Improving Personalized Search on the Social Web Based on Similarities between Users
Published in The 8th International Conference on Scalable Uncertainty Management (SUM), (EI ), 2014
Authors: Zhenghua Xu, Thomas Lukasiewicz and Oana TifreaMarciuska.
Abstract: To characterize a user’s preferences and the social summary of a document, the user profile and the general document profile are widely adopted in existing folksonomy-based personalization solutions. However, in many real-world situations, using only these two profiles cannot personalize well the search results on the Social Web, because (i) different people usually have different perceptions about the same document, and (ii) the information contained in the user profile is usually not comprehensive enough to characterize a user’s preference. Therefore, in this work, in order to improve personalized search on the Social Web, we propose a dual personalized ranking (D-PR) function, which adopts two novel profiles: an extended user profile and a personalized document profile. For each document, instead of using a general document profile for all users, our method computes for each individual user a personalized document profile to better summarize his/her perception about this document. A solution is proposed to estimate this profile based on the perception similarities between users. Moreover, we define an extended user profile as the sum of all of the user’s personalized document profiles to better characterize a user’s preferences. Experimental results show that our D-PR ranking function achieves better personalized ranking on the Social Web than the state-of-the-art baseline method.