The field of information retrieval is moving towards a more nuanced understanding of complex, multi-faceted concepts, with a focus on entity-aware and semantic matching approaches. Recent developments have highlighted the importance of modeling relationships between entities and concepts, and of incorporating structural information from knowledge graphs to enhance retrieval performance. Noteworthy papers in this area include: PairSem, which proposes a framework for pairwise semantic matching that captures complex scientific concepts. REGENT, which introduces a neural re-ranking model that uses entities as a semantic skeleton to guide attention. QDER, which unifies entity-oriented and multi-vector approaches to neural IR by integrating knowledge graph semantics into a multi-vector model.
Advances in Entity-Aware Information Retrieval
Sources
PairSem: LLM-Guided Pairwise Semantic Matching for Scientific Document Retrieval
Integrating Structure-Aware Attention and Knowledge Graphs in Explainable Recommendation Systems
LLMAtKGE: Large Language Models as Explainable Attackers against Knowledge Graph Embeddings
REGENT: Relevance-Guided Attention for Entity-Aware Multi-Vector Neural Re-Ranking
QDER: Query-Specific Document and Entity Representations for Multi-Vector Document Re-Ranking
Embedding-Based Context-Aware Reranker