We propose a system to make user’s ideas more diverse as a breakthrough during creating ideas. Discussions such as brainstorming are held to stimulate user’s idea, though such composition requires much effort. For solving the problem, the proposed system presents cross-category articles related to the user’s idea as the divergence support. When the users choose one of them and the keywords shown in the articles, the sentence prototype are generated with the keywords. Afterward, the sentence prototype or its revision is input as a query of searching for the analogy-recalling articles. By iterating the processes, the user’s idea goes diversely. For obtaining the cross-category articles related to the user’s idea, the article texts and user’s idea text is input to GPT2 models to convert to vectors. The category-dependent components extracted with correlation of the category codes of the article is subtracted from the vectors to generate the category independent vectors. The cross-category but similar articles are searched based on the category independent vectors. And also, the experimental evaluation results are shown in this paper.

Molecular fingerprints are widely used for predicting chemical properties, and selecting appropriate fingerprints is important. We generate new fingerprints based on the assumption that a performance of prediction using a more effective fingerprint is better. We generate effective interaction fingerprints that are the product of multiple base fingerprints. It is difficult to evaluate all combinations of interaction fingerprints because of computational limitations. Against this problem, we transform a problem of searching more effective interaction fingerprints into a quadratic unconstrained binary optimization problem. In this study, we found effective interaction fingerprints using QM9 dataset.