A Personalized Recommender Algorithm Based on Semantic Tree
Zhaoguo Xuan, Haoxiang Xia, Jing Miao · 2011
Most of today's content-based personalized recommender algorithms make recommendations with respect to the match-making between the user and the item profiles, which are generally represented with eigenvectors. In conventional methods, the semantic relationships between the terms are missing, with only the frequency of the terms being taken into account. This would be a key factor that causes the poor recommendation results. To cope with this drawback, we in this paper propose a novel recommender algorithm, in which the user and the item profiles are both denoted as semantic trees, so as to incorporate the semantic information between terms when evaluating the similarity between the profiles. By taking the semantic similarity into account, the experimental tests illustrate that the similarity measure is more accurate with the proposed method and more reliable recommendations can then be made.