Search

CN-121980081-A - Book intelligent recommendation method based on big data analysis

CN121980081ACN 121980081 ACN121980081 ACN 121980081ACN-121980081-A

Abstract

The invention relates to the technical field of big data analysis and intelligent information recommendation, and discloses an intelligent book recommendation method based on big data analysis. Collecting the discrete natural day borrowing behavior of a platform user, establishing a user and book number index, generating a history sequence according to the day, setting a fixed window intercepting sub-sequence and bit encoding on the sequence, counting the occurrence proportion of each path in the user, setting a threshold to form a frequent path library, extracting the tail behavior of the target user and encoding, matching the paths with the prefix in the library, collecting the next candidate and aggregating the support degree to be a recommendation score, removing the borrowed resource, generating a recommendation set according to the sorting, and preferentially selecting a representative interpretation path from the frequent path to form a book-score-interpretation triplet, recording recommendation and date, wherein the user has newly added borrowing, namely the update sequence and the log are circularly executed.

Inventors

  • MA YEXIANG
  • WANG XIAOJUAN

Assignees

  • 无锡商业职业技术学院

Dates

Publication Date
20260505
Application Date
20260119

Claims (9)

  1. 1. The intelligent book recommendation method based on big data analysis is characterized by comprising the following steps: Acquiring borrowing behaviors of a platform user on discrete natural days, establishing a platform user number index and a platform book number index, and generating historical borrowing behavior sequences of all users according to the dates; setting a fixed window on a history borrowing behavior sequence, intercepting continuous subsequences according to different initial positions, and coding the subsequences according to time-bit times to form a behavior subsequence coding set with a bit time mark; Based on the behavior subsequence coding set, counting the occurrence proportion of each behavior coding path in the user sequence, constructing a support degree, screening out paths meeting a threshold value, and forming a frequent behavior path library; extracting tail borrowing behaviors of a target user, coding, matching frequent behavior paths which are identical to tail behavior coding path prefixes in a frequent behavior path library, and summarizing candidate recommended books given at the next position; Aggregating the frequent behavior path support degree associated with the candidate recommended books, and calculating the recommended score of each book to obtain a recommended score set; Removing historical borrowed books of the target user from the recommendation score set, sorting the rest books according to scores, and generating a final recommendation book set according to the preset number; Searching associated frequent behavior paths aiming at a final recommended book set, selecting representative interpretation paths according to the support degree and the path length and combining the book numbering sequence, constructing recommended interpretation triples comprising recommended books, recommended scores and interpretation paths, and summarizing the recommended interpretation triples into a recommended interpretation set; And recording the final recommended book set, the recommended interpretation set and the generation date into a recommended log, and monitoring the history borrowing behavior sequence and log record of the target user after newly added borrowing.
  2. 2. The intelligent book recommendation method based on big data analysis of claim 1, wherein the collection platform user borrows on discrete natural days, a platform user number index and a platform book number index are established, and historical borrowing sequences of all users are generated according to dates, and the method specifically comprises the following steps: Acquiring all user lists participating in recommendation calculation on a platform, distributing unique user numbers for each user, establishing a one-to-one correspondence between the user numbers and actual user identifications, and forming a platform user number index; Acquiring all book lists participating in recommendation calculation on a platform, distributing unique book numbers for each book, establishing a one-to-one correspondence between book numbers and actual book identifications, and forming a platform book number index; Collecting borrowing behaviors occurring on corresponding natural days for each user on each discrete natural day, and correspondingly recording book numbers of borrowed books on the corresponding natural days; and for each user, according to the sequence of natural days from the beginning date of recording, carrying out time sequence arrangement on the borrowing book numbers of the corresponding user in each discrete natural day, concatenating the arranged borrowing records to form a history borrowing behavior sequence of the corresponding user, and simultaneously recording the length of the history borrowing behavior sequence as the number of the discrete natural days of the corresponding user participating in recording.
