Library intelligent management method and platform based on Internet

Abstract

The invention discloses an intelligent library management method and platform based on the Internet, and relates to the technical field of the Internet. A trend semantic vector Trd is constructed through a structured hot word set Kws, a book semantic vector set BVecSet is generated based on a book text field set BTxt, a semantic similarity value Sim is calculated, and a library semantic vector set BVecSet is generated based on a library text field set BTxt; and automatic semantic matching of the book content and the Internet hotspot is realized. According to the method, a social propagation intensity value Soc and a borrowing trend change value Dyn are combined, a comprehensive popularity prediction function Hot is constructed, a book popularity set HotSet is formed, a recommended book set Rec is screened out by comparing the book popularity set HotSet with a recommendation threshold value Thr, and compared with lagging response depending on historical borrowing data, external user behavior feedback is introduced through the social propagation intensity value Soc, so that the book popularity prediction function HotSet is obtained. And a multi-factor recommendation mechanism is enhanced in combination with a comprehensive popularity prediction function Hot, so that the accuracy and foresight of book recommendation are improved, and active response to external hotspots is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet, in particular to an Internet-based library intelligent management method and platform. BACKGROUND

[0002] In today's highly developed information technology, modern libraries are gradually transforming from traditional "book access and borrowing space" into "information resource service platforms" that integrate information management, resource scheduling, user service, intelligent recommendation and other functions. In particular, in the field of "intelligent information services" in digital libraries, Internet-based data-driven management methods are becoming a key breakthrough. Such methods, with network behavior analysis, user portrait modeling, real-time hot spot tracking and other core capabilities, are promoting the transition of library services from "static access" to "dynamic intelligence". Book recommendation services, as the most intuitive interactive module for users, play a role in resource guidance, knowledge distribution and interest stimulation. In highly digitalized scenarios such as universities and city comprehensive libraries, predicting users' potential borrowing interests through systematic means has become an important issue to improve library usage and reader satisfaction.

[0003] In the current mainstream library information system, book recommendation logic often relies on traditional algorithm models such as historical borrowing data mining, user borrowing record analysis, and reader preference tag matching. Such models are usually lagging behind user behavior, and only after obvious borrowing behavior data appears can "hot books" be pushed to the recommendation page. Therefore, the core problem of book recommendation is the passive response to borrowing behavior, rather than active prediction based on environmental changes. For example, after the movie "Oppenheimer" was released, many users flocked to the library to find books related to atomic energy, Manhattan Project, and nuclear physics. However, due to the fact that traditional systems do not have real-time awareness of Internet hot words and related social dynamics, the recommendation mechanism often displays related books when the heat has already begun to decline, missing the best recommendation opportunity and the peak of user attention. This directly leads to core problems such as "lack of timeliness, lagging perception, and inaccurate matching" in the recommendation system, especially when facing "sudden" "event-driven" reader interest fluctuations, the system is difficult to respond efficiently, affecting book utilization efficiency and reader experience quality. SUMMARY

[0004] To overcome the deficiencies of the prior art, the present application provides an Internet-based library intelligent management method and platform, which solves the problems mentioned in the background art.

[0005] To achieve the above purpose, the present application is implemented by the following technical scheme: an Internet-based library intelligent management method, comprising the following steps: S1, an Internet hotword collection module, a hotword semantic conversion module, a library matching module, an Internet propagation monitoring module, and a book recommendation module; S2, based on each hotword in the structured hotword set Kws, using a semantic embedding model to encode, obtaining a word vector Emb, after traversing all hotwords in the structured hotword set Kws, constructing a trend semantic vector Trd; S3, extracting the book text field set BTxt of the books in the library collection database, performing semantic encoding, constructing the book semantic vector set BVecSet of the books, and performing cosine similarity calculation on the book semantic vector set BVecSet and the trend semantic vector Trd to obtain a semantic similarity value Sim; S4, monitoring the external propagation behavior of each book on the Internet platform, collecting the propagation data of the text field set of the book on the Internet platform, and calculating a social propagation intensity value Soc; S5, based on the semantic similarity value Sim, the social propagation intensity value Soc, and the borrowing trend change value Dyn, constructing a comprehensive hotness prediction function Hot of the books, after integrating the comprehensive hotness prediction functions Hot of all the books, obtaining a hotness set HotSet, and performing traversal comparison with a preset recommendation threshold Thr, filtering all the books according to the comparison result to obtain a recommended book set Rec.

[0006] Preferably, the S1 includes S11; S11, through a set data collection interface, accessing multiple mainstream Internet platforms including microblog, B station, Zhihu and Douban Internet platforms, real-time extracting keyword content data containing high-frequency trend information, the keyword content data including hot search title, topic label, post abstract, video title and popular comment content, uniformly collecting to form a text set WTxt, and performing word segmentation processing on each text data in the text set WTxt, the word segmentation processing using a natural language processing tool to segment keywords to obtain a keyword candidate set, and performing merging operation on the keyword candidate set obtained after processing all text data in the text set WTxt to obtain an original hotword set Raw.

[0007] Preferably, the S1 includes S12; S12, performing structured processing on the obtained original hotword set Raw, removing semantically invalid, low-frequency and noisy words, and forming a structured hotword set Kws; The structured processing includes deduplication processing, word frequency statistics and screening, and invalid word filtering, after the structured processing is completed, the keyword items that meet the word frequency condition and do not belong to stop words are retained, marked as hotwords, and then composed into the structured hotword set Kws; The deduplication processing eliminates the repetition in the original hotword set Raw and generates a unique number for each keyword; The word frequency statistics and screening statistics the frequency of occurrence of each keyword in the original hotword set Raw, and compares the frequency of occurrence with the preset minimum word frequency threshold. The keywords that meet the condition of frequency of occurrence ≥ minimum word frequency threshold are retained to generate the frequency screening set RawFil after frequency filtering; The invalid word filtering filters out the syntax conjunction words, auxiliary words and empty adverb related semantic load keywords in the frequency screening set RawFil by calling the preset stop words Stop.

