Endicott College at 2014 TREC Session Track

Abstract

Endicott College submitted three runs to Task 1 of the 2014 TREC Session Track. All runs re-ranked the baseline runs provided by the track organizers. One of the runs made use of a click graph to re-rank results for RL1, RL2, and RL3. The other two used relevance models computed over snippets from the session, and boosted their RL3 run using click graph recommendations. In the absence of clicks (e.g., RL1 and clickless sessions in RL2 and RL3), two of the runs used pseudo relevance feedback over the session, while the other used the unmodified baseline ranking. All runs used a similar pre-processing procedure, which we describe in Section 2. We then discuss our click graph re-ranking technique for the ECxCGxPRF run in Section 3 and the session relevance modeling technique for our ECxSRMxOS and ECxSRMxPRF runs in Section 4. We follow that with an analysis of the performance of our runs compared to each other, as well as the track minimums, medians, and maximums. Finally, we wrap up with some closing thoughts and future directions in Section 6.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 2014
Accession Number
ADA618733

Entities

People

  • Henry Feild

Organizations

  • Endicott College

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Abstracts
  • Algorithms
  • Data Sets
  • Demographic Cohorts
  • Failure Analysis
  • Feedback
  • Frequency
  • Information Operations
  • Instructions
  • Iterations
  • Preprocessing
  • Probability
  • Random Walk
  • Standards

Readers

  • Aerospace Research.
  • Information Retrieval
  • Systems Analysis and Design