Research On Hybrid Recommendation Algorithms Based On Auto-Encoder

With the rapid development of Internet information technology and the popularization of various electronic devices,many Web sites and APPs have emerged as the times require,bringing great convenience to people's daily life,making people's clothing,food,shelter and transportation more convenient and faster.This in turn promotes the growth of users and information,and the complexity and dynamics of the content are also increasing.The era when information and people's relations are inseparable has become a non-material tool for people's survival and development.However,the explosive growth of information has also brought troubles to people,traditional search engines and tools are time-consuming and labor-intensive to find,and simple retrieval systems can no longer meet the needs of users;online resource information changes at any time.It also increases the difficulty people need to find accurately;Frequent changes in people's interests also bring great difficulties to accurate recommendations.In response to the above problems,researchers have proposed many solutions in recent years,but due to the complexity of recommendation scenarios,item types,and completeness of data sets,the recommended accuracy of these solutions still has much room for improvement.This paper mainly studies the user's preference for movies,combines the deep learning technology with the recommendation system,and uses the user information,movie information and user's rating information of the movie provided by MovieLens as the input of the recommendation system to match the matching movie.Resources recommend users.The main work of this paper is:(1)Firstly,considering the long-term stability of user interest,a long-term model of user interest in movies is extracted based on the noise reduction self-encoder;(2)Secondly,considering the dynamics of user interest,a short-term model of user interest in movies is proposed,and short-term interest features are extracted based on time-series-sensitive RNN technology in deep learning;(3)Finally,the user's long-term and short-term interest model of the movie is mixed,that is,the short-term interest change disturbance is added on the basis of long-term stability of the user's interest,and a hybrid model recommended by the user is constructed;(4)Through the actual experimental design,verify the model selection,the actual parameter selection and the correctness and effectiveness of the proposed hybrid recommendation model.