Modeling Context in Answer Sentence Selection Systems on a Latency Budget
This work addresses the efficiency-accuracy trade-off in open-domain Question Answering systems for users needing low-latency responses, representing an incremental advancement.
The paper tackled the problem of incorporating contextual information into Answer Sentence Selection (AS2) models to improve accuracy without significantly increasing latency, achieving a 6% to 11% improvement over noncontextual state-of-the-art models.
Answer Sentence Selection (AS2) is an efficient approach for the design of open-domain Question Answering (QA) systems. In order to achieve low latency, traditional AS2 models score question-answer pairs individually, ignoring any information from the document each potential answer was extracted from. In contrast, more computationally expensive models designed for machine reading comprehension tasks typically receive one or more passages as input, which often results in better accuracy. In this work, we present an approach to efficiently incorporate contextual information in AS2 models. For each answer candidate, we first use unsupervised similarity techniques to extract relevant sentences from its source document, which we then feed into an efficient transformer architecture fine-tuned for AS2. Our best approach, which leverages a multi-way attention architecture to efficiently encode context, improves 6% to 11% over noncontextual state of the art in AS2 with minimal impact on system latency. All experiments in this work were conducted in English.