CCASL: Counterexamples to Comparative Analysis of Scientific Literature - Application to Polymers

The exponential growth of scientific publications has made the exploration and comparative analysis of scientific literature increasingly complex and difficult.For instance, eliciting two scientific publications that diverge on widely accepted concepts within their domain turns out to be more and more difficult despite its great interest.We are interested in the automatic detection of these discrepancies using the latest artificial intelligence (AI) techniques. Given a particular scientific domain, we focus on large-scale analysis of the tables present in related scientific publications and propose to capture domain knowledge with arbitrary functions.In this setting, we propose a five-step method, called CCASL: (1) Modeling the domain knowledge with functions expressed as approximate functional dependencies (FDs), (2) Acquiring a corpus of scientific documents related to the proposed functions, (3) Analysing all tables occurring in the PDF documents and producing a consolidated table from them, (4) Detecting counterexamples of the FDs in the consolidated table, and (5) Conducting a comparative analysis of the pairs of papers containing the detected counterexamples. We have applied CCASL to a subfield of polymer research, known as Epoxy-Amine networks (EA). In collaboration with material scientists, we have identified an intuitive function \(f_{EA}\) that relates the storage modulus \((SM)\), the structure of the polymer \((V_{EA})\), and its glass transition temperature \((T_g)\). Based on this function, we have implemented all the 5 steps of CCASL. First results show that CCASL is proving to be a powerful approach for bibliographic confrontation in the field of polymers.