Research On Recommendation Algorithm Based On The Mixture Of Text Sentiment Analysis And Matrix Decomposition

The conventional recommendation algorithms still have the problems of cold start,sparsity and scalability.Therefore,this paper proposed a hybrid recommendation algorithm based on Jaccard similarity and matrix decomposition to solve the sparsity problems by introducing users' preference information.Then,in order to improve the accuracy of English text sentiment analysis algorithm,an English sentiment analysis algorithm based on sentence vector and LSTM is proposed.Finally,considering the comment sentiment information contained in user reviews,this paper proposed a recommendation algorithms based on comment sentiment of users' and matrix decomposition.The primary research works of the thesis are given as follows:1.In order to introduce rating preference information of similar users,this paper proposed a hybrid recommendation algorithm based on Jaccard similarity and matrix decomposition.the algorithm uses the Jaccard similarity to mines user clusters with score similarity,and uses matrix decomposition method to predict the local score of similar users.Finally,the filled ratings matrix is calculated by matrix decomposition.Moreover,an example is given to verify the feasibility of HRA-JSMD algorithm.The experimental results show that the HRA-JSMD algorithm has better recommendation quality and less computation time.2.To improve the prediction accuracy of text sentiment analysis algorithm for English texts,this paper proposed a method named English Sentiment Analysis Based on Sentence Vector and LSTM.After English text pre-process,each word is assigned a number in order to construct sentence vectors and extracts their sentiment features by using TF-IDF.Finally,this algorithm uses LSTM neural network to train the data.Moreover,an example is given to verify the feasibility of SA-SVL algorithm.Finally,the experiments result show that SA-SVL algorithm can obtain better prediction accuracy in English text datasets.3.According to the user's historical comment information,this paper proposed a method named Recommendation Algorithms Based on Comment Sentiment of Users and Matrix Decomposition.The algorithm uses text sentiment analysis algorithm to introduce text sentiment value,and defines the deviation value of comment sentiment and rating standard value of user to integrate sentiment value into the actual ratings.Finally,uses matrix decomposition algorithm to predict the rating.I' he feasibility of RACSMD algorithm is verified by an example analysis of the algorithm.The experimental results show that the proposed RACSMD algorithm can effectively improve the recommend quality.The research contributions of this paper arc as follows.Jaccard similarity measure method is introduced to solve the problem that traditional recommendation algorithms only rely on historical rating to predict.The sentence vectors are constructed according to word frequency and sentiment featurcs in sentcence vectors are mined througn TF-IDF method.Basced ton the theory of recommendation algorithm,the sentiment information of user comments is introduced to get more consistent with the subjective score of users.Finally.the experiment result proved that the algorithm is effective and feasible.