Research On Music Recommendation Integrating Sentiment Analysis Of User Reviews

With the rapid development of the Internet,more and more users download their favorite music products on the Internet.However,the increasing number of music products on the Internet also makes it difficult for users to filter out their favorite music.Traditional search engines have been unable to satisfy users to choose from massive amounts of data.But such a huge Internet can also allow researchers to obtain a large amount of music-related data.Through these data,it has become an important research content to develop a system that allows users to personalize music recommendations according to their personal interests.In the existing music recommendation methods,most of them are content-based recommendations,that is,users query the system by providing their favorite music,and the system recommends relevant or similar music recommendation lists to the user in response.Rely on the similarity between music to generate recommendations that match user interests.This recommendation method can make relatively good recommendations when each music has complete historical data.However,in reality,the historical data of many music is relatively sparse,making it difficult to make accurate recommendations,and the recommendation results are relatively single.In order to solve the above problems,this thesis proposes a music recommendation model based on user reviews,which analyzes the unique information contained in user reviews,and user reviews often include the user’s true feelings,and uses this information as music-related data to make recommendations,that is,in the extended It not only provides a way to obtain music-related historical data,but also improves the degree of personalization of recommendations to a certain extent.This thesis uses review text as basic information to design a recommendation algorithm that combines user-based collaborative filtering and item-based recommendation.Among them,the user-based collaborative filtering recommendation is to use the Bi-LSTM model to conduct sentiment analysis on the review text,and after obtaining the emotional rating expressed by the review text,combine it with the user’s rating to form a user-scoring matrix,and use the Pearson correlation coefficient to calculate the user similarity to obtain a similar user set,and then use the similar user set to predict the music score of the target user,and make recommendations based on the score.However,collaborative filtering recommendation is often accompanied by problems such as cold start and data sparseness.To solve this problem,this thesis proposes a second model,Item-based recommendation is to extract the topic of the review text through the LDA topic model,so as to obtain the topic probability distribution of music,and then calculate the similarity between different music by calculating the probability distribution distance,and use the method of weighted summation to predict the user’s preference.The score of the album is used to get the recommendation list,that is,a more accurate recommendation list can be obtained according to the user’s lower behavior,finally,the two recommendation methods are mixed through the waterfall mixing method,and the output list of the first model is used as the second model.The input of each model is recommended to obtain an accurate recommendation result.Finally,the model proposed in this thesis is compared and analyzed.By comparing the recommendation effects of the two separate models and the fusion,and comparing the effects of different sentiment analysis and text similarity calculation methods,the final results show that based on user reviews.The recommendation method can effectively improve the accuracy of the music recommendation system.