Explicit vs. Implicit Biographies: Evaluating and Adapting LLM Information Extraction on Wikidata-Derived Texts

2025-09-18

AI Summary

This work addresses the challenge of implicit semantic reasoning in information extraction (IE). We systematically investigate the performance disparity of large language models (LLMs) on explicit versus implicit biographical texts. To this end, we introduce the first 10k-scale multimodal biographical dataset with explicit/implicit relation annotations, revealing substantial performance degradation—up to 30% F1 drop—across mainstream LLMs (LLaMA-2.3, DeepSeek-V1, Phi-1.5) on implicit relation identification. To bridge this gap, we propose a lightweight low-rank adaptation (LoRA)-based fine-tuning framework specifically designed for modeling implicit relational semantics, enhancing both generalization and interpretability. Extensive experiments demonstrate that LoRA fine-tuning yields an average 12.7% F1 improvement over base LLMs and significantly boosts zero-shot inference capability on unseen implicit patterns. Our results empirically validate the effectiveness and scalability of parameter-efficient adaptation for implicit semantic understanding in IE.

Technology Category

• Natural Language Processing: Information Extraction
• Machine Learning: Large Multimodal Models (LMMs)
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Text Implicitness has always been challenging in Natural Language Processing (NLP), with traditional methods relying on explicit statements to identify entities and their relationships. From the sentence "Zuhdi attends church every Sunday", the relationship between Zuhdi and Christianity is evident for a human reader, but it presents a challenge when it must be inferred automatically. Large language models (LLMs) have proven effective in NLP downstream tasks such as text comprehension and information extraction (IE). This study examines how textual implicitness affects IE tasks in pre-trained LLMs: LLaMA 2.3, DeepSeekV1, and Phi1.5. We generate two synthetic datasets of 10k implicit and explicit verbalization of biographic information to measure the impact on LLM performance and analyze whether fine-tuning implicit data improves their ability to generalize in implicit reasoning tasks. This research presents an experiment on the internal reasoning processes of LLMs in IE, particularly in dealing with implicit and explicit contexts. The results demonstrate that fine-tuning LLM models with LoRA (low-rank adaptation) improves their performance in extracting information from implicit texts, contributing to better model interpretability and reliability.

Problem

• Evaluating LLM performance on implicit vs explicit information extraction
• Analyzing impact of fine-tuning on implicit reasoning generalization
• Improving model interpretability in handling textual implicitness challenges

Innovation

• Fine-tuning LLMs with LoRA adaptation
• Synthetic datasets for implicit/explicit verbalization
• Improving information extraction from implicit texts