Overview of the TREC 2013 Contextual Suggestion Track

Abstract

For participants familiar with the 2012 Contextual Suggestion Track we have provided a list of the main changes to this year's track: Contexts no longer include a temporal component (day of week, time of day, and season), contexts consist of only a location. Users were recruited from a crowdsourcing service (Mechanical Turk) as well as from the University of Waterloo student body. Suggestion attractiveness judgements are given on a 5-point, rather than a 3-point, scale. Submissions based off of the ClueWeb12 corpus were allowed in addition to submissions based off of the open web. A modified Time-Biased Gain (TBG) metric was used in addition to P@5 and MRR. This metric is described in section 4.3. The option to submit solely based on context or solely based or user profiles was removed. The file format used for profiles, contexts, and suggestions was switched from XML to CSV and JSON.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 2013
Accession Number
ADA599415

Entities

People

  • Adriel Dean-hall
  • Charles L. Clarke
  • Ellen Voorhees
  • Jaap Kamps
  • Nicole L Simone
  • Paul M. Thomas

Organizations

  • University of Waterloo

Tags

DTIC Thesaurus Topics

  • Abstracts
  • Crowdsourcing
  • Digestive System Processes
  • Information Operations
  • Information Retrieval
  • Judgment
  • Language
  • Latitude
  • Maryland
  • Ratings
  • Social Media
  • Social Networks
  • Standards
  • United States
  • Universities
  • Vector Spaces
  • Websites

Readers

  • Database Systems and Applications
  • Information Retrieval
  • Military History of the United States in the 20th Century.