The World Wide Web is a virtually inexhaustible source of information for any conceivable knowledge domain. However, pinpointing interesting items is a challenging problem. We deal with this problem from basic definitions, through specific practical issues, up to concrete applications. We first ask: what is interestingness? Our answer is a composition of two functions: Relevance and Surprise. Relevance refers to arbitrarily chosen domains of interest, without any justification besides personal taste: one may be interested in basketball, or in the symphonies of Beethoven, from the Eroica and later. In more concrete terms a domain may be characterized by a set of keywords. Any particular item obtained by Web searching is also characterized by another, possibly smaller set of keywords. Relevance can be calculated by the degree of matching between these two sets. Surprise refers to outstanding items far from the typical average case, while still being relevant. Among metals, conducting electricity and heat, and shining when polished, mercury has the relevant properties, but surprises by being liquid instead of solid at room temperature. Surprise can be calculated by the degree of mismatch between the referred sets, e.g. by a symmetric difference. Concerning chemicals, we searched them by complete or partial structural formulas – not names – and groups of atoms related to potential activity. To be usable as search inputs, 2D and 3D structures were linearized e.g. by SMILES. It turns out that direct search with chemical structures obtains very focused and manageable numbers of results. One may logarithmically control these numbers by varying sizes of input groups of consecutive atoms, even when these are randomly cut from the whole molecule. Tested applications were chemicals with potential pharmacological activity. Web search results ordered by decreasing Interestingness compared with results obtained by standard search engines, have shown item rankings significantly different. Our ranking actually highlighted items of interest. We conclude discussing novel ideas, practical solutions and obtained results for representative case studies.

Systematic generation of surprises: Application to chemicals with activity of interest

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.