in that, The application relates to a method for recommending a homepage display document to a target user. The method comprises the following steps: step 1, extracting basic keywords and description keywords of each document, and then fusing the documents according to the basic keywords and the description keywords to obtain fused documents; step 2, analyzing the description type of the target user according to historical reading records of the target user, then performing description portrait on the target user according to the description type of the target user, and when the target user performs retrieval, obtaining each retrieval recommended document of the target user according to the description portrait of the target user and each retrieval content, and then recommending and arranging each retrieval recommended document based on reading preferences of the target user analyzed according to the historical reading records of the target user; and step 3, analyzing the reading association between the target user and other users based on the historical reading records of the target user, and then recommending a homepage display document to the target user. The documents comprise text documents, video documents, audio documents and picture documents. The basic keyword set of each document is constructed according to the basic keywords of each document and the near-synonyms of the basic keywords stored in a web database.

2. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 1, wherein, The style keywords and the emotional keywords of each document are extracted according to the description keywords of each document, the near-synonyms of the style keywords and the emotional keywords stored in the web database are combined, the style keyword set and the emotional keyword set of each document are constructed, and the description keyword set of each document is comprehensively recorded.

3. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 2, wherein, The basic keyword set and the description keyword set of each document are matched with each other to obtain the fused documents. The basic keywords and the description keywords of each document are substituted into a document association evaluation model, and the association evaluation coefficients between the documents are output through a document evaluation model expression. The association evaluation coefficients of each document are compared with a set document association evaluation coefficient threshold value, when the association evaluation coefficient is greater than or equal to the set document association evaluation coefficient, each corresponding document is recorded as each fusible document, otherwise, each corresponding document is recorded as an unfusible document, and each fusible document is fused to obtain the fused documents. The description type of the target user comprises a style description type and an emotional description type.

4. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 3, wherein, The description type of the target user is analyzed according to the historical reading records of the target user, and then the description portrait of the target user is performed according to the description type of the target user. The historical reading records of the target user are obtained from a data center, the description keyword sets of each document read by the target user in the historical reading records are analyzed, the style description keywords and the emotional description keywords of each historical document are extracted based on the description keyword sets of the historical documents, and then the appearance frequencies of the style description keywords and the emotional description keywords are counted, each style keyword and each emotional keyword with an appearance frequency greater than or equal to a preset frequency are recorded as the description portrait labels of the target user. ​ 5. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 4, wherein, ​ 6. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 5, wherein, ​ ​ 7. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 6, wherein, The target user's each search recommended document is analyzed according to the target user's description image and each search keyword, and the specific process is as follows: Firstly, the search content of the target user is preprocessed to obtain each search keyword, and then each search keyword and the description image label of the target user are respectively converted into index word vectors V n , style description word vectors Q ξ and sentiment description word vectors V σ , where n, ξ and σ represent the number of index word vectors, the number of style description word vectors and the number of sentiment description word vectors respectively, and n, ξ and σ are positive integers. Meanwhile, each basic keyword, each style keyword and each sentiment keyword of each fusion document are respectively converted into document basic vectors R jm , document style description vectors P jω and document sentiment description vectors , where j represents the number of each fusion document, m, ω and represent the number of document basic vectors, the number of document style vectors and the number of temperature sentiment vectors respectively, and j, m, ω and are positive integers. The index word vectors and the description word vectors of the target user and the document basic vectors and the document description vectors of each fusion document are substituted into the vector correlation degree matching model, and the search correlation evaluation coefficient L j of the target user and the jth fusion document is calculated through the expression of the vector correlation degree matching model: , where M and N represent the total number of index word vectors and the total number of document basic vectors respectively, and represent the total number of style description word vectors of the target user and the total number of document style description vectors respectively, and ο and θ represent the total number of sentiment description word vectors of the target user and the total number of document sentiment description vectors respectively. The target user and the search association evaluation coefficient of each fusion document are sorted from large to small, and the display order of each fusion document for the target user's current search is obtained accordingly.

8. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 7, wherein, The target user's reading preference is analyzed based on the target user's historical reading records, and each search recommended document is recommended and arranged based on the reading preference, and the specific process is as follows: The historical reading records of the target user are obtained from the data center, the reading times and reading time of the target user for each type of document are obtained based on the historical reading records of the target user, the reading preference evaluation model is substituted, the reading preference evaluation coefficient of the target user for each document is output through the expression of the reading preference evaluation model, and the ratio of the reading preference evaluation coefficient of the target user for each document is obtained, and the display form of each fusion document on the target user's current search page is set according to the ratio of the reading preference evaluation coefficient of the target user for each document.

9. The intelligent multimodal document fusion and personalized retrieval system as claimed in claim 8, wherein, The reading association between the target user and other users is analyzed based on the target user's historical reading records, and the homepage display document recommendation for the target user is carried out, and the specific analysis process is as follows: The description word vector of each user is analyzed according to the analysis method of the target user's description word vector, the reading description similarity evaluation coefficient between the target user and each user is analyzed based on the target user's description word vector and each user's description word vector, each user whose reading description similarity coefficient is higher than the set reading description similarity coefficient threshold is recorded as each reference user, each fusion document read by each reference user in the preset time period is obtained, and the homepage display document for the target user is randomly recommended.

10. A method of intelligent multi-modal document fusion and personalized retrieval performed by the intelligent multi-modal document fusion and personalized retrieval system of any one of claims 1-9, characterized in that, It includes: Step one, extract each basic keyword and each description keyword of each document, and then fuse each document according to each basic keyword and each description keyword to obtain each fusion document; Step two, analyze the target user's description type according to the target user's historical reading records, and then describe the target user according to the target user's description type, and when the target user searches, analyze the target user's each search recommended document according to the target user's description image and each search content, and analyze the target user's reading preference based on the target user's historical reading records, so as to recommend and arrange each search recommended document based on the reading preference; Step three, analyze the reading association between the target user and other users based on the target user's historical reading records, and then recommend the homepage display document for the target user.

Images