Understanding and Predicting Characteristics of Test Collections in Information Retrieval
This work addresses the problem of understanding and predicting the quality of IR test collections, which is important for organizers of evaluation campaigns like TREC to optimize resource allocation and design better evaluations.
This paper investigates how the number of participating teams and other factors influence the quality of test collections in information retrieval evaluations. It also explores whether the reusability of a test collection can be predicted before human relevance judgments are collected, showing high accuracy predictions for successive years using the same document collection.
Research community evaluations in information retrieval, such as NIST's Text REtrieval Conference (TREC), build reusable test collections by pooling document rankings submitted by many teams. Naturally, the quality of the resulting test collection thus greatly depends on the number of participating teams and the quality of their submitted runs. In this work, we investigate: i) how the number of participants, coupled with other factors, affects the quality of a test collection; and ii) whether the quality of a test collection can be inferred prior to collecting relevance judgments from human assessors. Experiments conducted on six TREC collections illustrate how the number of teams interacts with various other factors to influence the resulting quality of test collections. We also show that the reusability of a test collection can be predicted with high accuracy when the same document collection is used for successive years in an evaluation campaign, as is common in TREC.