A Smart TV Program Recommendation System Combined With Video Stream Processing

Due to the development of information technology and internet technology,along with the sharp increase of data amount,it is difficult to get the information and services which meet the actual demands of people.How to search and deal with these data efficiently and carry out personalized recommendation become the pressing problems need to be solved.Thus,a smart TV program recommendation system combined with video stream processing is proposed.The system is able to realize the collection of user's behavior records and recommend personalized video programs the user.The recommendation system is running in a distributed parallel computing framework Spark to improve the operation efficiency of the system as a whole.The research of this paper is mainly focused on the following three aspects.1.Mark recognition algorithm based' on the recursive convolution neural network is studied.Data acquisition module of the recommendation system is built on the basis of the algorithm.Aiming at the deficiencies of the traditional mark recognition algorithm in terms of recognition accuracy and speed,mark recognition algorithm based on the recursive convolution neural network is proposed in this paper.The recursive network model cascades two convolution neural network models.In order to improve the recognition precision,each level of network is connected by a kind of linked list.Different mark samples are used for the training of the two-stage model.The experiments prove that the accuracy of the recursive convolution neural network designed in this paper exceeds 99.3%with the average recognition time less than 0.15s.2.A recommendation algorithm based on the hybrid model is built.A recommendation algorithm based on the hybrid model is proposed in the paper focused on the characteristics that neighborhood model is easily influenced by data sparsity through the research of recommendation algorithm.Two recommendation algorithms of the hybrid model have different pertinences.Alternating Least Square(ALS)algorithm is able to obtain the global data feature and has good recommendation effect for sparse data.The algorithm based on the neighborhood model emphasizes local relevance and is suitable for the recommendation task in which that data is not sparse.Different recommended strategies in the hybrid model are used by distinguishing different sparse degrees of users' data to enhance the recommendation accuracy and practicability of the model.3.A smart TV program recommendation system based on parallel computing framework Spark is built.Spark is a distributed computing framework developed by AMP laboratory which is able to process data quickly and efficiently.Aiming at the computing tasks of huge amount of data involved in the entire recommendation process,Spark is used in the construction of the recommendation system in this paper.Firstly,the basic concept and application method of Spark are introduced.Then mark recognition algorithm based on the recursive convolution neural network and the hybrid recommendation algorithm model designed in this paper are realized on the Spark platform.And the promotion effect of parallel programs to the running efficiency of the system is verified through contrast experiment.Experiments prove that the recursive convolution neural network model proposed in this paper is able to meet our requirements in recognition accuracy and speed.The hybrid recommendation algorithm model is able to effectively avoid the precision decline caused by data sparsity.At the same time,the operation efficiency of the system is effectively improved through the use of Spark parallel computing framework.