An extraction and novelty evaluation framework for technology knowledge elements of patents

Technology knowledge elements play an important role in technology innovation. However, there is still challenges about their extraction and evaluation. Traditional methods exhibit limitations in precisely linking key technologies with functions, and they usually focus on measuring the overall novelty of patent documents rather than individual technology details, leading to poor interpretability and practicality of research outcomes. In this work, we present a framework that extracts technology knowledge triples and evaluates the novelty of triples based on deep learning model. This framework first identifies key sentences that reflect innovation from patent claims and then extracts technology knowledge elements from these sentences. A novelty index is then designed to evaluate the novelty of these technology knowledge elements based on the probability of their occurrence and the similarity to existing knowledge. The experimental results demonstrate the effectiveness of the proposed method. The extracted technology knowledge elements can use to construct an innovation knowledge graph, which provides practical applications in engineering knowledge retrieval, design and innovation support.