| In the field of E-commerce, with the emergence of information overload, recommender system has been widely used. Nowadays,recommender system is mainly recommended by analyzing user rating,ignoring the important role of user reviews in the process of recommendation, thus affecting the accuracy of recommendation results.In view of this problem, this paper mainly carries on the following researches:(1) We propose a method called Sentiment Analysis based on Optimal Statistics(SAOS) and Semantic Meaning Selection(SASMS),which can quantify the results of sentiment analysis and apply to rating prediction. Realizing the automatic sentiment analysis and rating prediction for user reviews.(2) A method of reviews recommendation based on topic model is proposed. It generates topic distribution by topic extraction of user reviews and forms user's topic distribution preference. Then using the distance between topic distribution of review and topic distribution preference of user, combining the helpful of user reviews, to sort the list of reviews, thus realizing the personalized recommendation for user reviews.(3) We propose a method of item similarity calculation based on user reviews. It uses the topic model to generates topic distribution feature of items, which using to calculate the similarity between items.Then applying the similarity on the item-based collaborative filtering recommendation.(4) A method of item recommendation based on user rating and reviews is proposed. It uses sentiment analysis on user reviews to get the prediction rating, generating extend rating by combing the prediction rating and the real user rating, using the extend rating to carry out collaborative filtering recommendation, then adjusting recommendation result by using the topic distribution distance between users and items.This paper makes in-depth analysis on user reviews, using the integration of user reviews and collaborative filtering technology to build a recommender system based on user reviews. |
PDF Full Text Request