Search

CN-122022954-A - Peripheral product recommendation method based on historical booking information

CN122022954ACN 122022954 ACN122022954 ACN 122022954ACN-122022954-A

Abstract

The invention relates to the technical field of big data and hotel reservation, in particular to a peripheral product recommendation method based on historical booking information, which comprises the following steps of S1, acquiring historical booking record data and real-time state data of a user; the method comprises the steps of S2, carrying out multidimensional feature extraction based on historical study record data of a user to generate a user historical feature vector, S3, carrying out preference vectorization modeling on each scenic spot type based on the user historical feature vector to generate a user preference vector, S4, carrying out environment vectorization modeling based on real-time state data of the user to generate a real-time environment feature vector, S5, carrying out scenic spot and surrounding product recommendation based on the user preference vector and the real-time environment feature vector to generate corresponding recommended scenic spots and recommended surrounding products thereof, and recommending the corresponding recommended scenic spots and the recommended surrounding products to the user. The invention ensures that the recommendation result meets the historical preference of the user and gives consideration to the emotion tendency of the user, and simultaneously dynamically responds to the current environment of the user, thereby avoiding the problem of disjointing the recommendation result and the actual scene of the user.

Inventors

  • MA YUHAN
  • MA ZHAODE
  • LUO JIAYOU
  • XIAO KUNPENG

Assignees

  • 重庆惠迎客信息科技有限公司

Dates

Publication Date
20260512
Application Date
20260204

Claims (10)

