Abstract
The enormous rapid growth of the online world and universal computing brought a wide range of choices for Internet users to obtain information of interest. However, the huge amount of new information released every day "big data" is greater than the human information processing capacity. As a result, it becomes harder and harder for users to obtain the required information quickly and they are also facing the problem of information overload. Collaborative Filtering (CF) systems play an important role in overcoming the information overload phenomenon by providing users with relevant information based on their preferences. CF is one of the best recommendation approaches that automate the process of the word-of-mouth paradigm. The most critical tasks in CF is finding similar users with similar preferences and then predict user rating to provide a personalized list of ranked items to the users. previous studies have almost exclusively focused these tasks separately to enhance the quality of recommendation. Nevertheless, we argue that these two tasks are not completely independent, but are part of an incorporated process. The purpose of this study is to propose a recommendation method that bridge the gap between the tasks of rating prediction and ranking to better grasp the best similar users to the target user by combing the advantage potential information of users review text clustering and user numerical ratings to enhance the CF recommendation methods proposed in the literature. The experimental results on three different datasets from Amazon show a considerable improvement over the baseline CF approaches in terms of recall, precision, and F1-measure.
Key words: Recommender systems; Collaborative filtering; Review text; Clustering; Top-N recommendation
|