Content Based And Collaborative Filtering Based Video Hybrid Recommendation Research

In the digital media era,video media is popular with the public for its distinctive features such as intuitive content and vivid stereo.As an important platform for entertainment and leisure,video websites provide users with a large number of highquality video resources.However,with the continuous expansion of video libraries,it is often difficult for users to quickly find videos of interest,then research on personalized video recommendation algorithms has emerged.For the recommendation algorithm,whether the potential features of users or videos can be quickly extracted from the historical data is crucial,which will directly affect the recommended performance of the algorithm.However,in the practical application,the video recommendation algorithm usually faces problems such as extremely sparse data,inaccurate feature extraction,and performance bottleneck of a single recommendation algorithm,resulting in poor final recommendation.Based on this background,this paper will analyze the advantages and disadvantages of various recommendation algorithms,and achieve the fusion of data sources by integrating multiple single recommendation algorithms into a video hybrid recommendation algorithm,enhance the accuracy of feature extraction,and improve the performance of recommendation algorithms.The main works as follows:For problem of data sparsity,a video hybrid recommendation algorithm based on collaborative filtering and content recommendation is proposed.The algorithm calculates the similarity between the videos by considering the category information of the video,the content profile information and the user historical score data comprehensively,and then predicts and fills the scores of the missing items into the user-video score matrix,thereby alleviating the user-video scoring data sparsity issues.For the feature extraction problem,a recommendation algorithm based on PVMF(Paragraph Vector-Probabilistic Matrix Factorization)is proposed.The algorithm uses the Doc2 Vec model to extract the video content features from the video profile information,and uses the extracted video content features to initialize the video latent feature matrix in the probability matrix factorization algorithm,thus combining the content-based recommendation algorithm and the probability matrix factorization algorithm.The feature extraction capability helps the aggregated video hybrid recommendation algorithm to learn more accurate video potential features in the optimization process.For the "bottleneck" problem of single recommendation algorithm performance,a video hybrid recommendation algorithm based on content and PV-MF is proposed.The algorithm uses a neural network model to combine two single recommendation algorithms to learn the nonlinear combination characteristics between different recommendation algorithms.The experimental results show that the Root Mean Square Error of this algorithm is lower than two original algorithms by 13.93% and 11.01% respectively.The three video hybrid recommendation algorithms combine different recommendation algorithms from the perspectives of data source fusion,feature enhancement and recommendation results blending,and achieve better recommendation performance than original algorithms.Therefore,the research on hybrid recommendation algorithm has some referenced significance.