  1. 1. A peripheral product recommendation method based on historical booking information is characterized by comprising the following steps: S1, acquiring historical booking record data and real-time state data of a user; s2, carrying out multidimensional feature extraction based on historical booking record data of a user to generate a user historical feature vector comprising a time feature vector, a place feature vector and a behavior feature vector; s3, carrying out preference vectorization modeling on each scenic spot type based on the user history feature vector to generate a user preference vector containing the preference of each scenic spot type; S4, carrying out environment vectorization modeling based on real-time state data of a user, and generating a real-time environment feature vector comprising a position fitness value, a weather matching value, a hot spot heat value and a holiday sensitivity value; s5, recommending scenic spots and surrounding products based on the user preference vector and the real-time environment feature vector, generating corresponding recommended scenic spots and recommended surrounding products thereof, and recommending the scenic spots and the recommended surrounding products to the user.
  2. 2. The method for recommending a surrounding product based on history room information as set forth in claim 1, wherein in step S2, the user history feature vector is generated by: S201, preprocessing historical booking record data of a user; S202, extracting a corresponding time feature vector for each preprocessed historical booking record, wherein the time feature vector comprises a season identification, a holiday identification and a time attenuation factor; the calculation formula of the time attenuation factor is as follows: ; wherein: Represent the first A time attenuation factor of the historical booking record; Representing the attenuation coefficient; representing the time difference between the current time and the history reservation record; s203, extracting a corresponding place feature vector according to each preprocessed history booking record, wherein the place feature vector comprises scenic spot type labels; S204, extracting corresponding behavior feature vectors for each preprocessed history booking record; the calculation formula of the behavior feature vector is expressed as follows: ; wherein: Represent the first Behavior feature vectors recorded in a historical booking record; Representing the set behavior coefficients; Represent the first The entry duration of the historical reservation record; Represent the first Reservation prices recorded in the strip history reservation room; representing the average price of all the historical booking records of the user; Represent the first Emotional scores of the historical booking records; and S205, splicing the time feature vector, the place feature vector and the behavior feature vector of each history reservation record to obtain the user history feature vector of each history reservation record.
  3. 3. The method for recommending a surrounding product based on history room information as set forth in claim 1, wherein in step S3, the user preference vector is generated by: S301, acquiring preset The category of each scenic spot; S302, calculating a basic preference vector of a user for each scenic spot category based on time feature vectors, place feature vectors and behavior feature vectors of all the historical reservation records; The calculation formula of the base preference vector is expressed as: ; wherein: Representing user category for scenic spots Is a basic preference vector of (1); representing a total number of historic reservation records; Represent the first Behavior feature vectors recorded in a historical booking record; Represent the first Category match indicator of strip historic booking record, item The scenic spot type label of the historical booking record is Then =1, Otherwise =0; Represent the first A time attenuation factor of the historical booking record; S303, obtaining comment texts corresponding to all historical booking records of the user, and carrying out emotion enhancement on basic preference vectors of the user for each scenic spot category through the comment texts to generate user preference vectors.
  4. 4. The method for recommending a surrounding product based on the history room information as set forth in claim 3, wherein in step S303, the method comprises the following steps: S3031, extracting keywords of scenic spot categories from comment texts recorded in all historical bookings of the user, and calculating comment keyword weights of the user for all scenic spot categories through a TF-IDF algorithm; The formula is: ; wherein: Representing scenic spot categories Is a comment keyword weight of (2); Representing scenic spot categories Word frequency of the corresponding keywords in all comment texts; Representing scenic spot categories The inverse document frequency of the corresponding keywords in all comment texts; S3032, carrying out emotion enhancement on basic preference vectors of the user for all the scenic spot categories based on comment keyword weights of the user for all the scenic spot categories to obtain emotion enhancement preference vectors; The formula is: ; wherein: Representing user category for scenic spots Is a emotion enhancement preference vector of (a); Representing user category for scenic spots Is a basic preference vector of (1); representing emotion enhancement coefficients; s3033, carrying out normalization processing on each emotion enhancement preference vector to obtain a normalized preference vector; The formula is: ; wherein: Representing user category for scenic spots Is included in the normalized preference vector of (a); 、 Representing user category for scenic spots And Is a emotion enhancement preference vector of (a); Representing the total number of scenic spot categories; S3034, taking the normalized preference vector of the user for all the scenic spot categories as the user preference vector 。
  5. 5. The method for recommending surrounding products based on historic booking information according to claim 4, wherein in step S4, a real-time environment feature vector is generated by the following steps: S401, acquiring real-time state data of a user, wherein the acquired real-time state data of the user comprises a real-time position of the user, and weather types, hot event data and holiday data corresponding to the real-time position of the user; S402, calculating corresponding position fitness based on the real-time position of the user and the booking positions of all the historical booking records of the user ; S403, calculating a corresponding weather matching degree value based on the weather type corresponding to the real-time position of the user ; S404, calculating corresponding hot spot heat value based on hot spot event data corresponding to the real-time position of the user ; The formula is: ; wherein: representing a hot spot heat value; Represent the first Initial values of the hotness of the individual events; Represent the first Weights of the individual events; Representing the number of events; s405, calculating corresponding holiday sensitivity values based on holiday data corresponding to the real-time position of the user ; S406, adapting the position of the user Weather match value Hot spot heat value And holiday sensitivity value Splicing to generate real-time environment feature vector of user 。
  6. 6. The method for recommending a surrounding product based on the history room information as set forth in claim 5, wherein in step S402, the position fitness is calculated by: s4021 calculating center longitude based on the booking position of all the historical booking records of the user And center latitude ; The formula is: ; ; wherein: 、 Represent the first Longitude and latitude corresponding to the strip history booking record; representing a total number of historic reservation records; S4022 calculating center longitude And center latitude Longitude difference from user's real-time location And difference in altitude ; The formula is: ; ; wherein: 、 longitude and latitude representing the user's real-time location; s4023 based on longitude difference And difference in altitude Calculating a corresponding deviation distance; The formula is: ; wherein: Representing the deviation distance; representing the earth radius; S4024 normalizing the offset distance to position fitness ; The formula is: 。
  7. 7. The method for recommending surrounding products based on history room information as set forth in claim 1, wherein in step S5, recommended sights and recommended surrounding products thereof are generated by: S501, acquiring a preset scenic spot database, wherein the scenic spot database comprises the positions of all scenic spots, scenic spot category labels, outdoor labels, real-time weather data and popularity; s502, calculating the estimated distance between the real-time position of the user and each scenery spot in the scenery spot database, and taking the scenery spot with the estimated distance smaller than the threshold value as a candidate scenery spot; The formula is: ; ; ; wherein: representing user real-time location and first Estimated distances of individual attractions; 、 longitude and latitude representing the user's real-time location; 、 Represent the first Longitude and latitude of the individual scenic spots; representing the earth radius; S503, calculating category matching scores, distance scores, weather adaptation scores and popularity scores of all candidate scenic spots; s504, calculating weight values of category matching scores, distance scores, weather adaptation scores and popularity scores based on user preference vectors and real-time environment feature vectors; S505, carrying out weighted calculation on the category matching score, the distance score, the weather adaptation score and the popularity score of each candidate scenic spot based on the weight value to obtain a corresponding basic comprehensive score; S506, selecting a plurality of candidate scenery spots with highest basic comprehensive scores as recommended scenery spots; S507, taking the related peripheral products preset by the recommended scenic spots as the recommended peripheral products.
  8. 8. The method for recommending a surrounding product based on the history room information as set forth in claim 7, wherein in step S503, the method specifically comprises the following steps: s5031 from user preference vector Acquiring normalized preference vectors corresponding to the category labels of the candidate scenic spots as category matching scores of the candidate scenic spots; S5032 based on the estimated distance between the user' S real-time location and the candidate sight location Calculating the distance score of the candidate scenic spots; The formula is: ; wherein: Representing candidate sights Distance score of (2); Representing a user's real-time location and candidate attractions Is a function of the estimated distance of (2); S5033, acquiring a preset weather adaptation score based on the outdoor label of the candidate scenic spot and real-time weather data; s5034, taking the preset popularity of the candidate scenic spots as the popularity score.
  9. 9. The method for recommending a surrounding product based on the historic booking information according to claim 8, wherein in step S504, the method specifically comprises the following steps: S5041 initializing basic weights 、 、 ; S5042 by user preference vector And base weight Calculating an adjustment weight ; The formula is: ; s5043 through real-time environmental feature vector And base weight Calculating an adjustment weight ; The formula is: ; S5044 based on the adjustment weight Adjusting weights And a base weight 、 、 Calculating weight value 、 、 ; The formula is: ; ; 。
  10. 10. The method for recommending surrounding products based on the historic booking information according to claim 9, wherein in step S505, the basic integrated score of the candidate scenic spot is calculated by the following formula: ; wherein: Representing candidate sights Is a basic composite score of (1); 、 、 、 Representing candidate sights Category matching score, distance score, weather fit score, and popularity score.