[0008] Preferably, S2 includes S21 and S22; S21, based on each hotword in the structured hotword set Kws, uses a semantic embedding model for encoding processing, calls a semantic embedding function Enc, traverses each hotword in the structured hotword set Kws, and inputs the hotword into the semantic embedding model as input data for processing, outputs the semantic vector Svec(j) of the jth hotword, integrates the semantic vectors of all hotwords, and combines them into the word vector set Emb={Svec(1), Svec(2), …, Svec(j)|j∈N}, N represents the total number of hotwords; Wherein, the semantic embedding model is used for converting the jth hotword Kws(j) into a semantic vector representation, and the semantic embedding model includes using Word2Vec, BERT, GloVe and BERT derivative model based on Chinese field fine-tuning for encoding processing, and calling the semantic embedding function Enc in the form of calling the semantic embedding function Enc(Kws(j)), Kws(j) represents the jth hotword in the structured hotword set Kws; S22, vector aggregation processing is performed on the obtained word vector set Emb to generate a semantic trend vector representing the semantic trend of the entire hotword set, which is marked as trend semantic vector Trd; The trend semantic vector Trd is obtained by using The weighted average aggregation strategy formula is obtained, wherein w(j) represents the weight of the jth hotword.

[0009] Preferably, S3 includes S31; S31, each book in the library collection database is extracted, and the field data of each book is extracted in turn, the field data including book title, book label, book classification and book abstract, and the field data of each book is sequentially spliced to form the complete text field of the ith book, which is marked as the book text field btxt(i) of the ith book. After integrating all the books, the book text field set Btxt={btxt(1), btxt(2), …, btxt(i)|i∈M} is obtained, M represents the total number of books. The constructed book text field set BTxt is encoded using the semantic embedding model in step S2, and a semantic embedding function Enc is called to generate a vector representation of each book in the semantic space, marked as the book semantic vector BVec(i) of the ith book. After integrating all book vectors, the book semantic vector set BVecSet is obtained; The book semantic vector set BVecSet has the specific form BVecSet = {BVec(1), BVec(2), …, BVec(i) | i ∈ M}.

[0010] Preferably, S3 includes S32; S32, based on the book semantic vector set BVecSet and the trend semantic vector Trd, the semantic similarity between the book semantic vector BVec(i) of the ith book and the trend semantic vector Trd is calculated, marked as the semantic similarity value Sim(i) of the ith book. After performing the semantic similarity operation on each vector in the book semantic vector set BVecSet, the book trend correlation set SimSet is formed; The book trend correlation set SimSet has the specific form SimSet = {Sim(1), Sim(2), …, Sim(i) | i ∈ M}, M represents the total number of books; The semantic similarity value Sim(i) of the ith book is calculated by the following cosine similarity formula: ; In the formula, ||·|| represents the Euclidean norm of the vector.

[0011] Preferably, S4 includes S41 and S42; S41, based on the book text field btxt(i) of each book in the book text field set BTxt, the propagation behavior of the content including the book text field btxt(i) in multiple Internet platforms is collected and counted in real time by calling a preset Internet platform monitoring interface, and the forwarding number Rtw(i), the mention frequency Mnt(i) and the active sharing number Shr(i) of the ith book are obtained. The propagation behavior feature vector MVec(i) = {Rtw(i), Mnt(i), Shr(i)} of the ith book is constructed; By traversing the book text field set BTxt, the propagation behavior feature vector MVec(i) of each book is extracted, and the social propagation behavior feature set Mset is constructed; The social communication behavior feature set Mset is specifically a social communication behavior feature set Mset={MVec(1), MVec(2), …, MVec(i)|i∈M}, and M represents the total number of books. S42, the social communication behavior feature set Mset is normalized to eliminate differences between different dimensions, and the normalized standard forwarding number NRtw(i), the normalized standard mention frequency NMnt(i) and the normalized active sharing number NShr(i) of the i-th book are subjected to communication intensity evaluation processing, to obtain the social communication intensity value Soc(i) of the i-th book on the Internet platform. After integrating the social communication intensity values Soc(i) of all books, a social communication intensity set SocSet={Soc(1), Soc(2), …, Soc(i)|i∈N} is obtained. The social communication intensity value Soc(i) of the i-th book is obtained by the following calculation formula: ; In the formula, s1, s2 and s3 represent the weight coefficients of the standard forwarding number NRtw(i), the standard mention frequency NMnt(i) and the standard active sharing number NShr(i) of the i-th book respectively, and s1+s2+s3=1. The specific values are set by the user.

[0012] Preferably, the S5 comprises S51; S51, by extracting the borrowing amount of each book in a fixed period in the book borrowing record, the borrowing trend change set DynSet of each book is counted, and the obtained social communication intensity set SocSet is normalized to eliminate differences between different dimensions, and then fitted with the book trend correlation set SimSet, to construct the comprehensive hotness prediction function Hot(i) of the i-th book. After integrating the comprehensive hotness prediction functions Hot(i) of all books, a hotness set HotSet={Hot(1), Hot(2), …, Hot(i)|i∈N} is obtained. The comprehensive hotness prediction function Hot(i) of the i-th book is obtained by the following calculation formula: ; In the formula, log represents the logarithmic function.

[0013] Preferably, the S5 comprises S52; S52, based on the hotness set HotSet, a traversal comparison operation is performed, the comprehensive hotness prediction function Hot(i) of the i-th book is compared with a preset recommendation threshold Thr, and all recommended books are selected according to the comparison result to obtain a recommended book set Rec. The recommended books are obtained through the following comparison screening mode: When the comprehensive hotness prediction function Hot(i) of the i-th book ≥ the recommended threshold Thr, it is determined that the i-th book has potential hotness to enter the homepage recommendation area, the i-th book is marked as a recommended book, and is counted into the recommended book set Rec; When the comprehensive hotness prediction function Hot(i) of the i-th book < the recommended threshold Thr, it is determined that the i-th book does not have potential hotness to enter the homepage recommendation area, the i-th book is marked as a non-recommended book, and is not counted into the recommended book set Rec.

