Conversational Graph-Based Navigation Over Semantically Connected Content (COGNOSCO)

Motivation

The Internet as we know it has led to revolutionary changes in the world and our everyday life. We now have access to billions of resources, which can be discovered via search engines. As data volumes and complexity grow, it becomes increasingly difficult to extract valuable information to derive knowledge from data repositories. Metaphorically speaking, one has to first navigate through an ocean of information to find something with true relevance.

Computer scientists have been consistently developing new techniques to address this problem. Modern semantic search engines go beyond simple question answering or keyword matching by understanding the intent and contextual meaning of user queries. The emerging paradigm of conversational search goes even a step further. It promises to satisfy information needs using human-like information-providing dialogs. Through multi-turn interactions with conversational agents, users can resolve ambiguities, narrow down the relevant search space, and extract novel insights.

Research Focus and Goals

COGNOSCO (from Latin cognōscere: to discover/learn/know), a research project in collaboration with Springer Nature, aims at investigating conversational search systems that enable explorative navigation in scientific knowledge bases. Digital publication repositories provide access to millions of scientific documents from conferences, journals, workshops, or books. Various scenarios how scientists discover and learn new insights from these resources are considered. In this context, conversational search systems can interact with humans naturally and offer assistance. With the help of knowledge graphs, structured as well as unstructured textual data are semantically linked during the information search and presented in an understandable format. The search process itself is controlled by an interactive text-based dialog optimized with regard to the user's information goal. In addition to the theoretical contributions of the project, a prototype system is developed and evaluated in real-world settings.

Necessary steps for this objective are expressed through the research questions below:

What are suitable conversational interaction mechanisms for exploration and navigation within scientific knowledge bases?
What is a reasonable architecture for such a conversational search system considering state-of-the-art technologies?
Which criteria are expedient from a design and technical point of view for the development of the conversational interface?
How does the underlying semantic data model have to look like to allow dynamic linking of different data sources during the search process?
Which evaluation methods can be used to determine the utility of the developed system for individual users and organizations?