Interpretation of Implicit Conditions in Database Search Dialogues

Abstract

This study focuses on database (DB) search dialogue and proposes to employ user utterance information that does not directly mention the DB field of the back-end system but is useful for constructing DB queries. We name this type of information implicit conditions, the interpretation of which enables the dialogue system to be more natural and efficient in communication with humans. We formalise the interpretation of implicit conditions as the classification of user utterances into the related DB field while simultaneously identifying the evidence for such classification. Introducing this new task is one of the contributions of this paper. We implemented three models for this task: an SVM-based model, an RCNN-based model and a sequence-to-sequence model with an attention mechanism. In an evaluation via a corpus of simulated dialogues between a real estate agent and a customer, the sequence-to-sequence model outperformed than the other models.