Show simple item record

DUTH at TREC 2013 Contextual Suggestion Track

dc.contributor.authorDrosatos, George
dc.contributor.authorStamatelatos, Giorgos
dc.contributor.authorArampatzis, Avi
dc.contributor.authorEfraimidis, Pavlos S.
dc.description.abstractIn this report we give an overview of our participation in the TREC 2013 Contextual Suggestion Track. We present an approach for context processing that comprises a newly designed and fine-tuned POI (Point Of Interest) data collection technique, a crowdsourcing approach to speed up data collection and two radically different approaches for suggestion processing (a k-NN based and a Rocchio-like). In the context processing, we collect POIs from three popular place search engines, Google Places, Foursquare and Yelp. The collected POIs are enriched by adding snippets from the Google and Bing search engines using crowdsourcing techniques. In the suggestion processing, we propose two methods. The first submits each candidate place as a query to an index of a user’s rated examples and scores it based on the top-k results. The second method is based on Rocchio’s algorithm and uses the rated examples per user profile to generate a personal query which is then submitted to an index of all candidate places. The track evaluation shows that both approaches are working well; especially the Rocchio-like approach is the most promising since it scores almost firmly above the median system and achieves the best system result in almost half of the judged contextprofile pairs. In the final TREC system rankings, we are the 2nd best group in MRR and TBG, and 3rd best group in [email protected], out of 15 groups in the category we participated.en_UK
dc.relation.ispartofseriesProceedings of the Text REtrieval Conference (TREC), NIST;
dc.titleDUTH at TREC 2013 Contextual Suggestion Tracken_UK

Files in this item


This item appears in the following Collection(s)

Show simple item record
Except where otherwise noted, this item's license is described as