Improving Response Selection in Multi-Turn Dialogue Systems by Incorporating Domain Knowledge

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 497–507 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 497 Improving Response Selection in Multi-turn Dialogue Systems by Incorporating Domain Knowledge Debanjan Chaudhuri Smart Data Analytics Group University of Bonn & Fraunhofer IAIS Germany chaudhur@cs.uni-bonn.de Agustinus Kristiadi Smart Data Analytics Group University of Bonn Germany kristiadi@uni-bonn.de Jens Lehmann Smart Data Analytics Group University of Bonn & Fraunhofer IAIS Germany jens.lehmann@cs.uni-bonn.de Asja Fischer Department of Mathematics Ruhr University Bochum Germany asja.fischer@rub.de Abstract Building systems that can communicate with humans is a core problem in Artificial Intel- ligence. This work proposes a novel neural network architecture for response selection in an end-to-end multi-turn conversational dia- logue setting. The architecture applies context level attention and incorporates additional ex- ternal knowledge provided by descriptions of domain-specific words. It uses a bi-directional Gated Recurrent Unit (GRU) for encoding context and responses and learns to attend over the context words given the latent response representation and vice versa. In addition, it incorporates external domain specific informa- tion using another GRU for encoding the do- main keyword descriptions. This allows better representation of domain-specific keywords in responses and hence improves the overall per- formance. Experimental results show that our model outperforms all other state-of-the-art methods for response selection in multi-turn conversations. 1 Introduction In a conversation scenario, a dialogue system can be applied to the task of freely generating a new re- sponse or to the task of selecting a response from a set of candidate responses based on the previ- ous utterances, i.e. the context of the dialogue. The former is known as generative dialogue sys- tem while the latter is called retrieval-based (or response selection) dialogue system. Both approaches can be realized using a modu- lar architecture, where each module is responsible for a certain task such as natural language under- standing, dialogue state-tracking, natural language Context Utterance 1: My networking card is not working on my Ubuntu, can somebody help me? Utterance 2: What’s your kernel version? Run uname -r or sudo dpkg -l |grep linux-headers |grep ii |awk ’{print $3}’ and paste the output here. Utterance 3: It’s 2.8.0-30-generic. Utterance 4: Your card is not supported in that kernel. You need to upgrade, that’s like decade old kernel! Utterance 5: Ok how do I install the new kernel?? Response Just do sudo apt-get upgrade, that’s it. Table 1: Illustration of a multi-turn conversation with domain specific words (UNIX commands) in italics. generation, etc., or can be trained in an end-to-end manner optimized on a single objective function. Previous work, belonging to the latter category, by Lowe et al. (2015a) applied neural networks to multi-turn response selection in conversations by encoding the utterances in the context as well as the possible responses with a Long Short-term Memory (LSTM) (Hochreiter and Schmidhuber, 1997). Based on the context and response encod- ings, the neural network then estimates the proba- bility for each response to be the correct one …

Content truncated to 500 words