Research And Implementation Of Personalized Recommendation System Based On Hybrid Algorithm

With the rapid development of Information Technology(IT),the scale of the Internet is expanding,accommodating more users and information.Massive information resources have led to the problem of information overload.The recommendation system is an effective method to resolve it.However,existing recommendation systems still have some issues,i.e.,sparsity and cold start.In response to the above-mentioned problems,a similarity calculation method based on the Bhattacharyya coefficient is proposed to solve the data sparsity.Then,a tag-aware recommender model is proposed to solve the sparse,redundant,and semantic ambiguity in tag recommendation,alleviating the cold start.Finally,combining the above methods,a complete system is built to realize personalized movie recommendations for users.The main contributions in this thesis can be summarized as follows:(1)To solve the sparse data in the similarity calculation of traditional collaborative filtering algorithms,a similarity calculation method based on the Bhattacharyya coefficient is proposed.Specifically,the grading level of the items is classified by the custom classification criteria at first,which can realize the fast calculation of similarity between items in the same category and reduce the calculation cost.Then,the Bhattacharyya coefficient is used to calculate the item similarity from the perspective of the probability distribution,which breaks the restriction of common scoring items.Finally,considering the repeatability of the similarity calculation results among different items,the co-rated items enhancement function is used to optimize the similarity calculation method.Experimental results show that,compared with the existing methods,the improved similarity calculation can effectively solve data sparsity,improving the recommendation accuracy.(2)To solve the heterogeneous information fusion in recommendation system,a Hybrid Tag-aware Recommender Model(HTRM)is introduced.Specifically,HTRM uses the word embedding model to embed the score and the label,respectively.Then,the text feature extraction of the item label is carried out through the autoencoder.Meanwhile,the Long Short-Term Memory(LSTM)is used to conduct the feature extractions for user label behavior.Finally,a fully connected neural network is developed to predict the scores of fused users and items.Experimental results demonstrate that the optimized recommendation model can effectively reduce the prediction error of user rating,alleviating the cold start problem.(3)Based on the above proposals,a personalized movie recommendation system based on the B/S architecture is implemented,including the front-end and back-end.Specifically,the front-end is developed by the Vue framework,with which we build the Web-based interfaces.The back-end business is mainly implemented using the Spring framework.The core recommendation engine part adopts a collaborative filtering algorithm based on Bhattacharyya coefficient similarity calculation to enable the offline recommendation service.Moreover,it also provides real-time recommendation services using HTRM.In this way,our system successfully realizes the personalized movie recommendation for users.