Modeling Multi-turn Conversation with Deep Utterance Aggregation
This addresses the challenge of building more intelligent dialogue systems, particularly for domains like e-commerce, but it is incremental as it builds on existing retrieval-based matching approaches.
The paper tackled the problem of multi-turn conversation understanding for retrieval-based dialogue systems by proposing a deep utterance aggregation model to form fine-grained context representations, resulting in outperforming state-of-the-art methods on three benchmarks.
Multi-turn conversation understanding is a major challenge for building intelligent dialogue systems. This work focuses on retrieval-based response matching for multi-turn conversation whose related work simply concatenates the conversation utterances, ignoring the interactions among previous utterances for context modeling. In this paper, we formulate previous utterances into context using a proposed deep utterance aggregation model to form a fine-grained context representation. In detail, a self-matching attention is first introduced to route the vital information in each utterance. Then the model matches a response with each refined utterance and the final matching score is obtained after attentive turns aggregation. Experimental results show our model outperforms the state-of-the-art methods on three multi-turn conversation benchmarks, including a newly introduced e-commerce dialogue corpus.