Chemistry and chemical engineering researchers are currently using computer aided tools to automate tasks such as writing complex chemical reactions [1] and identifying chemical formulae in free form text [2] in order to make the research process more efficient. However, managing the vast amount of information in the primary research literature remains an ever increasing challenge. As an example, whenever a catalyst is being reported in a journal paper, the information about whether it succeeded or not is present in some form in the text. Although this information is both easily readable and understandable by humans, it is very hard for a computer to identify the name of the catalyst or the reaction it was used in. To successfully extract information from unstructured data such as text, we must first train the computer. Statistical algorithms can be trained on rules that chemical formulae follow, such as: the formula must be made up of element symbols present in the periodic table. From any given text, we can extract features such as the presence of element symbols, word endings such as –ane, -ene etc. and use these features for training. In this work, we present the results of automatic identification of chemical entities and classification of text based on the abstracts of articles and show the feasibility of this method.

References:

[1] S-H Hsu, B. Krishnamurthy, P. Rao, C. Zhao, S. Jagannathan, J. Caruthers, V. Venkatasubramanian, “A Systematic Approach For Automated Reaction Network Generation”, 16th European Symposium on Computer Aided Process Engineering and 9th International Symposium on Process Systems Engineering, 2006, 973-978.

[2] Sun, B., Tan, Q., Mitra, P., and Giles, C.L. (2007) “Extraction and Search of Chemical Formulae in Text Documents on the Web,” The 16th International World Wide Web Conference, (WWW'07).