Evaluation Analysis of Large Language Models Based on TKG and Document Vectorisation

The integration of Temporal Knowledge Graphs (TKGs) and document vectorization techniques into the evaluation of large language models (LLMs) presents a significant advancement in natural language processing. This study explores how TKGs, which encapsulate temporal relationships and time-stamped events, can enhance the contextual understanding of LLMs. By converting documents into numerical vectors through methods like TF-IDF, Word2Vec, and BERT embeddings, we aim to improve the semantic coherence and temporal accuracy of these models. Our methodology involves collecting and preprocessing temporal and document data, followed by vectorizing this information and fine-tuning LLMs to incorporate the temporal dynamics and rich contextual embeddings. The results indicate that integrating TKGs and advanced vectorization significantly boosts the temporal and relational understanding of LLMs, leading to better performance across various applications.