Enhancing GLSLIM Using User Preference Change Marking Algorithm

In this paper, we proposed a data preparation method, User Preference Change Marking (UPCM) algorithm, which aims to boost the performance of top-N recommendation. Integrating with a global-local recommendation, our method shows that the prepared datasets favor the recommendation method significantly. Global-and-local recommendation method not only considers the item-item similarity but also the user cluster one. The method uses two important characteristics of data to discover the trend of user preferences: (i) categorical dimension of the data, such as movie genre, product catalog, etc, and (ii) timestamp of the relation, such as timestamp of when a user rated a movie. The categorical dimension is useful in projecting additional information about the rating instead of focusing on the item relation solely. The timestamp is used to calculate rating frequencies and then activity trend. These two combined help our algorithm to remove inactive users from the dataset. As a result, the prediction performance is improved especially for the users who provide sufficient rating data.