Research And Application Of Film Sequence Recommendation Model Based On Knowledge Graph

With the continuous development of information technology and the Internet environment,the accompanying information overload problem needs to be solved urgently.The common data sparse and cold start problems all affect the recommendation quality to a certain extent.Most of the existing recommendation algorithms are more inclined to consider item information,while ignoring user information.At the same time,they focus more on mining the static correlation between users and items,but ignore the user interest caused by the progress of time.The attenuation of,and the dynamic change of preferences,can not achieve the desired effect.In order to solve the above problems,this thesis proposes a user preference recommendation model based on double-end knowledge graph.The main research contents are as follows:In this thesis,a combined knowledge graph convolutional networks for Double side recommendation(CKD)algorithm is proposed,which extracts features on both the user side and the item side.The extraction of user features is obtained through the diffusion process of user preferences in the knowledge graph.For the extraction of item features,based on the knowledge graph convolutional network model,its neighbor information is aggregated into item nodes to generate embedding vectors.Finally,user interest propagation and item feature aggregation are alternately shared to share current known information,thereby improving the quality of recommendation.Based on the CKD model,this thesis presents a recommendation algorithm(CKD-LSP)that integrates the user’s Long and Short term Preference,and this method solves the problem of user interest caused by the passage of time.The gradual decay of,and the dynamic change of its preferences.This model first uses the CKD model to mine the user’s long-term potential preferences;then uses the hidden Markov model to obtain the user’s short-term interest preferences;finally,the above two are combined to provide personalized recommendations for target users.The algorithm not only covers the user’s overall personalized characteristics,but also grasps the user’s time-series characteristics with the help of user behavior sequences,thereby improving the recommendation effect.The model is compared in two datasets,Movie Lens-1M and Book-Crossing.The AUC and F1 values are used to compare it with the baseline method.The experiments show that the effectiveness of the algorithm is significantly improved.Finally,based on the research,this thesis designs and completes a video recommendation system based on double-end knowledge graphs and user preferences.The trained CKD-LSP model is imported into the recommendation system to provide users with satisfactory personalized video recommendations.After the requirement decomposition of the whole movie recommendation system implemented in this thesis,the design of each functional module and the design of the database,the implementation of the entire movie recommendation system on the user side and the management side is realized through development,which further verifies the effectiveness of the algorithm application.