Bachelor's Thesis Nils Rehtanz

Using Knowledge Graphs to Improve News Search and Exploration with Voice-Based Conversational Agents

Abstract

Knowledge graph driven conversational agents in general as well as knowledge graphs in particular for news article related tasks have been studied in the past separately. However, their combination, i.e., knowledge graph driven conversational agents for news exploration, seems to be an underinvestigated area. This thesis addresses how to create a knowledge graph for a voice-based conversational agent for news exploration and how to implement the agent to use this knowledge graph for news search and recommendation. For this purpose, the current state-of-the-art in voice-based news search and exploration is analyzed, and the essential potentials for improvement are identified. Furthermore, interaction patterns for voice-based German news search and exploration are evaluated, serving as a basis for the knowledge graph and the conversational agent. On this foundation, the knowledge graph is conceptualized and constructed, and the conversational agent is developed. User tests have been carried out to gain insights into how to further improve the system in future work.

Research Questions

1. What is the current state-of-the-art in voice-based news search and exploration?
2. What are suitable interaction patterns for voice-based German news search and exploration?
3. How to construct a German news knowledge graph as the database for a conversational agent?
4. How to build a voice-based conversational agent for news search and exploration with the knowledge graph?
5. Which insights can be gained from user tests for improving the conversational agent?