Pharos: Social Map-Based Recommendation for Content-Centric Social Websites Wentao Zheng † Michelle Zhou ‡ Shiwan Zhao † Quan Yuan † Xiatian Zhang † Changyan Chi † † IBM Research – China ‡ IBM Research – Almaden ABSTRACT Recommendation technologies are widely used in online so- cial websites (e.g., forums and blogs) to help users locate their interests among overwhelming amounts of information. However, it is difficult to make effective recommendations for new users (a.k.a. the cold start problem) due to a lack of user information (e.g., preferences and interests). Further- more, the complexity of recommendation algorithms may not be easily explained, leaving users with trust issues in rec- ommendation results. To tackle the above two challenges, we are building Pharos, a social map-based recommender system. A social map summarizes users’ content-related so- cial behavior (e.g., reading, writing, and commenting) over time as a set of latent communities. Each community de- scribes the content being discussed and the people involved. Discovering, ranking, and recommending “popular” latent communities on a social map, Pharos enables new users to grasp the dynamics of a social website, alleviating the cold start problem. In addition, the map can also be used as a context for making and explaining recommendations about people and content. We have deployed Pharos internally and the preliminary evaluation shows the usefulness of Pharos. Author Keywords Cold start, explanation, recommender systems, social web- sites, trust ACM Classification Keywords H.3.3 Information Search and Retrieval: Information filter- ing; H.5.3 Information Interfaces and Presentations: Group and Organization Interfaces—Collaborative computing INTRODUCTION In recent years, content-centric social websites (e.g., forums, wikis, and blogs) have flourished with an exponential growth of user-generated information. It becomes increasingly more difficult for people to navigate these sites and locate desired information. Thus, researchers have developed recommen- dation technologies to help people better find desired infor- mation at online social websites [4, 5]. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Workshop SRS’10, February 7, 2010 Hong Kong, China Copyright 2010 ACM 978-1-60558-995-4... $10.00 However, there are two prevalent challenges when building such a recommender system. First, it is difficult to make effective recommendations if a system knows little about its users or the items to be recommended. This is also known as the cold start problem [10, 11]. In the case of a social website, a recommender system knows little about a user who is new to the website or has few connections to others. As a result, it would be difficult for the system to guess what a user is looking for and make effective recommendations. Second, it is difficult to explain recommendation rationales to end users to make the recommendation more trustworthy. A recommender system usually utilizes complex algorithms or inferences to make recommendations. It is difficult for average users to interpret and comprehend the process and entrust themselves to the recommended results [5, 7]. To address the above two challenges, we are building Pharos, a social map-based recommender system. Here, we use the term social map to refer to a dynamically generated ma- rauder’s map 1 of a content-centric social website. Such a social map summarizes users’ content-related social behav- ior (e.g., reading, writing, and commenting) over time as a set of “latent communities” (Figure 1-a). Each latent com- munity characterizes the implicit connections among a set of users and the content they generated. It thus consists of two parts, explaining what is being talked about (content summa- rized in green) and who are involved (people summarized in blue). Based on the generated social map, Pharos then uses a set of criteria (e.g., people’s social status and content popularity) to recommend “hot” communities, “hot” content or people in each community. For example, Figure 1-(b, c) shows the lists of content and people being recommended re- spectively. As a result, a user especially a new user can use the social map to get a quick glimpse of a content-centric social website and learn about the website’s dynamics (e.g., popular content and people). In this sense, Pharos uses a so- cial map to recommend “hot communities” and relevant “hot items” (i.e., content and people) to all users, alleviating the cold start problem. Furthermore, the social map provides a natural context for users to grasp the recommendation results (e.g., hot communities and hot items), since it helps explain the existence of latent communities and their characteristics (e.g., highly influential content and individuals). In short, Pharos offers two unique contributions. First, it ad- dresses the cold start problem in part by using a social map to summarize a social website, which in turn helps new users 1 http://en.wikipedia.org/wiki/Magical objects in Harry Potter 1