Finding, Weighting and Describing Venues: CSIRO at the 2012 TREC Contextual Suggestion Track

Abstract

We report on the participation of CSIRO in the TREC 2012 contextual suggestion track, for which we submitted four runs. Two submissions were baselines that investigate the performance of a commercial system (namely the Google Places API), and whether the current experimental setup encourages diversity. The remaining two submissions were more complex approaches that explore the importance of time and personal preference. For the former, check-in statistics provided by Foursquare were used to identify which times of day and which days of week venues are more likely or less likely to be frequented. For the latter, textual similarity was used to weight venues with respect to positive and negative examples provided for each profile. Our submissions all fall either slightly above or slightly below the mean, depending on how they are judged. Interestingly, our baselines consistently outperform our more complex submissions which suggests that a) venue quality (as given by Google review score) is a more important signal than either time or personal preference, at least in the context of this evaluation, and b) that the evaluation is biased to a specific type of venue, namely pubs.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 2012
Accession Number
ADA579317

Entities

People

  • Cecile Paris
  • David Milne
  • Paul M. Thomas

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Abstracts
  • Algorithms
  • Australia
  • California
  • Complex Systems
  • Computing-Related Activities
  • Data Science
  • Governments
  • Graphs
  • Histograms
  • Information Retrieval
  • Information Science
  • New York
  • Sensitivity
  • Standards
  • Statistics
  • Test And Evaluation

Readers

  • Computational Modeling and Simulation
  • Information Retrieval
  • Mathematics or Statistics