11 July 2019
Everytime we think Natural language processing use-cases gotexhausted, there comes another one. Recently, a group of researchers ( Vahe Tshitoyan, John Dagdelen, Leigh Weston, Alexander Dunn, Ziqin Rong, Olga Kononova, Kristin A. Persson, Gerbrand Ceder & Anubhav Jain came up with NLP use-case where they scan the set of scientific literature ( of material science) and try to find out materials which can be good candidate of new materials, fro e.g., let’s say electrochemicals.
In the past, we’ve seen how to develop a skip-gram model using Word2Vec. Here, the target words, which are basically materials are converted into vectors and cosine similarity is found between these vectors. These Vectors can be mapped to various context word like cathodes, electrochemical, etc.
Based on cosine similarities, researchers were able to find some interesting correlation. Those materials which appeared with word ‘thermoelectrics’ but the researcher didn’t conclude the materials in abstract can be found using this method. It is now possible to predict candidate materials which need to be further investigated.
You can refer this paper to understand further.