How to measure the semantic similarities between scientific papers and patents in order to discover uncommercialized research fronts: A case study of solar cells

In this paper, we perform a comparative study to measure the semantic similarity between academic papers and patents. Research fronts which do not correspond any patents can be uncommercialized and opportunities for industry. Therefore it is significant to investigate the relationship between the scientific outcomes and the pieces of industrial technology. We compare structures of citation network of scientific publications with those of patents by citation analysis, measure the similarity between sets of academic papers and ones of patents by natural language processing, and discuss the validity of the results with experts. After the documents (papers/patents) in each layer are categorized by a citation-based method, we compare three semantic similarity measurements between a set of academic papers and a set of patents: Jaccard coefficient, cosine similarity of tfidf vector, and cosine similarity of log-tfidf vector. A case study is performed in solar cells to develop a method investigating the corresponding relationship between papers and patents. As a result, the cosine similarity of tfidf is the best way to discover the corresponding relationship. This proposed approach enables us to obtain, at least, the candidates of unexplored research fronts, where academic researches exist but patents do not.