Word-level information extraction from science and technology announcements corpus based on CRF

Conditional Random Field (CRF) has been applied widely in information extraction and natural language processing. However, according to corpus types, it has not been made much use of on corpus about science and technology declarations. In this paper, we extract word-level information from amounts of science and technology announcements corpus, and analyze the performance of CRF, comparing with Naïve Bayes as a baseline. According to our experiments, we show that CRF has much high precision except for a few unknown data. Also, Naïve Bayes model is satisfactory in closed domains, but it always makes mistakes when the data belong to a less weighted class.