Know Your Personalization: Learning Topic level Personalization in Online Services

摘要

Online service platforms (OSPs), such as search engines, news-websites, ad-providers, etc., serve highly personalized content to the user, based on the profile extracted from her history with the OSP. Although personalization (generally) leads to a better user experience, it also raises privacy concerns for the user-she does not know what is present in her profile and more importantly, what is being used to personalize her content. In this paper, we capture OSP's personalization for an user in a new data structure called the personalization vector (η), which is a weighted vector over a set of topics, and present efficient algorithms to learn it. Our approach treats OSPs as black-boxes, and extracts η by mining only their output, specifically, the personalized (for an user) and vanilla (without any user information) contents served, and the differences in these content. We believe that such treatment of OSPs is a unique aspect of our work, not just enabling access to (so far hidden) profiles in OSPs, but also providing a novel and practical approach for retrieving information from OSPs by mining differences in their outputs. We formulate a new model called Latent Topic Personalization (LTP) that captures the personalization vector in a learning framework and present efficient inference algorithms for determining it. We perform extensive experiments targeting search engine personalization, using data from both real Google users and synthetic setup. Our results indicate that LTP achieves high accuracy (R-pre = 84%) in discovering personalized topics. For Google data, our qualitative results demonstrate that the topics determined by LTP for a user correspond well to his ad-categories determined by Google.