I'm new to nlp and I'm working on an ambitious project which consists in analyzing the "role" of a set of chemicals in patents.
Task: Classify a set of chemical compounds as "inputs", "outputs" or "none" of patents
Data: Patent data with full natural language description
Tools: python
Idea of approach: Compute a set of features of chemicals and then classify chemicals as inputs or outputs with a Decision Tree Classifier
Problem: Computing the features to predict the role of the chemicals. I thought of using spaCy and create two new named entities but the same chemical can have different functions in different patents (maybe a problem?). I also thought of retrieving the sentences of the patents that contain chemicals and create a naive "profile" of the chemical measuring its position in the sentence, text and relation to recurrent words in the special sentences.