Domain-sensitive topic management in a modular conversational agent framework

TL;DRAbstract

Flexible nontask-oriented conversational agents require content for generating responses and mechanisms that serve them for choosing appropriate topics to drive interactions with users. Structured knowledge resources such as ontologies are a useful mechanism to represent conversational topics. In order to develop the topic-management mechanism, we addressed a number of research issues related to the development of the required infrastructure. First, we address the issue of heavy human involvement in the construction of knowledge resources by proposing a four-stage automatic process for building domain-specific ontologies. These ontologies are comprised of a set of subtaxonomies obtained from WordNet, an electronic dictionary that arranges concepts in a hierarchical structure. The roots of these subtaxonomies are obtained from Wikipedia’s article links or wikilinks; this under the hypothesis that wikilinks provide a sense of relatedness from the article consulted to their destinations.

Chat with Paper

AI Agents for this Paper