Query-biased Summarization Track

Overview

The purpose of this track is to evaluate methods of query-biased summarization of text documents.

For this track the standard procedure is used.

Collection

Only documents listed in the task description should be annotated by participants.

Task Description for Participating Systems

Each participant is granted access to the collections and a list of tasks. Each task is a (query, document) pair. For each task a participating system must generate an annotation of the document for the given query.

The list of the tasks is based on the set of queries which were used in the ROMIP'2007 workshop.

Expected result for each task is a plain text snippet not longer than 300 characters. Any html tags are not valid (including <br> and <p>).

Evaluation Methodology

• instructions for assessors:
  Assessors evaluate document relevance to a query basing on the annotation of the document (generated by a participating system) without seeing the document itself.
  For each result being evaluated assessors obtain a title (no longer than 100 characters) and an annotation (not longer than 300 characters).
• relevance scale
• yes / probably yes / perhaps yes / no / impossible to evaluate
• yes / no / impossible to evaluate
• official metrics:
  measure of concordance of metrics obtained for full documents in the ad hoc track and for annotations in this track

Data Formats

• collections
• tasks
• results