  3. 3. The intelligent book recommendation method based on big data analysis according to claim 2, wherein a fixed window is set on the history borrowing behavior sequence, continuous subsequences are intercepted according to different initial positions, the subsequences are coded according to time-bit times, and a behavior subsequence coding set with a bit-time mark is formed, and the method specifically comprises the following steps: Setting the maximum length of a behavior window, setting the maximum length of the behavior window to be not less than the natural number of days of two days, and taking the maximum length of the behavior window as the upper limit length of a continuous subsequence intercepted from the history borrowing behavior sequence; for the historical borrowing behavior sequence of each user, sequentially selecting different starting dates from the earliest recording position, and intercepting continuous borrowing subsequences according to the window length from the shortest continuous days to the preset maximum window length on each starting date, so that each intercepted subsequence is composed of borrowing behaviors adjacent in time; marking the position numbers of each book in the subsequence in sequence from the first borrowing action of the subsequence according to the time sequence, and forming ordered pairs of the book numbers and the position numbers in the subsequence to form a book sequence code with rank information; Summarizing book sequence codes generated by all users, all starting dates and all window lengths, and constructing a behavior subsequence code set consisting of a plurality of behavior subsequences with bit sub-marks.
  4. 4. The intelligent book recommendation method based on big data analysis of claim 3, wherein the behavior subsequence coding set is based on statistics of occurrence proportion of each behavior coding path in a user sequence, construction of support degree and screening out paths meeting a threshold value to form a frequent behavior path library, and the method specifically comprises the following steps: selecting any behavior coding path, and regarding the selected behavior coding path as an ordered pair sequence formed by arranging a plurality of book numbers and position numbers in a window according to a position sequence; generating a rank coding result on the complete history borrowing behavior sequence of each user according to the same rule as the subsequence, searching whether the behavior coding path appears in a continuous segment form in the rank coding result generated in the previous step, and adding one to the statistics count of the corresponding user when the rank coding result contains the behavior coding path; Counting the number of the users of each behavior coding path in the platform user set, performing proportional calculation on the counted number of the users and the total number of the users participating in recommendation calculation to obtain the occurrence proportion of the behavior coding paths in the platform user set, and taking the occurrence proportion as a support index of the corresponding behavior coding paths; Setting the proportion of the single user in all users as a minimum support threshold, converging the behavior coding paths with the support degree not lower than the minimum support threshold into a frequent behavior path set, and forming a frequent behavior path library.
  5. 5. The intelligent book recommendation method based on big data analysis according to claim 4, wherein the extracting tail borrowing behaviors of the target user and coding are performed, frequent behavior paths which are identical to tail behavior coding path prefixes are matched in a frequent behavior path library, and candidate recommended books given in the next position are summarized, and the method specifically comprises the following steps: Selecting a target user needing to execute recommendation calculation, starting from the tail of a historical borrowing behavior sequence of the target user, setting the maximum length of tail behaviors according to a result obtained by subtracting one day from the maximum length of a preset behavior window, and acquiring the latest continuous borrowing behavior as a tail behavior subsequence; Performing bit encoding on the tail behavior subsequence according to a time sequence, sequentially marking position numbers from the first borrowing behavior of the tail behavior subsequence, and constructing a tail behavior encoding path formed by ordered pairs of book numbers and position numbers; searching all frequent behavior paths with the initial section completely consistent with the tail behavior coding path in a frequent behavior path library to form a frequent behavior path set taking the tail behavior coding path as a prefix; when frequent behavior paths matched with the tail behavior coding paths exist, reading book numbers corresponding to the next positions, which are immediately behind the tail behavior length, of each matching path, summarizing the next book numbers given by all the matching paths, and forming a candidate recommended book set of the target user at the current moment.
  6. 6. The intelligent book recommendation method based on big data analysis according to claim 5, wherein the converging candidate recommended book associated frequent behavior path support degree, each book recommendation score is calculated to obtain a recommendation score set, and the method specifically comprises: Searching a frequent behavior path set taking a tail behavior coding path of a target user as a prefix for each book in the candidate recommended book set, and obtaining all frequent behavior path subsets of the corresponding book predicted at corresponding positions; Reading the support degree value of the corresponding frequent behavior path in the platform user set for each frequent behavior path in the frequent behavior path subset belonging to the same candidate recommended book, and regarding the read support degree value as a partial score of the contribution of the corresponding frequent behavior path to the candidate recommended book; sequentially accumulating all partial scores of the same candidate recommended book to obtain a total recommended score of the corresponding candidate recommended book in the current target user scene, and forming a recommended score result between the target user and the corresponding book; when no frequent behavior paths matched with the tail behavior coding paths of the target users exist, the recommendation scores of all books are uniformly recorded as zero values.
