Dialogue Analysis with Graph Databases: Characterising Domain Items Usage for Movie Recommendations

Nowadays, the use of graph databases combined with textual corpora analysis seems to play a pivotal role in supporting dialogue systems design and implementation. However, dialogues are rarely put in an explicit relationship with the graph structures representing the knowledge domain. In this work, we show how native graph databases provide a framework for a deeper understanding about the use of domain items during dialogue. We describe a multiple-source data collection procedure and we describe how linguistic concepts related to common ground can be found in graph structures. We also describe different patterns that can be detected in the obtained graph structures and discuss their implications in the design of dialogue systems for the movie recommendation task.