Constructing knowledge graph from linux kernel commit message

• Main Author: Pentium, Gede Bagus Bayu
• Other Authors: Liu Yang
• Format: Final Year Project
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Computer science and engineering
• Online Access: http://hdl.handle.net/10356/74057
• Institution: Nanyang Technological University

With a large amount of data available, a lot of security-related information can be extracted from the data. The main problem is a large portion of them (80%-90%) are stored in an unstructured manner. One of the well-known forms of unstructured data is in the form of text. Textual data can contain much information with using a small amount of space. But textual data are mainly stored in human language, with this machine are having a hard time to extract information. Many natural language processing is done to extract information from the text. When extracting information data representation is playing a huge role. One of the most popular data representation from textual data is knowledge graph. Constructing knowledge graph from unstructured textual data can help the machine to understand the information contained in the data. This project is aimed to extract knowledge graph from Linux Kernel commit message. With consists of more than 700,000 commit message, this is a huge amount of data to be processed. If the information is successfully extracted, the information contained will benefit a lot in computer security. The knowledge graph extraction consists of four processes. They are data cleaning, entity extraction, relation extraction, and knowledge graph construction. Entity extraction is a process to recognize named entities from the text into pre-defined categories. For entity extraction, a combination of automated labeling and machine learning (CRF classifier) are used. Relation extraction is a process to detect and classify semantic relationship between the pre-extracted entities in text. For relation extraction, both schema-based and schema-free relation is extracted. After the extraction, 1,247,864 entities and 1,747,009 relations are extracted. With a convincing result of 74.29% F-measure score, the knowledge extraction is considered to be performing well under given circumstances.