Research On Feature Interactive Personalized Recommendation System Based On Recall Ranking Two-Tier Model

Information overload is a common problem in the 21 st century.In news,music,film and television,e-commerce and other fields,the amount of Internet content has shown explosive growth.Recommendation system can help Internet content providers to locate users’ interests and provide content services in line with users’ needs,and also help users avoid the influence of information overload.In a specific scenario,the recommendation system can automatically push the content that the user is interested in.In this era,recommendation system emerges at the historic moment.It is of great practical significance to study recommendation algorithm and recommendation system architecture for promoting the development of recommendation system.Therefore,this paper proposes a new recall layer and a new sorting layer algorithm based on the classic recall and sorting model of recommendation system,and applies them to a specific recommendation scenario.In this paper,a recall layer algorithm based on Item2 vec is proposed.The algorithm first uses Item2 vec technology to obtain the embedding vector of each candidate item,and then uses the locally redundant sensitive hash neighbor computing recall layer algorithm proposed in this paper to quickly screen out candidates similar to the target item from massive data.A series of experiments on four classical data sets show that the comprehensive performance of the proposed algorithm is better than the existing locally sensitive hashing algorithm.In this paper,we propose a multi-order model for predicting click-through rate based on attention mechanism.The model predicts click-through rate of candidate items from recall layer and obtains N most relevant recommendations to complete the sorting task.The algorithm fully considers the influence of the first-order feature,second-order feature and high-order feature of the candidate on the final prediction results.At the same time,attention mechanism was introduced in the second stage of feature crossover to further enhance the performance of the model.In the experimental part,in order to ensure the fairness of experimental results,under the same experimental conditions,this paper reproduced five classical CTR prediction models or similar models,and carried out comparative experiments on two classical data.Experimental results show that the proposed sorting layer algorithm based on multi-order feature interaction is superior to the existing algorithm models.Based on the above proposed recall layer algorithm and sorting layer algorithm,this paper implements a movie recommendation system.This paper first fully expounds the function and performance requirements of the recommendatory system,and puts forward a recommendatory system architecture scheme with clear structure and complete functions.For each component of the recommendation system,this paper has carried on the detailed design and implementation,and has carried on the full elaboration to the internal principle.The core part of the recommendation system uses the recall layer and sorting layer algorithms proposed in this paper,which can ensure the performance of the architecture.At the same time,the functions of each component of the architecture are clear and easy to expand,which has a good reference for the current recommendation system architecture scheme.