[0014] An internet-based library intelligent management platform, comprising an internet hotword collection module, a hotword semantic conversion module, a library matching module, an internet propagation monitoring module and a book recommendation module; The internet hotword collection module, the hotword semantic conversion module, the library matching module, the internet propagation monitoring module and the book recommendation module; The hotword semantic conversion module encodes each hotword in the structured hotword set Kws using a semantic embedding model to obtain a word vector Emb, and constructs a trend semantic vector Trd by traversing all hotwords in the structured hotword set Kws; The library matching module extracts a book text field set BTxt of books in a library collection database, performs semantic encoding, and constructs a book semantic vector set BVecSet of the books; and performs cosine similarity calculation on the book semantic vector set BVecSet and the trend semantic vector Trd to obtain a semantic similarity value Sim; The internet propagation monitoring module monitors the external propagation behavior of each book on the internet platform, collects propagation data of the text field set of the book on the internet platform, and calculates a social propagation intensity value Soc; The book recommendation module constructs a comprehensive hotness prediction function Hot of the books based on the semantic similarity value Sim, the social propagation intensity value Soc and a borrowing trend change value Dyn, obtains a hotness set HotSet after integrating the comprehensive hotness prediction functions Hot of all the books, and performs traversal comparison with a preset recommended threshold Thr to screen all the books according to the comparison result and obtain a recommended book set Rec.

[0015] The present application provides an internet-based library intelligent management method and platform, which has the following beneficial effects: (1) Through the structured hot word set Kws, the trend semantic vector Trd is constructed, and based on the book text field set Btxt, the book semantic vector set BVecSet is generated, the semantic similarity value Sim is calculated, and the automatic semantic matching of book content and Internet hotspots is realized. Combined with the social communication intensity value Soc and the borrowing trend change value Dyn, the comprehensive hotness prediction function Hot is constructed, the book hotness set HotSet is formed, and through comparison with the recommended threshold Thr, the recommended book set Rec is screened out. Compared with the existing technology which relies on historical borrowing data for lag response, this method uses the trend semantic vector Trd and the semantic similarity value Sim to improve the real-time of recommendation, introduces external user behavior feedback through the social communication intensity value Soc, and strengthens the multi-factor recommendation mechanism by combining the comprehensive hotness prediction function Hot, effectively improves the accuracy and forward-looking of book recommendation, and realizes the active response and dynamic push to external hotspots.

[0016] (2) By extracting the book text field set Btxt in the library collection database, and using the same semantic embedding model as step S2 for encoding processing, the book semantic vector set BVecSet is constructed, and then the cosine similarity between the book semantic vector set BVecSet and the trend semantic vector Trd is calculated to generate the semantic similarity value Sim of each book, realizing the deep alignment and semantic quantitative matching of book content and Internet semantic trend. Unlike traditional technology which only relies on book metadata or topic tags for keyword comparison, this method uses the trend semantic vector Trd as a high-dimensional semantic benchmark, effectively bridges the semantic gap between Internet hot words and book content, and enables the book recommendation mechanism to understand the context semantics and capture similar semantic topics.

[0017] (3) By monitoring the external communication behavior of the book text field set Btxt in the Internet platform, the communication feature vector set MSet is constructed, and based on the number of forwarding, the frequency of mentioning and the number of active sharing, the social communication intensity value Soc is calculated. The social feedback dimension is introduced for the comprehensive hotness evaluation of the book, which is different from the existing recommendation mechanism which only relies on library borrowing or static score information. The introduced social communication intensity value Soc can dynamically reflect the user attention and interaction activity of the book on social media, realizing the real-time quantitative supplement of the external hotness of the book. By constructing the communication feature vector set MSet, the attention trend behind the user's spontaneous communication behavior is captured, effectively improving the agility of the recommendation system in responding to hotspots and the ability of external hotspot linkage, thereby enhancing the real-time, sensitivity and public feedback adaptability of the recommendation results.

[0018] (4) The semantic similarity value Sim, the social communication intensity value Soc and the borrowing trend change value Dyn are jointly modeled to construct a comprehensive hotness prediction function Hot of the book, the hotness prediction function Hot values of all books are integrated to form a hotness set HotSet, and the hotness set HotSet is compared with a recommendation threshold Thr to screen a recommended book set Rec to form a final dynamic recommendation result. Compared with the existing recommendation system which usually scores and sorts based on a single index, the unified hotness evaluation function Hot is constructed to realize the uniform modeling of multiple source heterogeneous indexes, and ensure that the recommendation judgment has a more comprehensive user behavior perspective and semantic understanding ability. Through the linkage judgment mechanism of the hotness set HotSet and the recommendation threshold Thr, the recommendation boundary can be adjusted in real time according to the trend intensity and user interest, the selection of the recommended book set Rec is more targeted and timely, and the response capability of the recommendation system to external hotspots and user demand changes is enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A step schematic diagram of the intelligent management method of the library based on the Internet according to the present application; Figure 2 A block diagram schematic diagram of the intelligent management platform of the library based on the Internet according to the present application; Figure 3 A trend schematic diagram of the social communication intensity value Soc of the book. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part 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] Embodiment 1: The present application provides an intelligent management method of a library based on the Internet, please refer to Figure 1 , comprising the following steps: S1, an Internet hotword acquisition module, a hotword semantic conversion module, a library matching module, an Internet communication monitoring module and a book recommendation module; S2, based on each hotword in the structured hotword set Kws, the semantic embedding model is used for coding to obtain the word vector Emb, and after traversing all hotwords in the structured hotword set Kws, the trend semantic vector Trd is constructed; S3, extract the book text field set Btxt of the books in the library collection database, perform semantic coding, construct the book semantic vector set BVecSet of the books; and perform cosine similarity calculation on the book semantic vector set BVecSet and the trend semantic vector Trd to obtain the semantic similarity value Sim; S4, monitor the external dissemination behavior of each book in the Internet platform, collect the dissemination data of the text field set of the book in the Internet platform, and calculate the social dissemination intensity value Soc; S5, based on the semantic similarity value Sim, the social dissemination intensity value Soc and the borrowing trend change value Dyn, construct the comprehensive hotness prediction function Hot of the books, and after integrating the comprehensive hotness prediction function Hot of all the books, obtain the hotness set HotSet, and compare it with the preset recommendation threshold Thr, filter all the books according to the comparison result, and obtain the recommended book set Rec.

