Search

CN-119963282-B - Ice and snow travel product recommendation method and system based on neural collaborative filtering and linear confidence upper bound

CN119963282BCN 119963282 BCN119963282 BCN 119963282BCN-119963282-B

Abstract

The invention discloses an ice and snow travel product recommending method and system based on neural collaborative filtering and linear confidence upper bound, relates to the technical field of electronic commerce product recommending, and aims to solve the problem of exposure deviation in the existing ice and snow travel product recommending method. The method fully considers the problem of exposure deviation in ice and snow travel product recommendation, and excessively recommending hot products can lead to recommendation to present long tail phenomenon, so that the flow is concentrated on the high-exposure products, and the selection range of users is limited. Meanwhile, the low exposure product is difficult to enter a recommendation list of a user due to the influence of exposure deviation, so that unfair recommendation is formed. The method and the device combine the multi-feature information of the user, capture the real preference of the user, and reduce the influence of exposure deviation on the recommendation result. The neural synergistic network is designed to extract multi-attribute information of users, and a linear confidence upper bound algorithm is adopted to generate rewarding value characteristics, so that the exploration capacity of low-exposure products is improved, and meanwhile, the recommendation precision and individuation are maintained. According to the recommendation method, the user multi-feature information, the product attribute data and the rewarding value features are fused into the recommendation model, so that the exposure deviation problem in the traditional recommendation model is optimized, and the fairness and diversity of recommendation are improved.

Inventors

  • LI PENG
  • LI XIAOSHAN
  • YAO LU

Assignees

  • 哈尔滨商业大学

Dates

Publication Date
20260512
Application Date
20241231

Claims (10)

