‘ChemXtract’ A System for Extraction of Chemical Events from Patent Documents

Proceedings of Recent Advances in Natural Language Processing, pages 988–995 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_106 988 1 Abstract ChemXtraxt main goal is to extract the chemical events from patent documents. Event extraction requires that we first identify the names of chemical compounds involved in the events. Thus, in this work two extractions are done and they are (a) names of chemical compounds and (b) event that identify the specific involvement of the chemical compounds in a chemical reaction. Extraction of essential elements of a chemical reaction, generally known as Named Entity Recognition (NER), extracts the compounds, condition and yields, their specific role in reaction and assigns a label according to the role it plays within a chemical reaction. Whereas event extraction identifies the chemical event relations between the chemical compounds identified. Here in this work we have used Neural Conditional Random Fields (NCRF), which combines the power of artificial neural network (ANN) and CRFs. Different levels of features that include linguistic, orthographical and lexical clues are used. The results obtained are encouraging. 1 Introduction Chemical information extraction is a challenging task. Unstructured data in the biomedical domain contain descriptions of chemical entities and the extracting these entities from textual data repositories, in particular from the patents, is becoming increasingly important for researchers and for the industry. Human annotation of patents to generate annotated corpus and populate chemical databases is a tedious task and this can be made easy and fast through the use of automated language processing. The process of automatically extracting the mentions of a particular semantic type in text is known as Information Extraction (IE). IE includes the extraction of names of chemical compounds and assigns a label according to the role it plays within the chemical reaction, popularly known as named entity recognition (NER) and also event relation extraction, where it extracts the chemical event relation that takes place between the chemical compounds. ChemXtract extracts the chemical compound names and its event relation in patent documents. In this paper we discuss in detail the methods and techniques used in ChemXtract. The extraction identify and label chemical compounds and their specific types, i.e. to assign the label of a chemical compound according to the role which it plays within a chemical reaction, the temperature and reaction time at which the chemical reaction is carried out, the yields obtained for the final chemical product and the label of the reaction. The challenges in extracting the chemical compounds are many and it further increases when it is from patent documents. The language used in patents is very different from the language used in scientific literature. When writing scientific papers, authors strive to make their words as clear and straightforward as possible, whereas patent authors often seek to protect their knowledge from being fully disclosed [34]. Thus the main challenges for natural language processing (NLP) in patent documents arise from its writing style such as long and complex sentences and long list of chemical compounds. As the characteristics of sentences in …

Content truncated to 500 words