KNOWLEDGE GRAPH CONSTRUCTION FROM SITUATED MULTIMODAL DIALOGUE
Abstract
Computationally mediated dialogue (e.g. chat) has emerged as a fundamental mechanism for team performance in terms of coordination, planning and action. Being able to extract information from both dialogue and its surrounding context (e.g. shared documents, organizational embedding) can enable new forms of analytics and help power intelligent interventions. Current research has focused on the development of conversational agents and not on the analytics over dialogue itself. In particular, there is a paucity of research that considers the multimodal nature of a dialogue’s context with respect to dialogue itself.. This project focuses on the extraction of information about dialogue and its context in the form of a knowledge graph – a knowledge base describing entities (e.g. people, organizations) and their relationships to one another. Knowledge graph construction combines the areas of information extraction, knowledge fusion, and graph refinement, thus making it a challenging problem. Specifically, the project aims to define a set of benchmark tasks and associated datasets for extracting knowledge graphs from dialogue along with its context. Additionally, it will develop a state-of-the-art baseline system for the defined tasks. The project, thus, will provide a concrete measurable foundation for research in this area.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2021
- Source ID
- FA86552017005
Entities
People
- Paul Groth
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Amsterdam