Research And Implementation Of Video Recommendation System Based On Interest Fusion And Spread

In the Internet era,with the increase in the number of videos,users need to spend a lot of time and energy to find the videos they want to see.In order to solve this problem,personalized recommendation technology came into being.Personalized recommendation is to recommend video information of interest to users according to their interest characteristics and viewing behavior.However,the existing recommendation algorithms do not take into account the importance of cross features,which will affect the final recommendation effect.Because not all cross features can work on the final prediction,useless cross features may produce noise.To solve this problem,this thesis proposes an I-RippleNet model based on hybrid recommendation algorithm.The I-RippleNet model obtains the potential interest preference of one-hop propagation by cross-integrating features into one-hop propagation,and then performs multi-hop propagation in turn to form interest diffusion to obtain the final click probability prediction and realize the Topk recommendation of video.The main research contents of this thesis are as follows :Firstly,AFM algorithm is a classical model based on attention mechanism for feature crossover,but there is still a problem of insufficient information utilization in feature extraction.Therefore,this thesis proposes a new feature crossover model FAFM.The F-AFM model introduces user history features based on the cross-fusion of target user features and video features.The target user features and user historical features are self-crossed to obtain the weight value of each user feature,and the corresponding attention score is obtained according to the weight value.Then,the target user features and video features affected by historical features are crossed on F-AFM.More accurately capture the user ’s interest.Secondly,a new I-RippleNet model is proposed.I-RippleNet can spread interest according to seed nodes on the knowledge graph.Starting from the seed node,the onehop to multi-hop diffusion propagation is carried out through the directed edges on the knowledge graph.Based on this,this paper combines F-AFM algorithm with RippleNet,and obtains the potential interest preference of one-hop propagation by integrating FAFM into one-hop propagation.Then,interest diffusion is performed on I-RippleNet to obtain the final prediction probability,which also provides interpretability for recommendation.Finally,this thesis verifies the algorithm on the Movielens dataset and compares it with various algorithms.It is proved that the proposed I-RippleNet hybrid recommendation model can effectively alleviate the impact on the recommendation effect due to the lack of consideration of cross features and improve the accuracy of recommendation prediction.This thesis constructs a personalized video recommendation system,applies the algorithm to the actual scene,and designs the demand analysis and function of the video recommendation system.At the same time,considering the management of the data information of the recommendation system,the background data management system is designed.Finally realized the personality for users...