  7. 7. The intelligent book recommending method based on big data analysis according to claim 6, wherein the step of removing the historical borrowed books of the target user from the recommendation score set, sorting the remaining books according to scores, and generating a final recommendation book set according to a preset number, comprises the following steps: Traversing borrowing records of the target user on each discrete natural day based on the historical borrowing behavior sequence of the target user, and summarizing to obtain all book number sets which the target user has borrowed in the past to form a historical book set of the target user; Deleting book numbers in the historical book set of the target user in the platform book number index to obtain a legal candidate book set which is not borrowed by the target user, and regarding books in the legal candidate book set as books meeting recommendation conditions on borrowing records; Selecting books with recommendation scores greater than zero from legal candidate book sets to form a candidate book set with positive recommendation scores, and when the candidate book set with the positive recommendation scores is empty, marking the current round of recommendation output set as an empty set and not outputting specific recommendation books; When the recommendation scores of the two books are the same and are both in the sequencing set, the books are arranged according to the sequence of the book numbers in the platform book number index, and the books with smaller numbers are arranged in front; And according to the preset maximum recommended number and the number of the candidate books with positive recommended scores, taking the smaller value of the maximum recommended number and the number of the candidate books with positive recommended scores as the actual recommended number of the current round, and sequentially selecting a plurality of previous books from the sorted candidate book sets to form a final recommended book set of the target user.
  8. 8. The intelligent book recommendation method based on big data analysis according to claim 7, wherein the searching of the associated frequent behavior paths for the final recommended book set, selecting representative interpretation paths according to the support and path length and combining the book number sequence, constructing recommended interpretation triples including recommended books, recommended scores and interpretation paths, and summarizing the recommended interpretation triples into recommended interpretation sets, specifically comprises: When the final recommended book set is empty, recording a recommended interpretation set aiming at a target user as an empty set, only retaining information of recommendation failure or no recommendation result, and not constructing a path interpretation structure; When the final recommended book set is not empty, searching a frequent behavior path taking a tail behavior coding path of a target user as a prefix in a frequent behavior path library, and selecting a frequent behavior path predicted to correspond to the recommended book at a tail subsequent position from the frequent behavior paths to form a frequent behavior path set related to the corresponding recommended book; Calculating the support degree of each frequent behavior path for a frequent behavior path set related to the recommended book, finding out the maximum value of the support degree, screening out a frequent behavior path subset with the maximum support degree, and regarding the frequent behavior path subset with the maximum support degree as a candidate frequent behavior path set; In the candidate frequent behavioral path set, the path length of each path is counted, the minimum path length value is obtained, a path subset with the path length equal to the minimum path length value is screened out, and the path subset with the path length equal to the minimum path length value is taken as the candidate frequent behavioral path set with the highest support and the shortest path; When a plurality of paths are still contained in the candidate frequent behavioral path set with highest support and shortest paths, books in the paths are mapped into corresponding platform book number sequences in sequence, dictionary ordering is carried out on the sequences according to the comparison sequence from small numbers to large numbers, and the forefront path in the ordering result is selected as the unique interpretation path of the corresponding recommended book; And constructing triples consisting of the recommended books, the recommended scores of the corresponding recommended books and the unique interpretation paths for the recommended books with the unique interpretation paths, and summarizing the triples constructed for all the recommended books to form a recommended interpretation set corresponding to the target user.
  9. 9. The intelligent book recommendation method based on big data analysis of claim 8, wherein the recording of the final recommended book set, the recommended interpretation set and the generation date into the recommended log, the monitoring of the history borrowing behavior sequence and log record of the target user after the newly added borrowing, specifically comprises the following steps: When the current round of recommendation results aiming at the target user are generated, recording a discrete natural day index corresponding to a recommendation generation date, combining the target user identification, a final recommendation book set and the recommendation generation date into a recommendation log record, and additionally writing the recommendation log into a recommendation log library; When the target user is detected to borrow books on a new discrete natural day, the book number of the new borrowed book is added to the tail part of the original historical borrowing behavior sequence of the target user, an updated user behavior sequence is formed, and the behavior sequence length is increased by one; When the new borrowing behavior is detected and the behavior sequence is updated, the updated user behavior sequence is used as input data, and the user behavior data acquisition, the behavior subsequence extraction, the frequent behavior path screening, the candidate book prediction, the recommendation score calculation, the final recommendation book set construction and the recommendation interpretation output are re-executed to form a new recommendation and updating cycle.

