Extracting Problem Linkages to Improve Knowledge Exchange between Science and Technology Domains using an Attention-based Language Model

Abstract

Science and technology activities can be considered problem-solving activities, and scientific papers and patent publications can be viewed as providing explicit knowledge gained from the problem-solving of academia and industry respectively. However, even in the same field, the approach to the same problem is not consistent between a paper and the patented technology. The creation of information silos in science and technology generates inefficiency in human intellectual production. Therefore, this study examines whether insights from technical problems can be shared with academics to solve scientific problems. We propose a concept to link the problems between these two domains using a linguistic approach for knowledge discovery that connects science and technology. We extracted scientific papers from the Association for Computational Linguistics dataset, and patent literature from the Derwent Innovation platform. From these, pairs of problem defining sentences were identified and extracted using an attention-based language model. For example, we were able to extract examples of issues that do not necessarily arise from scientific papers, such as annotation difficulties in the analysis of social network data, but can be hinted at by patented techniques prior to the paper. These results suggest that scientific problems and industrial solutions can provide mutual insight. This knowledge discovery approach is recommended not only for benefiting corporate activities but also for grasping research trends.

Keywords

• problem extraction
• attention-based language model
• information matching
• scientometrics
• literature-based discovery (lbd)