TREC Challenge 2013

Problem Description

The goal of the TREC Challenge 2013 was to investigate search techniques considering as context only the user’s location, as well as, user interests via personal preferences and past history. More precisely, given a set of user profiles and a set of contexts, the objective is to suggest new context on users, tailored to reflect their preferences.


Our team ranked in the second place regarding MRR and TBG and third place regarding the P@5 measure in the TREC Challenge 2013 using personalized recommendations based on Rocchio-like and k-NN user-item queries. The corresponding publication has been presented in the Twenty-Second Text REtrieval Conference 2013, held in November 19-22, 2013.

There were 15 competing teams and the ranking of the top three teams is given below:

Team Affiliation MRR TBG P@5
UDInfo University of Delaware 0.6320 2.4474 0.5094
DUTH Democritus University Thrace & Athena Research Center 0.5955 1.8508 0.4090
simpleScore University of Lugano 0.5871 1.8374 0.4332