Reducing Wrong Labels using Conflict Score in Distant Supervision for Relation Extraction in Bangla Language

The research area of information extraction (IE) aims to extract structured information such as types of entities and relations between them, from unstructured textual data like newswires, blogs, governmental documents etc. Relation extraction (RE) deals with the automatic detection of relationships between concepts mentioned in free texts. Knowledge-based distant supervision (DS) uses structured data to heuristically label a training corpus. However, this heuristic can generate some noisy labeled data. In this paper, we propose a method using conflict score in DS to reduce the number of wrong labels for Bangla sentences.