Real time evaluation of quality of search terms during query expansion for streaming text data using velocity and relevance

Abstract

The traditional methods of evaluation of retrieved data using precision and recall cannot be used to evaluate the quality of twitter data retrieved using the streaming API due to the restriction on the access of historic tweets, and the dynamic nature of the microblog posts. For this purpose, we propose a novel method to quantify the quality of data retrieved independent of the underlying model. Using the change in velocity of tweets due to addition of a search term, and the relevance calculated by the underlying model, we evaluate the impact of each search term on the quality of data retrieved.