[0022] In this embodiment, the trend semantic vector Trd is constructed by the structured hot word set Kws, and the book semantic vector set BVecSet is generated based on the book text field set Btxt, the semantic similarity value Sim is calculated, and the automatic semantic matching of the book content and the Internet hotspots is realized. Combined with the social dissemination intensity value Soc and the borrowing trend change value Dyn, the comprehensive hotness prediction function Hot is constructed, the book hotness set HotSet is formed, and the recommended book set Rec is selected by comparing with the recommendation threshold Thr. Compared with the lag response of the existing technology depending on the historical borrowing data, this method improves the real-time performance of the recommendation by using the trend semantic vector Trd and the semantic similarity value Sim, introduces the external user behavior feedback through the social dissemination intensity value Soc, and strengthens the multi-factor recommendation mechanism by combining the comprehensive hotness prediction function Hot, effectively improves the accuracy and forward-looking of the book recommendation, and realizes the active response and dynamic push of the external hotspots.

[0023] Embodiment 2: Specifically, the S1 includes S11; S11, through the set data acquisition interface, access multiple mainstream Internet platforms including microblog, B station, Zhihu and Douban Internet platforms, real-time extract keyword content data containing high-frequency trend information, the keyword content data includes hot search title, topic label, post abstract, video title and hot comment content, after unified collection, form a text set Wtxt, and perform word segmentation processing on each text data in the text set Wtxt, the word segmentation processing utilizes natural language processing tool to perform word segmentation, obtains a keyword candidate set, and performs merging operation on the keyword candidate set obtained after processing all text data in the text set Wtxt, to obtain the original hot word set Raw; The high-frequency trend information refers to topic content accessed, discussed, liked, searched or forwarded by a large number of users in a unit of time, and the heat thereof is usually reflected by platform ranking, interaction amount and the like.

[0024] The S1 includes S12; The S12 performs structured processing on the obtained original hotword set Raw, removes semantically invalid, excessively low frequency and noisy word items, and forms a structured hotword set Kws with modeling value; The structured processing includes deduplication processing, word frequency statistics and screening, and invalid word item filtering. After the structured processing is completed, the key word items that meet the word frequency condition and do not belong to stop words are retained, marked as hotwords, and then constitute the structured hotword set Kws; The deduplication processing removes the repetition in the original hotword set Raw, and generates a unique number for each key word, which is used for subsequent word frequency statistics and screening; The word frequency statistics and screening statistics the occurrence frequency of each key word in the original hotword set Raw, and then compares the occurrence frequency with a preset minimum word frequency threshold. The key word that meets the condition of occurrence frequency ≥ minimum word frequency threshold is retained, and a frequency screening set RawFil after frequency filtering is generated; The invalid word item filtering filters out the syntax connecting words, auxiliary words and semantically empty adverb-related key word items in the frequency screening set RawFil by calling a preset stop word Stop.

[0025] In this embodiment, by accessing the interfaces of the microblog, B station, Zhihu and Douban Internet platforms, key word content data is collected and a unified text set WTxt is formed. Then, the natural language processing technology is used for word segmentation processing to generate an original hotword set Raw. Through the execution of deduplication, word frequency statistics screening and invalid word item filtering operations, the redundant, noisy and semantically empty key word items in the original hotword set Raw are removed. Finally, a structured hotword set Kws is constructed, which can be used as a high-quality trend semantic input set to provide a stable and effective semantic basis for the subsequent semantic encoding and matching module. Unlike the existing technology which depends on manual screening or simple word frequency extraction, the hotword set Kws after structured processing in this scheme clearly ensures that the key words have content carrying property and trend representativeness in semantics, significantly improving the effectiveness and anti-interference ability of the entire system in processing Internet hot spot semantics. The processing flow can effectively reduce the interference of platform noise words, false hot spots or syntax invalid content on the subsequent trend semantic modeling results, enhance the recognition ability of the system to the real user focus, and provide a more accurate pre-semantic corpus basis for the intelligent recommendation module.

[0026] Embodiment 3: Specifically, the S2 includes S21 and S22. S21, based on each hot word in the structured hot word set Kws, using a semantic embedding model for encoding processing, calling a semantic embedding function Enc, traversing each hot word in the structured hot word set Kws, as input data into the semantic embedding model for processing, outputting the semantic vector Svec(j) of the jth hot word, after integrating the semantic vectors of all hot words, merging into the word vector set Emb={Svec(1), Svec(2), …, Svec(j)|j∈N}, N represents the total number of hot words; Wherein, the semantic embedding model is used for converting the jth hot word Kws(j) into a semantic vector representation, the semantic embedding model includes using Word2Vec, BERT, GloVe and BERT derived model based on Chinese field fine-tuning for encoding processing, calling semantic embedding function Enc, the specific form of calling semantic embedding function Enc(Kws(j)), Kws(j) represents the jth hot word in the structured hot word set Kws, the semantic embedding model can map the words with similar semantics into adjacent vectors in space by learning the co-occurrence relationship or context dependence of words in a large amount of text, realizing the mathematical computability of semantic relationship. S22, performing vector aggregation processing on the obtained word vector set Emb to generate a semantic trend vector representing the whole hot word set, marked as trend semantic vector Trd. The trend semantic vector Trd is obtained by using The weighted average aggregation strategy formula is obtained, wherein w(j) represents the hotness weight of the jth hot word, which can be set by the user based on the normalized word frequency and network search volume.

