Visualizing Basic Words Chosen by Latent Dirichlet Allocation for Serendipitous Recommendation

The use of recommender systems to support users' search and selection of items in an information overloaded environment is widespread. Usually, precision and recall are utilized for evaluating a recommender system. However, an alternative measure should be considered, because a user's satisfaction is the most important factor to be considered when constructing a recommender system. Briefly, there exists a novel technique, serendipitous recommendation that considers this factor. In this paper, we propose a new method for constructing a novel serendipitous recommendation technique. In our method, we utilize a map of basic words that shows the semantic relationships between words. The basic words selected by Latent Dirichlet Allocation (LDA) are arranged on the map by principal components analysis (PCA). As a result, they are composed of semantically connected word pairs. We believe that this map is useful for searching and selecting items, because the user can find serendipitous words.