An ID-tree index strategy for supporting similarity searches in information filtering systems

Finding the nearest neighbor is one of the most important search operations in the field of information filtering. Aggarwal et al. have proposed SG-table strategy to index user profiles to support finding similar profiles for a query, reducing the number of accessed profiles. However, it can only improve some particular queries because of collisions in the index. Therefore, in this paper, to avoid the collision problem, we propose a signature-based structure, called the identifier tree (ID-tree), to index the signatures of user profiles via partitioning them into several subgroups globally. From our simulation, we have shown that the ID-tree strategy can reduce the most number of accessed profiles in the similarity search as compared with the SG-table strategy.