A Similarity Search System Based on the Hamming Distance of Social Profiles

Abstract

The goal of a similarity search system is to allow users to retrieve data that presents a required similarity level in a certain dataset. For example, such dataset may be applied in the social media scenario, where huge amounts of data represent users in a social network. This paper uses a Vector Space Model (VSM) to represent users' profiles and the Random Hyper plane Hashing (RHH) function to create indexes for them. Both VSM and RHH compose an alternative to address the challenge of performing similarity searches over the huge amount of data present in the social media scenario: the Hamming similarity. In order to evaluate the effectiveness of our proposal, this paper brings examples of reference profiles, used for performing queries, and presents results regarding the correlation between cosine and Hamming similarity and the frequency distribution of Hamming distances among identifiers of users' profiles. In short, the results indicate that Hamming similarity can be useful for the development of similarity search systems for social media.