The Combination of Temporal Knowledge Graph and Retrieval-based Generation for Technology Supply and Demand Texts Matching

Matching supply and demand texts is a key challenge for online technology trading platforms, so we've designed a more efficient method to improve performance. The current best method uses the RAG framework for short text expansion and matching, but it struggles with complex technical terms due to limits in embedding similarity-based retrieval. To fix this, we introduced the TKG-RAG framework, which uses TKG-based retrieval instead. We first build TKG with peer technical knowledge, adding a time dimension. Then, we retrieve more relevant and timely content from TKG by considering connected neighbors and time constraints. Finally, we use an LLM to generate expanded text from the technical demand text and retrieved content. Our framework has proven to outperform previous methods with real-world data.