An Online Video Recommendation System Based On Deep Neural Network

With the coming of the information age,the amount of video data is increasing day by day.On the one hand,video providers hope to obtain user preferences,match up to the user to push his interest in video,improve the user experience,enhance the user stickiness of the company;on the other hand,online video users hope from the massive video content,quickly find themselves interested in the video,reduce unnecessary time overhead.It is these urgent needs that make personalized video recommendation become a hot research topic.In this background,this paper studies the personalized recommendation algorithm of video,and proposes two kinds of personalized recommendation algorithms based on the deep learning technology in Natural Language Processing.1)Video personalized recommendation algorithm based on depth semantic model.The algorithm combines depth learning and content-based recommendation algorithm,and constructs a depth network model to extract the feature of video and user's information.At the semantic level,the distributed representation of the video and the user is realized,and the potential connection between the user and the video is deeply excavated.2)Video personalized recommendation algorithm based on probability of language model.The algorithm first will get the video sequence,the user history to watch the analogy to natural language sequence,the sequence of each video as a statement of the separate words,Then combined with the depth study on natural language processing technology,according to the correlation between different video size,building and training the neural network model,distributed characteristics of the video said.And on this basis,using video vector similarity between collaborative filtering recommendation.Respectively in this paper,first of all,through the theoretical derivation of the proposed two kinds of recommendation algorithm,this paper expounds in detail the principle of thetheoretical analysis of the feasibility of them.Then according to the related principle of the algorithm,based on the collected the user history click behavior of Tencent video data and video text description data has carried on the experimental design.And converting the proceeds respectively with the results of the User CF,Item CF,TF-IDF algorithm compares the results of analysis,etc.The experimental results show that the presented two kinds of recommendation algorithm is superior to contrast the result of the algorithm,they can be very good video recommendation task.