كاميار درويشي

Due to the increasing use of Internet and growing expansion in generated data, the necessity of systems to comprehend and extract information from raw data becomes perceptible. Additionally, available applications for Information Extraction (IE), Natural Language Understanding (NLU) and Information Retrieva‎l (IR), mostly require to percept the relation between entities. Many approaches are used to extract information from text, which one of them is Relation Extraction. Relation Extraction between entites is a significant step for knowledge extraction. Also, Relation Extraction is a crucial phase to improve the performance of Question Answering (QA), intelligent query and Recommender systems. Main purpose of this thesis is to propose a supervised method to detect the types of relation among entites based on context in Farsi. For this reason, BERT which is a transfer learning model is implemented on Perlex dataset. In this thesis, an approach is proposed to improve the performance of transformer learning model by considering its structure and also indirect feeding of data to proposed model which outperforms former methods available for relation extraction.

استخراج رابطه , دسته بندي روابط , مدل انتقال يادگيري , پردازش زبان هاي طبيعي

relation extraction , Relation classification , transfer learning , Natural Language Processing

Kamyar Darvishi

Dr. Minaei