[0027] In this embodiment, for each hot word in the structured hot word set Kws, the semantic embedding function Enc is called and multiple semantic embedding models are used for encoding processing to obtain the semantic vector representation of the hot word, and the word vector set Emb is formed; the weighted average aggregation operation is performed on the word vector set Emb based on the normalized weight of word frequency and network search volume, and finally the trend semantic vector Trd is generated as the unified semantic reference for subsequent book semantic similarity calculation and matching. Unlike the frequency analysis of hot information at the keyword level in the prior art, this scheme realizes high-dimensional semantic compression and trend abstraction of the hot word set Kws by constructing the trend semantic vector Trd, so that the system can grasp the overall Internet public opinion semantic direction and overcome the problems of one-sided hot spot interpretation and context disconnection caused by single keyword semantic limitation. At the same time, the trend semantic vector Trd has a quantifiable, comparable and traceable mathematical form, which significantly enhances the semantic consistency and dynamic adaptability of the semantic matching and semantic recommendation model, and provides a stable and highly abstract semantic core representation basis for subsequent book content semantic alignment.

[0028] Embodiment 4: Specifically, the S3 comprises S31; S31, extract each book in the library collection database, and sequentially extract the field data of each book, the field data comprising book title, book label, book classification and book abstract, and sequentially splice the field data of each book to form the complete text field of the i-th book, marked as the book text field btxt(i) of the i-th book, integrate all books to obtain the book text field set Btxt={btxt(1), btxt(2), …, btxt(i)|i∈M}, M represents the total number of books; The constructed book text field set Btxt is encoded using the semantic embedding model in step S2, and a semantic embedding function Enc is called to generate the vector representation of each book in the semantic space, marked as the book semantic vector BVec(i) of the i-th book, and the vector of all books is integrated to obtain the book semantic vector set BVecSet; Wherein, the semantic embedding function Enc is called in the form of calling the semantic embedding function Enc(btxt(i)); The book semantic vector set BVecSet is in the form of book semantic vector set BVecSet={BVec(1), BVec(2), …, BVec(i)|i∈M}.

[0029] The S3 comprises S32; S32, based on the book semantic vector set BVecSet and the trend semantic vector Trd, calculate the semantic similarity between the book semantic vector BVec(i) of the i-th book and the trend semantic vector Trd, marked as the semantic similarity value Sim(i) of the i-th book, after performing the semantic similarity operation on each vector in the book semantic vector set BVecSet, form the book trend correlation set SimSet; The book trend correlation set SimSet is in the form of book trend correlation set SimSet={Sim(1), Sim(2), …, Sim(i)|i∈M}, M represents the total number of books; The semantic similarity value Sim(i) of the i-th book is calculated by the following cosine similarity formula: ; In the formula, ||·|| represents the Euclidean norm of a vector, the semantic similarity value Sim(i) of the i-th book ∈ [-1, 1] represents the cosine value of the angle between the book semantic vector BVec(i) of the i-th book and the trend semantic vector Trd, and the higher the similarity, the more consistent the semantic vector BVec(i) of the i-th book is with the trend semantic vector Trd of the current hot trend. The purpose of the formula is to use the obtained semantic similarity value Sim(i) as one of the scoring bases for subsequent book screening, sorting and homepage recommendation.

[0030] An example of calculating the semantic similarity between a book semantic vector and a trend semantic vector is described below. Taking book i as an example, the book text field btxt(i) is processed by a semantic embedding model to obtain The book semantic vector BVec(i) = [0.52, 0.13, 0.69, 0.34, 0.27]. The trend semantic vector Trd is generated by encoding the structured hot word set Kws: The trend semantic vector Trd = [0.41, 0.11, 0.65, 0.38, 0.30]. The cosine similarity formula is used for semantic matching: Sim(i) ≈ 0.8862 / 0.7569 ≈ 1.17. In this embodiment, the book text field set BTxt of the books in the library database is extracted, and the same semantic embedding model as in step S2 is used for encoding processing to construct a book semantic vector set BVecSet. Then, the cosine similarity between the book semantic vector set BVecSet and the trend semantic vector Trd is calculated to generate the semantic similarity value Sim of each book, thereby realizing the deep alignment and semantic quantitative matching of the book content and the Internet semantic trend. Unlike the traditional technology that only relies on book metadata or topic tags for keyword comparison, this step uses the trend semantic vector Trd as a high-dimensional semantic reference to effectively bridge the semantic gap between the Internet hot words and the book content, so that the book recommendation mechanism has the ability to understand the context semantics and capture similar semantic topics. Through the introduction of the semantic similarity value Sim, the system can not only accurately quantify the text semantics, but also greatly improve the adaptation accuracy of semantic recommendation and the robustness of content matching, thereby strengthening the recognition ability of the recommendation mechanism for complex semantic associations.