  1. 1. The ice and snow travel product recommending method based on neural collaborative filtering and linear confidence upper bound is characterized by comprising the following steps of: The method comprises the steps of collecting user characteristic information and user historical scoring data, wherein the user characteristic information covers gender, age and occupation, and the user historical scoring data comprises scoring on tourist attractions, browsing, clicking and purchasing records of related E-commerce products; step two, collecting characteristic information of ice and snow travel products; Constructing feature vectors of users and ice and snow travel products, and encoding each feature into a low-dimensional dense vector through an embedded layer by utilizing the user attribute information and the ice and snow travel product attribute information collected in the first step and the second step to form the user feature vectors and the ice and snow travel product feature vectors; step four, calculating similarity by combining the behavior characteristics and the content characteristics of the user so as to acquire user preference; Fifthly, introducing a linear confidence upper bound (LinUCB) algorithm to calculate rewarding features, and mining products which are possibly interested by a user but have lower exposure rate through dynamic balance exploration and utilization; Embedding the reward value data into a neural collaborative filtering network model, generating reward value embedded vectors of users and products, and combining the reward value embedded vectors with respective nearest neighbor feature vectors to form matrix decomposition and user and product representation required by a multi-layer perceptron; Step seven, the user and the ice and snow travel product representation obtained in the step six are used for learning the interactive function between the user and the ice and snow travel product to obtain the output vector of the multilayer sensor; Step eight, splicing output vectors of the matrix decomposition and the multi-layer perceptron, and inputting the output vectors into a full-connection layer to obtain the predictive score between a user and ice and snow travel products and rewarding values: step nine, the output layer completes a prediction task, optimizes and outputs the preference probability of a user on ice and snow travel products according to a binary cross entropy loss function, predicts and converts the preference probability into two kinds of problems, and the prediction probability range is set to be 0, 1; step ten, sorting the final preference probability obtained in step nine from high to low, and selecting the top ranking Recommending ice and snow travel products in the bit, wherein Representing the recommended number of ice and snow travel products.
  2. 2. The ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound of claim 1, wherein in the first step, users are initially classified, users with the same gender, age or occupation are classified into the same group, and the gender, age and occupation characteristics of the users are mapped to the [ 0-1 ] interval.
  3. 3. The ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound according to claim 1 or 2, wherein in step two, The ID of the ice and snow travel product is used as a unique identifier of each product, the system is helped to accurately track and analyze the exposure and user interaction conditions of the ice and snow travel product, the type information is used as an important factor for selecting the ice and snow travel product by a user, the recommended correlation and the user satisfaction can be improved, and the characteristic information of the ice and snow travel product comprises the ID and the type of the ice and snow travel product, the ticket of the ice and snow travel product, the winter theme event, the traffic convenience, the navigation service, the scenic spot guiding identification, the special catering and the barrier-free facilities.
  4. 4. The ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound of claim 3, wherein in step three, For the characteristics of ice and snow travel products, product ID, type, activity type, traffic convenience and navigation service attribute are integrated into a single high-dimensional vector representation; After the feature vectors of the users and the ice and snow travel products are obtained in the third step, the feature vectors are reduced in dimension through a full connection layer in consideration of the requirements of optimizing the model performance and maintaining reasonable calculation efficiency, and the final feature representation of the products is obtained through ReLU activation function processing.
  5. 5. The ice and snow travel product recommending method based on neural collaborative filtering and linear confidence upper bound according to claim 4, wherein in step five, linUCB algorithm calculates potential reward values of each ice and snow travel product through historical user behavior and product attributes, in each recommending, the system preferentially selects products to recommend based on confidence interval upper bound of reward values, thereby improving recommending frequency of low exposure ice and snow travel products, The expected rewards obtained by each user corresponding to each ice and snow travel product are as follows: representing a set of ice and snow travel products to be selected; Represent the first Ice and snow travel product for secondary experiment Corresponding feature vector, preference vector Is unknown, but the estimated value is calculated through ridge regression in each round of interaction, and each ice and snow travel product maintains one ; The system finally selects the ice and snow travel product with the highest confidence interval upper bound for recommendation: Wherein the method comprises the steps of , Is composed of history information The matrix is formed by a matrix of, Is the dimension of the context feature vector, Representing the input training data, the training data, Representing the estimated parameters of the device, Super parameters for trade-off exploration-utilization; the history information comprises a feature set composed of an interaction record of a user and an article, a context feature and a parameter estimated value in a model training process; further, the optimal prize value is calculated using a ridge regression method, with the optimization objective as follows: Wherein, the Is that The result of the trial is defined as the prize value The final system will record the item that obtained the greatest prize value, And Representing the acceptance and rejection of the recommendation by the user respectively, Is that A feature vector matrix observed by the secondary experiment, wherein each row corresponds to one feature vector; I.e. Regularization in which The diagonal matrix can effectively avoid the phenomenon of overfitting; For a pair of And (3) conducting derivation to obtain: If at first Individual ice and snow travel products have been recommended The probability of obtaining rewards for the product can be calculated as follows: For any arbitrary All present: Wherein, the Is the true probability of the product and, Is a super parameter, so the selection mechanism of LinUCB algorithm satisfies the following conditions: Wherein, the 。
  6. 6. The ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound of claim 5, wherein in step seven, the output vector is: The seventh step is to obtain matrix decomposition and representation of users and ice and snow travel products corresponding to the multi-layer perceptron, wherein the input layer at the bottom layer consists of seven characteristics, namely user ID, user gender, user age, user occupation, ice and snow travel product ID, ice and snow travel product type and rewarding value characteristics, and to learn the interaction function between the users and the ice and snow travel products to obtain the output vector of the multi-layer perceptron : Wherein, the In order to activate the function, Is the first A matrix of layer weights is provided, Is that The output of the layer is provided with, Is a bias vector.
  7. 7. The ice and snow travel product recommending method based on neural collaborative filtering and linear confidence upper bound according to claim 6, wherein in step eight, output vectors of two parts of matrix decomposition and multi-layer perceptron are spliced together and input into a full connection layer to obtain a user Ice and snow travel product Prize value Prediction score between: Wherein, the For the weight vector of the output layer, A bias term representing an output layer; Is a Sigmoid function; is the output vector of the matrix decomposition.
  8. 8. The ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound of claim 7, Step nine, the output layer completes the prediction task, outputs the preference probability of users on ice and snow travel products, adopts a binary cross entropy loss optimization function, a positive sample indicates that the users-ice and snow travel products have interaction, a negative sample indicates that the users-ice and snow travel products have no interaction, the prediction is converted into two classification problems, and the prediction probability range is set as [0,1]: Wherein, the A set of positive samples is represented and, Representing a negative set of samples.
  9. 9. A system for recommending ice and snow travel products based on the upper bound of neural collaborative filtering and linear confidence is characterized by comprising a program module corresponding to the steps of any one of claims 1-8, wherein the steps in the method for recommending ice and snow travel products based on the upper bound of neural collaborative filtering and linear confidence are executed in operation.
  10. 10. A computer readable storage medium, characterized in that it stores a computer program configured to implement the steps of a method for recommending ice and snow travel products based on neural collaborative filtering and linear confidence upper bound as claimed in any one of claims 1-8 when called by a processor.

