Biological question answering with syntactic and semantic feature matching and an improved mean reciprocal ranking measurement

Abstract

Specific information on biomolecular events such as protein-protein and gene-protein interactions is essential for molecular biology researchers. However, the results derived by current keyword-based information retrieval engine contain a great deal of noisy information, which forces biologists to use a combination of several keywords to locate information. To resolve this problem, we propose a question answering (QA) system that offers more efficient and user-friendly ways to retrieve desired information. In addition, QA system measurements may suffer from the same score problem, so the evaluation of a QA system may be unfair. An improved mean reciprocal rank (MRR) measurement, mean average reciprocal rank (MARR), and an efficient formula to reduce the computational complexity of the MARR are proposed to address the same score problem. With our syntactic and semantic features, our system achieves a Top-1 MARR of 74.11% and Top-5 MARR of 76.68%. Compared to the baseline system, Top-1 MARR and Top-5 MARR increase by 16.17% and 18.61% respectively.