Direct optimization of F-measure for retrieval-based personal question answering

TL;DR

This paper introduces a neural retrieval-based question answering system for personal assistants that directly optimizes F1-score using reinforcement learning to improve retrieval accuracy from user memories.

Contribution

It presents a novel approach to optimize retrieval performance directly with reinforcement learning, enhancing personal assistant capabilities for long-term memory retrieval.

Findings

Improved F1-score on test datasets.

Enhanced retrieval accuracy from user memories.

Effective reinforcement learning optimization.

Abstract

Recent advances in spoken language technologies and the introduction of many customer facing products, have given rise to a wide customer reliance on smart personal assistants for many of their daily tasks. In this paper, we present a system to reduce users' cognitive load by extending personal assistants with long-term personal memory where users can store and retrieve by voice, arbitrary pieces of information. The problem is framed as a neural retrieval based question answering system where answers are selected from previously stored user memories. We propose to directly optimize the end-to-end retrieval performance, measured by the F1-score, using reinforcement learning, leading to better performance on our experimental test set(s).