Description

Peripheral product recommendation method based on historical booking information Technical Field The invention relates to the technical field of big data and hotel reservation, in particular to a peripheral product recommendation method based on historical booking information. Background With the popularization of online travel service platforms, hotel room booking recommendation systems have become the core for improving user experience and platform competitiveness. The current mainstream recommendation system mostly adopts collaborative filtering, content filtering or mixing algorithm, mainly focuses on personalized recommendation of hotels based on data such as user history booking records, scoring behaviors, search keywords and the like. Currently, the prior art begins to attempt to explore recommended scenes that extend to scenic spots and surrounding products using historical booking information. The concept has a certain theoretical basis that historical booking data implies travel behavior tracks and preference clues of users, for example, users who frequently check in hotels in a seashore area may prefer seashore activities, and booking records in specific seasons can reflect seasonal travel preferences. And part of practice generates preliminary recommendation through regional association rules or simple frequency statistics, and the preliminary recommendation shows limited value in the aspect of improving the attention degree of users to associated scenic spots. However, at present, the attempts stay in the data surface layer association, so that systematic and high-precision recommended logic cannot be formed, and the practical application effect is limited. The applicant analyzes the scenic spot and the recommending method of the peripheral products thereof in the prior art, and finds that the following problems mainly exist: 1) The user preference modeling dimension is single and static. The conventional method generally only carries out coarse granularity statistics (such as regional access frequency) on historical booking data, and cannot construct dynamic user portraits from multiple angles such as time evolution (such as seasonal preference change and recent behavior weight), place semantics (such as fine granularity interest mapping of scenic spot types), behavior depth (such as combination of comment emotion and consumption habit), and the like, so that distortion of a user preference model is caused, and real and changed interest trends of users are difficult to accurately capture. 2) The recommendation process lacks the ability to perceive and adapt to real-time environmental factors. The existing method generally ignores key dynamic variables such as the current geographic position, real-time weather conditions, social hot events, holiday atmosphere and the like of the user, and the recommendation result is easy to be disjointed with the current actual scene of the user. For example, recommending open-air scenic spots in rainy days, or still pushing remote options when the user is already in a scenic spot, reduces the recommendation practicality and user confidence. In summary, the prior art has shortcomings in deep mining of user preference and dynamic adaptation of environment, which results in low recommendation accuracy of scenic spots and peripheral products and weak scene fitting, and is difficult to meet urgent demands of users on intelligent and integrated travel services. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide a peripheral product recommending method based on historical booking information, which is characterized in that through multi-dimensional feature extraction and user preference modeling, a recommending result is enabled to be in line with the user history preference and also give consideration to the user emotion tendency, so that the accuracy and the user satisfaction of scenic spot and peripheral product recommendation are improved, meanwhile, through the construction of a real-time environment feature vector, the current environment of a user is dynamically responded, the problem that the recommending result is disjointed with the actual scene of the user is avoided, thereby providing highly-adaptive recommending content, improving the practicability and the user experience of scenic spot and peripheral product recommendation, and enabling the travel recommendation to be more fit with the instant requirement of the user. In order to solve the technical problems, the invention adopts the following technical scheme: a peripheral product recommendation method based on historical booking information comprises the following steps: S1, acquiring historical booking record data and real-time state data of a user; s2, carrying out multidimensional feature extraction based on historical booking record data of a user to generate a user historical feature vector comprising a time feature vector, a place feature vector and a