[0031] Embodiment 5: Please refer to Figure 1 and Figure 3 Specifically, the S4 includes S41 and S42. S41, based on the book text field set Btxt each book text field btxt(i) of each book, by calling the preset Internet platform monitoring interface, the content including the book text field btxt(i), the propagation behavior in multiple Internet platforms is collected and counted in real time, the forwarding number Rtw(i), the mention frequency Mnt(i) and the active sharing number Shr(i) of the i-th book are obtained, and the propagation behavior feature vector MVec(i) of the i-th book is constructed. By traversing the book text field set Btxt, the propagation behavior feature vector MVec(i) is extracted for each book, and the social propagation behavior feature set Mset is constructed. The specific form of the social propagation behavior feature set Mset is Mset={MVec(1), MVec(2), …, MVec(i)|i∈M}, M represents the total number of books. Among them, the forwarding number Rtw(i) of the i-th book represents the cumulative number of times that the book text field btxt(i) is forwarded and reprinted by the user. The mention frequency Mnt(i) of the i-th book represents the frequency of explicitly mentioning the keywords in the book text field btxt(i) by the user in the dynamic, comment or article. The active sharing number Shr(i) of the i-th book represents the number of times that the user actively spreads the content related to the book text field btxt(i) as an information source. External diffusion includes book recommendation posts and video introductions. S42, the social propagation behavior feature set Mset is normalized to eliminate the difference between different dimensions, and the normalized standard forwarding number NRtw(i), the normalized standard mention frequency NMnt(i) and the normalized active sharing number NShr(i) of the i-th book are processed for propagation intensity evaluation, and the social propagation intensity value Soc(i) of the i-th book in the Internet platform is obtained. After integrating the social propagation intensity values Soc(i) of all books, the social propagation intensity set SocSet={Soc(1), Soc(2), …, Soc(i)|i∈N} is obtained. The social propagation intensity value Soc(i) of the i-th book is obtained by the following calculation formula: ; In the formula, s1, s2 and s3 represent the weight coefficients of the standard forwarding number NRtw(i), the standard mention frequency NMnt(i) and the standard active sharing number NShr(i) of the i-th book respectively, and s1+s2+s3=1. The specific value is set by the user.

[0032] In this embodiment, by monitoring the external propagation behavior of the book text field set BTxt in the Internet platform, a propagation feature vector set MSet is constructed, and the social propagation strength value Soc is calculated based on the number of forwarding times, the number of mentions, and the number of active sharing, etc. The social feedback dimension is introduced for the comprehensive hotness evaluation of the book. Unlike the existing recommendation mechanism which only relies on the library borrowing or static scoring information, the introduced social propagation strength value Soc can dynamically reflect the user attention and interaction activity of the book on the social media, and realize the real-time quantification supplement of the external hotness of the book. By constructing the propagation feature vector set MSet, the system can capture the attention trend behind the spontaneous propagation behavior of the user, effectively improve the agility of the recommendation system in responding to hot spots and the ability of external hot spot linkage, and thus enhance the real-time, sensitivity and public feedback adaptability of the recommendation results.

[0033] Embodiment 6: Specifically, the S5 comprises S51; S51, by extracting the borrowing amount of each book in a fixed period in the book borrowing record, the borrowing trend change set DynSet of each book is counted, and after normalization processing with the obtained social propagation strength set SocSet, the difference between different dimensions is eliminated, and then fitting processing is performed with the book trend relevance set SimSet, the comprehensive hotness prediction function Hot(i) of the i-th book is constructed, and after integrating the comprehensive hotness prediction functions Hot(i) of all books, the hotness set HotSet={Hot(1), Hot(2), …, Hot(i)|i∈N} is obtained; The comprehensive hotness prediction function Hot(i) of the i-th book is obtained by the following calculation formula: ; In the formula, log represents the logarithmic function.

[0034] The S5 comprises S52; S52, based on the hotness set HotSet, a traversal comparison operation is performed, the comprehensive hotness prediction function Hot(i) of the i-th book is compared with the preset recommendation threshold Thr, and all recommended books are selected according to the comparison result, and the recommended book set Rec is obtained; The recommended book is obtained by the following comparison and selection method: When the comprehensive hotness prediction function Hot(i) of the i-th book is greater than or equal to the recommendation threshold Thr, it is determined that the i-th book has potential hotness to enter the front page recommendation area, the i-th book is marked as a recommended book, and is counted into the recommended book set Rec; When the comprehensive hotness prediction function Hot(i) of the i-th book is less than the recommendation threshold Thr, it is determined that the i-th book does not have potential hotness to enter the homepage recommendation area, and the i-th book is marked as a non-recommended book, which is not included in the recommended book set Rec.

[0035] In this embodiment, the semantic similarity value Sim, the social propagation intensity value Soc and the borrowing trend change value Dyn are jointly modeled to construct a comprehensive hotness prediction function Hot of the book, the hotness prediction function Hot values of all books are integrated to form a hotness set HotSet, and the hotness set HotSet is compared with a recommendation threshold Thr to screen a recommended book set Rec to form a final dynamic recommendation result. Compared with the existing recommendation system which usually scores and sorts based on a single indicator, a unified hotness evaluation function Hot is constructed to realize the modeling of multi-source heterogeneous indicators while ensuring that the recommendation judgment has a more comprehensive user behavior perspective and semantic understanding ability. Through the linkage determination mechanism of the hotness set HotSet and the recommendation threshold Thr, the recommendation boundary can be adjusted in real time according to the trend intensity and user interest, ensuring that the selection of the recommended book set Rec is more targeted and timely, and enhancing the response ability of the recommendation system to external hotspots and changes in user demand.

[0036] Embodiment 7: An Internet-based intelligent management platform for a library, please refer to Figure 2 , in particular: comprising an Internet hotword collection module, a hotword semantic conversion module, a library matching module, an Internet propagation monitoring module and a book recommendation module; The Internet hotword collection module, the hotword semantic conversion module, the library matching module, the Internet propagation monitoring module and the book recommendation module; The hotword semantic conversion module encodes each hotword in the structured hotword set Kws using a semantic embedding model to obtain a word vector Emb, and constructs a trend semantic vector Trd by traversing all hotwords in the structured hotword set Kws; The library matching module extracts a book text field set BTxt of books in the library collection database, performs semantic encoding, and constructs a book semantic vector set BVecSet of the books; and performs cosine similarity calculation on the book semantic vector set BVecSet and the trend semantic vector Trd to obtain a semantic similarity value Sim; The Internet propagation monitoring module monitors the external propagation behavior of each book on the Internet platform, collects propagation data of the text field set of the book on the Internet platform, and calculates a social propagation intensity value Soc; The book recommendation module constructs a comprehensive hotness prediction function Hot of the books based on the semantic similarity value Sim, the social transmission strength value Soc and the borrowing trend change value Dyn. After integrating the comprehensive hotness prediction functions Hot of all the books, a hotness set HotSet is obtained, and a preset recommendation threshold Thr is compared. According to the comparison result, all the books are filtered to obtain a recommended book set Rec.

