2026-10-14T18:09:03Z https://www.scidb.cn/oai

10.57760/sciencedb.j00133.000ld 2026-09-20T09:57:23Z

2026-09-20 Tech Novelty-Bench: A Benchmark Dataset for Simulated Scientific Novelty Search doi:10.57760/sciencedb.j00133.000ld en This dataset is the supporting data for the paper "Technology Novelty Bench Technology Novelty Search Intelligent Agent System Integrating Semantic Fusion and Contrastive Thinking Chain: Using Multimodal Private Domain Knowledge Base as a Scenario", which is the Tech Novelty Bench technology novelty search simulation evaluation benchmark constructed by the author team. Contains two subsets: (1) Novelty discrimination subset, with 500 pairs of samples covering electronic information, mechanical manufacturing, and biomedicine, each containing core technical features of novelty search points, comparative literature public numbers and key fragments, technical feature relationships (separated/intersecting), expert annotations (novel/partially novel), and fine-grained difference explanations; (2) Illusion inducement subset, consisting of 200 cross domain or fictional search points, used to evaluate the system's ability to intercept circuit breakers. The data is sourced from the China Patent Database (CNKI) and has undergone entity level desensitization processing. It can be used for evaluation and reproduction in areas such as technology novelty search, novelty discrimination, retrieval enhancement generation, and large language model illusion suppression.  Scientific Novelty Search → Evaluation Benchmark → Novelty Discrimination → Retrieval-Augmented Generation → Semantic Circuit Breaker; Tech Novelty Bench; technology novelty search; benchmark dataset Sun Tianzhen Li Dan Feng Yihan PUBLIC https://creativecommons.org/licenses/by/4.0/ dataset Science Data Bank