Context Graph

TL;DR

This work identifies the limitations of triple-based knowledge graphs in capturing contextual nuances and proposes Context Graphs (CGs) that incorporate time, location, and provenance. It introduces CGR^3, a paradigm that leverages large language models to retrieve contextual evidence, rank candidate entities, and perform context-aware reasoning for KG completion and KGQA. Empirical results on FB15k237 and YAGO3-10 show consistent gains across embedding models, with notable improvements for long-tail entities and a strong boost in KGQA benchmarks using Wikidata/Wikipedia grounding. The findings underscore the importance of context in knowledge representation and reasoning, and demonstrate the practical potential of integrating CGs with LLM-driven reasoning for more robust AI systems.

Abstract

Knowledge Graphs (KGs) are foundational structures in many AI applications, representing entities and their interrelations through triples. However, triple-based KGs lack the contextual information of relational knowledge, like temporal dynamics and provenance details, which are crucial for comprehensive knowledge representation and effective reasoning. Instead, \textbf{Context Graphs} (CGs) expand upon the conventional structure by incorporating additional information such as time validity, geographic location, and source provenance. This integration provides a more nuanced and accurate understanding of knowledge, enabling KGs to offer richer insights and support more sophisticated reasoning processes. In this work, we first discuss the inherent limitations of triple-based KGs and introduce the concept of CGs, highlighting their advantages in knowledge representation and reasoning. We then present a context graph reasoning \textbf{CGR$^3$} paradigm that leverages large language models (LLMs) to retrieve candidate entities and related contexts, rank them based on the retrieved information, and reason whether sufficient information has been obtained to answer a query. Our experimental results demonstrate that CGR$^3$ significantly improves performance on KG completion (KGC) and KG question answering (KGQA) tasks, validating the effectiveness of incorporating contextual information on KG representation and reasoning.

Figures (6)

• Figure 1: Examples of limitations of triple-based KGs. (a) gives an example that the loss of contextual information during KG construction processes may result in the extraction of contradictory triples; (b) gives an example that triple-based representation struggle to represent two facts that involve the same entities and relations but occur in different contexts; (c) gives an example that triple-based KG reasoning methods often learn rule patterns that frequently occur in KGs, but they tend to ignore contexts that may affect the validity of these rules; (d) gives an example that triple-based KG reasoning methods face difficulties in answering questions that involve relational knowledge or contextual information beyond the scope of the triples in KGs.
• Figure 2: An example of factual triples with entity and relation contexts
• Figure 3: The pipeline of the CGR$^3$ paradigm.
• Figure 4: Knowledge Graph Completion.
• Figure 5: Knowledge Base Question Answering.