[0037] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An Internet-based intelligent management method for a library, characterized in that: The method comprises the following steps: S1, an internet hotword collection module, a hotword semantic conversion module, a library matching module, an internet propagation monitoring module, and a book recommendation module; S2, based on each hotword in the structured hotword set Kws, a semantic embedding model is used for coding to obtain a word vector Emb, and after traversing all hotwords in the structured hotword set Kws, a trend semantic vector Trd is constructed; S3, a book text field set BTxt of books in a library collection database is extracted, semantic coding is performed, a book semantic vector set BVecSet of books is constructed, and cosine similarity calculation is performed between the book semantic vector set BVecSet and the trend semantic vector Trd to obtain a semantic similarity value Sim; S4, the external propagation behavior of each book on an internet platform is monitored, the propagation data of the text field set of the book on the internet platform is collected, and a social propagation intensity value Soc is calculated; S5, based on the semantic similarity value Sim, the social propagation intensity value Soc, and the borrowing trend change value Dyn, a comprehensive hotness prediction function Hot of the book is constructed, after integrating the comprehensive hotness prediction functions Hot of all books, a hotness set HotSet is obtained, and traversal comparison is performed between the hotness set HotSet and a preset recommendation threshold Thr, all books are selected according to the comparison result to obtain a recommended book set Rec.

2. The intelligent management method of an Internet-based library according to claim 1, characterized in that: The S1 comprises S11; S11, through a set data collection interface, a plurality of mainstream internet platforms including microblog, B station, Zhihu and Douban internet platforms are accessed, keyword content data containing high-frequency trend information is extracted in real time, the keyword content data includes hot search title, topic label, post abstract, video title and popular bullet screen content, is uniformly summarized to form a text set WTxt, and word segmentation processing is performed on each text data in the text set WTxt, the word segmentation processing utilizes a natural language processing tool to perform word segmentation to obtain a keyword candidate set, and the keyword candidate set obtained after processing all text data in the text set WTxt is subjected to a merging operation to obtain an original hotword set Raw.

3. The intelligent management method of an Internet-based library according to claim 2, characterized in that: The S1 comprises S12; S12, the obtained original hotword set Raw is subjected to structured processing, processing of eliminating semantically invalid, excessively low frequency and noise words is performed, and a structured hotword set Kws is formed; The structured processing comprises de-duplication processing, word frequency statistics and screening, and invalid word filtering, after the structured processing is completed, the keyword items that meet the word frequency condition and do not belong to stop words are retained, are marked as hotwords, and form the structured hotword set Kws; The de-duplication processing removes the repetition in the original hotword set Raw, and generates a unique number for each keyword; The word frequency statistics and screening statistics the occurrence frequency of each keyword in the original hotword set Raw, and compare the occurrence frequency with a preset minimum word frequency threshold, retain the keywords that meet the condition that the occurrence frequency is greater than or equal to the minimum word frequency threshold, and generate a frequency screening set RawFil after frequency filtering; The invalid term filtering filters out the syntax conjunctions, auxiliary words and empty adverb related non-semantic load key terms in the frequency screening set RawFil by calling the preset stop word.

4. The intelligent management method of an Internet-based library according to claim 3, characterized in that: The S2 includes S21 and S22. The S21 encodes and processes each hot word in the structured hot word set Kws using a semantic embedding model, calls a semantic embedding function Enc, traverses each hot word in the structured hot word set Kws, inputs the hot word into the semantic embedding model for processing as input data, and outputs the semantic vector Svec(j) of the jth hot word. After integrating the semantic vectors of all hot words, the word vector set Emb={Svec(1), Svec(2), …, Svec(j)|j∈N} is obtained, and N represents the total number of hot words. The semantic embedding model is used to convert the jth hot word Kws(j) into a semantic vector representation, and the semantic embedding model includes encoding and processing using Word2Vec, BERT, GloVe and BERT derivative model based on Chinese field fine-tuning, and the specific form of the semantic embedding function Enc is calling the semantic embedding function Enc(Kws(j)), and Kws(j) represents the jth hot word in the structured hot word set Kws. The S22 performs vector aggregation processing on the obtained word vector set Emb to generate a semantic trend vector representing the entire hot word set, which is marked as a trend semantic vector Trd. The trend semantic vector Trd is obtained by using The weighted average aggregation policy formula is obtained, in which w(j) represents the heat weight of the jth hot word.

5. The intelligent management method of an Internet-based library according to claim 4, characterized in that: The S3 includes S31. The S31 extracts each book in the library collection database, sequentially extracts the field data of each book, the field data includes book title, book label, book classification and book abstract, and sequentially splices the field data of each book to form the complete text field of the ith book, which is marked as the book text field btxt(i) of the ith book. After integrating all books, the book text field set BTxt={btxt(1), btxt(2), …, btxt(i)|i∈M} is obtained, and M represents the total number of books. The semantic embedding model in step S2 is used to encode and process the constructed book text field set BTxt, and the semantic embedding function Enc is called to generate the vector representation of each book in the semantic space, which is marked as the book semantic vector BVec(i) of the ith book. After integrating all book vectors, the book semantic vector set BVecSet is obtained. The specific form of the book semantic vector set BVecSet is BVecSet={BVec(1), BVec(2), …, BVec(i)|i∈M}.

