Conditional Random Field for Crime News Information Extraction with Enhancement of SMOTE

Crime Information Extraction is a task to extract some entities from crime incident. Getting information from digital crime news is a task that can help people to know what happen from crime-related information. This task is one of the Natural Language Processing tasks that help people to analyze crime information news documents. The purpose of this task is converting the unstructured data from crime news article to structured data text form. This task can be done with rule based or machine learning algorithm. In this research, we used Conditional Random Field and SMOTE that enhancement with some addition rules. Because the unbalanced data from CRF labeling so we need to handle the lack of data from some of categorize. We conduct training to extract 5W1H from 3 digital crime news which consists of 1500 news. We extract news into 13 classes and build a model based on the training data. The test is carried out with real data which is crawled automatically and produces an accuracy of 77.52% for the SMOTE model and 75.61% for the NONSMOTE model. The SMOTE development was successfully used to better extract crime information for data that was not in the training data.