DPAEG: A Dependency Parse-Based Adversarial Examples Generation Method for Intelligent Q&A Robots

Table 3

Performance of DrQA answering original questions and adversarial questions in terms of , , MRR, and MAP.

Questions	Dataset			MRR	MAP
Original questions	WebQuestionsSP	1	1	1	0.583
	CuratedTREC	1	1	1	0.587
	WikiMovies	1	1	1	0.576
Adversarial questions	WebQuestionsSP	0.208	0.255	0.220	0.094
	CuratedTREC	0.287	0.383	0.282	0.073
	WikiMovies	0.276	0.356	0.269	0.078

Since we only use the original questions that DrQA can answer correctly to generate adversarial questions, , , and MRR scores of DrQA answering original questions are 1.