Improving Contextual Suggestions using Open Web Domain Knowledge

Also published online by CEUR Workshop Proceedings (CEUR-WS.org, ISSN 1613-0073)Contextual suggestion aims at recommending items to users given their current context, such as location-based tourist recommendations. Our contextual suggestion ranking model consists of two main components: selecting candidate suggestions and providing a ranked list of personalized suggestions. We focus on selecting appropriate suggestions from the ClueWeb12 collection using tourist domain knowledge inferred from social sites and resources available on the public Web (Open Web). Specifically, we generate two candidate subsets retrieved from the ClueWeb12 collection, one by filtering the content on mentions of the location context, and one by integrating domain knowledge derived from the OpenWeb. The impact of these candidate selection methods on contextual suggestion effectiveness is analyzed using the test collection constructed for the TREC Contextual Suggestion Track in 2014. Our main findings are that contextual suggestion performance on the subset created using OpenWeb domain knowledge is significantly better than using only geographical information. Second, using a prior probability estimated from domain knowledge leads to better suggestions and improves the performance