San Francisco State University (SFSU) at Total Recall Track of TREC 2016

This paper describes the participation of San Francisco State University group in Text Retrieval Conference (TREC) 2016 Total Recall Track from National Institute of Standard and Technology (NIST). The TREC series provide large test collections and judgements for participant to design Information Retrieval (IR) systems for different proposes. The purpose of Total Recall Track is seeking text search system which achieves high recall with minimum number of return documents. This year, our team participates all automatic tasks, including 34 topics in athome task and 2 datasets in sandbox task. Our system is built based on the autonomous technology-assisted review (Auto TAR) model[1], which is also the baseline of Total Recall Track. In this paper, we will introduce several approaches which have improved the evaluation metrics compare to the baseline model. Our enhanced model combines seed expansion and feature engineering including adding n-gram, eliminating stop words, and preserving words contain digits.