Abstract

Modern language models use knowledge bases to provide fact-based and up-to-date information. The quality of knowledge-preprocessing and semantic enrichment is therefore decisive for system performance. This project deals with documents and knowledge repositories that regularly evolve and require systematic update mechanisms. Application domains range from law and medicine to journalism and scientific literature.

The goal is to develop a robust and transparent pipeline that filters, validates, and integrates new information into existing knowledge bases while maintaining consistency and temporal contextualization. The research focuses on three core areas: 

(1) efficient modeling of validity periods, versioning, and temporal dependencies between knowledge entities, 

(2) extraction of temporal and semantic information from documents to enable automated knowledge base updates, and 

(3) evaluation of the advantages and challenges of temporal knowledge bases in dialogue systems compared to conventional approaches. 

By combining temporal information extraction, semantic relationship modeling, and automated update mechanisms, CHRONO-FACT aims to develop knowledge management systems that can process dynamically evolving information while ensuring factual accuracy and traceability.

Research Questions

RQ1: How can validity periods, versioning, and temporal dependencies between knowledge entities be modeled efficiently?

RQ2: To what extent can temporal and semantic knowledge be extracted from documents to filter, verify, and integrate new information into an existing knowledge base?

RQ3: What advantages and challenges arise from using temporal knowledge bases in dialogue systems compared to conventional approaches?