6. The intelligent management method of an Internet-based library according to claim 5, characterized in that: The S3 includes S32. The S32 calculates the semantic similarity between the book semantic vector BVec(i) of the ith book and the trend semantic vector Trd based on the book semantic vector set BVecSet and the trend semantic vector Trd, which is marked as the semantic similarity value Sim(i) of the ith book. After performing the semantic similarity operation on each vector in the book semantic vector set BVecSet, the book trend correlation set SimSet is formed. The book trend relevance set SimSet is specifically in the form of SimSet={Sim(1), Sim(2), …, Sim(i)|i∈M}, wherein M represents the total number of books. The semantic similarity value Sim(i) of the i-th book is obtained by calculation according to the following cosine similarity formula: ; In the formula, ||·|| represents the Euclidean norm of a vector.

7. The intelligent management method of an Internet-based library according to claim 6, characterized in that: The S4 comprises S41 and S42. In S41, based on the book text field btxt(i) of each book in the book text field set BTxt, the propagation behavior of the content including the book text field btxt(i) in the multiple Internet platforms is collected and counted in real time by calling a preset Internet platform monitoring interface, the forwarding number Rtw(i), the mention frequency Mnt(i) and the active sharing number Shr(i) of the i-th book are obtained, and the propagation behavior feature vector MVec(i) of the i-th book is constructed as MVec(i)={Rtw(i), Mnt(i), Shr(i)}. The propagation behavior feature vector MVec(i) of each book is extracted by traversing the book text field set BTxt, and the social propagation behavior feature set Mset is constructed. The social propagation behavior feature set Mset is specifically in the form of Mset={MVec(1), MVec(2), …, MVec(i)|i∈M}, wherein M represents the total number of books. In S42, the social propagation behavior feature set Mset is normalized to eliminate the difference between different dimensions, and the normalized standard forwarding number NRtw(i), the normalized standard mention frequency NMnt(i) and the normalized standard active sharing number NShr(i) of the i-th book are subjected to propagation intensity evaluation processing, the social propagation intensity value Soc(i) of the i-th book in the Internet platform is obtained, and after the social propagation intensity values Soc(i) of all books are integrated, the social propagation intensity set SocSet is obtained as SocSet={Soc(1), Soc(2), …, Soc(i)|i∈N}. The social propagation intensity value Soc(i) of the i-th book is obtained by the following formula: ; In the formula, s1, s2 and s3 represent the weight coefficients of the standard forwarding number NRtw(i), the standard mention frequency NMnt(i) and the standard active sharing number NShr(i) of the i-th book respectively, and s1+s2+s3=1, and the specific values are set by the user.

8. The intelligent management method of an Internet-based library according to claim 7, characterized in that: The S5 comprises S51. S51, by extracting the lending amount of each book in the fixed period in the book lending record, the lending trend change set DynSet of each book is counted, and the obtained social communication intensity set SocSet is normalized, the difference between different dimensions is eliminated, and then fitted with the book trend correlation set SimSet, the comprehensive hotness prediction function Hot(i) of the i-th book is constructed, and after integrating the comprehensive hotness prediction functions Hot(i) of all books, the hotness set HotSet={Hot(1), Hot(2), …, Hot(i)|i∈N} is obtained; The comprehensive hotness prediction function Hot(i) of the i-th book is obtained by the following calculation formula: ; In the formula, log represents the logarithmic function.

9. The Internet-based intelligent management method for libraries according to claim 8, characterized in that: The S5 includes S52; S52, based on the hotness set HotSet, the traversal comparison operation is performed, the comprehensive hotness prediction function Hot(i) of the i-th book is compared with the preset recommendation threshold Thr, and all recommended books are selected according to the comparison result, and the recommended book set Rec is obtained; The recommended book is obtained by the following comparison and selection method: When the comprehensive hotness prediction function Hot(i) of the i-th book is greater than or equal to the recommendation threshold Thr, it is determined that the i-th book has potential hotness to enter the homepage recommendation area, the i-th book is marked as a recommended book, and is counted into the recommended book set Rec; When the comprehensive hotness prediction function Hot(i) of the i-th book is less than the recommendation threshold Thr, it is determined that the i-th book does not have potential hotness to enter the homepage recommendation area, the i-th book is marked as a non-recommended book, and is not counted into the recommended book set Rec.

10. An Internet-based intelligent library management platform applied to the Internet-based intelligent library management method of any one of claims 1-9. The internet hotword collection module, the hotword semantic conversion module, the library matching module, the internet transmission monitoring module and the book recommendation module are included. The internet hotword collection module, the hotword semantic conversion module, the library matching module, the internet transmission monitoring module and the book recommendation module are included. The hotword semantic conversion module encodes each hotword in the structured hotword set Kws using a semantic embedding model to obtain a word vector Emb. After traversing all hotwords in the structured hotword set Kws, a trend semantic vector Trd is constructed. The library matching module extracts the book text field set BTxt of the books in the library collection database, performs semantic encoding, and constructs the book semantic vector set BVecSet of the books. The cosine similarity between the book semantic vector set BVecSet and the trend semantic vector Trd is calculated to obtain a semantic similarity value Sim. The internet transmission monitoring module monitors the external transmission behavior of each book on the internet platform, collects the transmission data of the text field set of the book on the internet platform, and calculates the social communication intensity value Soc. The book recommendation module constructs a comprehensive hotness prediction function Hot of the books based on the semantic similarity value Sim, the social transmission strength value Soc and the borrowing trend change value Dyn. After integrating the comprehensive hotness prediction functions Hot of all the books, a hotness set HotSet is obtained, and a preset recommendation threshold Thr is compared. According to the comparison result, all the books are screened to obtain a recommended book set Rec.

Images