Relevancy Ranking of User Recommendations of Services based on Browsing Patterns Suresh Kumar Gudla, Joy Bose, Venugopal Gajam Samsung R&D Institute Bangalore,India suresh.gudla@samsung.com Srinath Srinivasa IIIT Bangalore Bangalore, India sri@iiitb.ac.in Abstract— There are a number of inbound web services, which recommend content to users. However, there is no way for such services to prioritize their recommendations as per the users’ interests. Here we are not interested in generating new recommendations, but rather organizing and prioritizing existing recommendations in order to increase the click rate. Since users have different patterns of browsing that also change frequently, it is good to have a system that prioritizes recommendations based on the current browsing patterns of individual users. In this paper we present such a system. We first generate the clusters of article topics using URLs from the users’ browsing history, which is then used to generate the relevancy scores of the recommendation services based on entropy. The relevancy scores are then fed to the service providers, which use them to prioritize their recommendations by ranking them based on the relevancy scores. We test the model using the browsing history for 10 users, and validate the model by calculating the correlation of the generated relevancy scores with the users’ manually provided topic preferences. We further use collaborative filtering to benchmark the usefulness of our ranking systems. Keywords— web services; clustering algorithms; relevancy ranking I. INTRODUCTION Currently there is no way for the content providers of web services, such as news aggregators, to know about the user’s changing interests in real time and provide tailored content to individual users, deduced from their browsing patterns. Most content providers or aggregators rank their content based on their business priorities or as per their limited knowledge about users of their services. All the content providers can know about individual user preferences consists of generalized categories of preferences based on users having certain profiles, or the list of preferred categories manually provided by the users. However, it is not always easy to categorize web content into neatly defined categories. Personalized recommendations for individual users rather than profile based recommendations are always preferable, especially since the user preferences may change frequently. Also, it is not easy to collect incremental data and modify the algorithm running on the server in real time. In this paper, we propose a method to enable the content providers to better understand the user and provide content in the order of the users’ predicted preferences. Here we are not providing or changing recommendations to the user, only ranking the existing recommendations of the content providers in an order that the users might find relevant. A recommendation system needs either search query parameters or the direct context of the user. In our system, we have a set of recommendations already generated on various parameters of business partners. Our novelty exists in ranking the already available set of recommendations based on the users' browsing patterns. We take the content of the URLs browsed by the user and run a clustering algorithm to generate the clusters of topics browsed for the recent period of time. Then, taking the recommendations of the service providers, we generate the probability distribution of each recommendation against the cluster of topics, finally computing the relevancy scores for the recommendations based on entropy. The computed relevancy scores are fed back to the service providers, who use it to rank and prioritize their recommendations. As the browsing patterns keep changing on a daily basis, the recommendations get re-ranked to suit the users' needs, which help in improving the click rate of these recommendations in line with the users' interests. In addition, the users' privacy is strictly maintained as the users' browsing patterns are not exposed to third party business partners. II. RELATED WORK There is a large body of work in the recommender systems (RecSys) community that seeks to understand the users’ interests in real time using various methods such as collaborative filtering, content based recommendations and user profiling, and accordingly recommend relevant services, emails or notifications. The focus of our work is somewhat different from these, since we do not generate more relevant recommendations; rather we rank the existing recommendations by analyzing the user’s recent browsing behavior. Moreover, our algorithm has to deal with changing user interests, and so has to update in real time. Pan et. al. [1] derived a content matching engine to generate context aware push notification recommendations which are in-line with the user’s current context. Sun et. al. [2] developed an approach to recommend personalized articles to scientific researchers, using the rankings generated from keyword similarity, journal similarity, and author similarity to measure the relevance of the articles to researchers. Jonnalagedda et. al. [3] presented a personalized news recommendation system, where the rankings of the articles