Description

Book intelligent recommendation method based on big data analysis Technical Field The invention relates to the technical field of big data analysis and intelligent information recommendation, in particular to an intelligent book recommendation method based on big data analysis. Background With the development of the informatization age, online reading platforms and digital libraries are rapidly popularized, and the reading and borrowing behavior data of users are increasingly abundant. Currently, book recommendation systems have become an important technical means for improving user experience and promoting book circulation and resource utilization. Most of the existing book recommendation technologies are based on mainstream models such as collaborative filtering, content analysis or deep learning, and recommend related books to users by analyzing information such as basic attributes, interest labels, historical scores and the like of the users. First, conventional collaborative filtering recommendation techniques typically rely on scoring matrices or click relationships between users and books, lacking in deep mining of the temporal and pattern structure of the sequence of user borrowing, reading behaviors. The borrowing or clicking actions of the user actually exhibit significant sequence and phasing, for example, the user may focus on borrowing a certain class of subject books during a certain period of time, or form individual reading habits during a particular time window. The existing method often regards behaviors as unordered sets, ignores time sequence association and potential rules among the behaviors, and leads to limited correlation and individuation degree of recommendation results. And secondly, content analysis analoging technology mainly focuses on static properties such as texts, topics, keywords and the like of books, and has low coupling degree with dynamic behavior sequences of users. The method based on the book content characteristics is difficult to fully reflect the dynamic changes of interest transfer, theme transfer and the like of the user in the actual use scene, and the recommending effect of a new book or a long-tail book is also difficult to guarantee. In addition, although the deep learning and neural network model can extract complex user characteristics and book relations, the deep learning and neural network model generally depends on a large number of training samples and high calculation force, the model decision process is difficult to explain, the application threshold is high, and the system flexibility and controllability are insufficient. Again, the prior art lacks an automatic mining and utilization mechanism for frequent behavior patterns in large-scale user behavior sequences. In particular, the borrowing and clicking actions of the user often involve some high frequency book combinations or borrowing paths, and these implicit modes have important guiding value for subsequent recommendations. The traditional method fails to design effective special algorithms such as behavior window segmentation, sequence segmentation, frequent pattern statistics and matching, and the like, so that a recommendation system in a big data environment is difficult to efficiently and accurately discover potential demands of users. Meanwhile, the current system generally cannot dynamically adjust the recommendation rules according to the differences, trends and individuation characteristics of the behavior patterns, so that fine and dynamic recommendation cannot be made for future behaviors of the user. The scheme aims to provide an intelligent book recommendation method based on big data analysis, which utilizes a user borrowing action sequence in discrete natural day dimensions, intercepts and orders the original time sequence through a fixed window to convert the original time sequence into a subsequence coding set with sequence marks, then carries out frequent path mining on all action coding paths based on a support index to construct a frequent action path library, quickly matches paths with the same prefix in the library aiming at the latest tail borrowing action of a target user and gathers book numbers at the next position to generate candidate recommendation, accumulates the support of the candidate book association paths to obtain personalized recommendation scores, finally outputs a final recommendation set after removing historical borrowed books and sorting the scores, generates interpretable path triples to support a result traceable source, and circularly updates recommendation results and new borrowing actions of the user. Disclosure of Invention The invention provides an intelligent book recommendation method based on big data analysis, which facilitates solving the problems mentioned in the background art. The invention provides a book intelligent recommending method based on big data analysis, which comprises the following steps: Acquiring borrowing behaviors of a