Description

Ice and snow travel product recommendation method and system based on neural collaborative filtering and linear confidence upper bound Technical Field The invention relates to the technical field of electronic commerce product recommendation, in particular to an ice and snow travel product recommendation method based on neural collaborative filtering and linear confidence upper bound. Technical Field In the current digital age, the digital energized culture tourism fusion development mode represented by the tourism electronic commerce is an important way for realizing the modernization of Chinese tourism, and the tourism electronic commerce industry caters to unprecedented development opportunities. With the increasing demands of people for personalized travel experiences, the effect of the ice and snow travel product recommendation method becomes particularly important. Particularly, under the promotion of popularization of sports in winter and propagation of ice and snow culture, ice and snow travel has great economic potential. However, the existing ice and snow travel product recommendation method still has certain defects in the ice and snow travel field. The current recommending method tends to recommend popular ice and snow travel products, and neglects the personalized demands and potential mass markets of users, so that users are difficult to obtain more personalized and targeted ice and snow travel product choices, and the contact range of users to mass ice and snow travel products is limited. In addition, the inadequacy of the recommendation of long-tail products aggravates the unfairness of information filtering, not only affects the diversity of ice and snow travel products, but also limits the optimization of the sales strategy of the platform in ice and snow travel markets. Therefore, the existing recommendation method has not effectively solved the following problems that 1) the user behavior data and the product attribute data are difficult to effectively fuse, and the recommendation result is lack of individuation. 2) The accuracy of preference assessment for new users or products is not high, making the recommendation system perform poorly in the face of emerging markets. The prior art with the document number of CN116738066B discloses a rural tourist service recommendation method, device, electronic equipment and storage medium, which comprises the steps of obtaining tourist resources and tourist data, constructing a tourist resource portrait, carrying out emotion analysis on feedback data to obtain an emotion analysis result, determining a scoring matrix according to the feedback data and the tourist resource portrait, constructing a tourist portrait according to basic feature tags and feature preference tags, calculating potential similarity, basic similarity and preference similarity of a first tourist according to the scoring matrix and the basic feature tags and feature preference tags of the tourist portrait, generating a similarity matrix of the first tourist by weighting and fusing three kinds of similarity, determining the first tourist similar to the second tourist according to the similarity matrix, and recommending resources for the second tourist. The method realizes the reliability and accuracy of the related data of the tourist resources, and provides more accurate country tourist resource recommendation service for tourists. The prior art does not effectively fuse user behavior data with product attribute data. The prior art with the document number of CN115438871A discloses an ice and snow scenic spot recommendation method and system for eliminating popularity deviation by fusing preferences, which are technically characterized by grouping gender, age and occupation of tourists respectively, mapping the gender, age and occupation of the tourists into values of 0-1, calculating and obtaining multi-feature preference similarity of the tourists according to the mapping values of the gender, age and occupation of the tourists, calculating and obtaining a tourist preference value according to the multi-feature preference similarity of the tourists and tourist history scoring data, constructing a tourist-preference value matrix according to the standardized tourist preference value, training a matrix decomposition model based on the multi-feature preference of the tourists according to the tourist-preference value matrix and the tourist history scoring data, obtaining a trained tourist-preference value matrix, and predicting and recommending new users according to the trained tourist-preference value matrix. The method effectively relieves the unfavorable situation that low-popularity ice and snow scenic spots are difficult to recommend to tourists. But there is no in-depth consideration in terms of the preferences of the new user or new product. Disclosure of Invention In view of the above problems, the invention provides an ice and snow travel product recommending method bas