University of Essex at the TREC 2012

The primary goal of our participation in the Session track is to further evaluate our anchor expansion technique proposed in the previous year [1]. In particular, we aim to test the effectiveness of this approach on a more realistic dataset collected this year. SWIRL 2012 noted that there is still a large gap between the study of users and the study of IR algorithms [2], so the session data collected for this year’s session track aimed at addressing some of the suggestions made in the SWIRL 2012 report . We exploit the anchor logs to derive query expansions that are relevant to user information needs throughout the session. Anchor text has shown to be effective for a variety of information retrieval tasks. This include the diversity task in the TREC Web track [3]. Anchor text can be considered as a replacement to user queries as often web authors use similar labels to describe web pages to those used by searchers to find them [4]. Moreover, Dang and Croft have recently shown how anchor text can be used to simulate user sessions. They have considered all the anchor text pointing to the same document as queries in the same user session [5]. We derive query expansions from the anchor logs by using anchor logs as simulation of query logs to derive related terms or phrases that represent the user information needs in the session (RL2). We also use further interaction data to create a better model of the user interests. We consider the documents displayed throughout the session as good indicators of the user’s interests which is a similar approach to the pseudo relevance feedback model, e.g. [6] (RL3). In addition, we consider only those clicked as useful indicators which is a similar approach to the implicit relevance feedback models, e.g. [7, 8] (RL4). Following this, we submitted two runs to the Session track: