An ontological framework for information extraction using fuzzy rule-based system and word sense disambiguation

• Main Author: Zaman, Gohar
• Format: Thesis
• Language: English; English; English
• Published: 2021
• Subjects: T Technology (General)
• Online Access: http://eprints.uthm.edu.my/8418/1/24p%20GOHAR%20ZAMAN.pdf; http://eprints.uthm.edu.my/8418/2/GOHAR%20ZAMAN%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/8418/3/GOHAR%20ZAMAN%20WATERMARK.pdf; http://eprints.uthm.edu.my/8418/

Automatic information extraction (IE) from online published scientific resources (mainly semi-structured and unstructured) like articles, proceedings, editorials etc. is among the hottest areas of research in text mining. This information is essential for various reasons like tagging, searching, indexing the documents and search engine optimization. In this regard, various techniques possessing considerable accuracy besides other merits, have been proposed in the literature. However, their efficiency is limited to domain-specific documents with static and well-defined formats. Whereas the accuracy is significantly challenged with a slight modification in the document format. Hence, it can be safely stated that so far, no scheme is robust enough for broader types, domains, and formats of documents from diverse publishing societies. To address this issue, an Ontological Framework for IE (OFIE) using a fuzzy rule-based system (FRBS) and an efficient word sense disambiguation (WSD) technique is proposed in this research. The FRBS module is responsible for IE in a precise manner by incorporating fuzzy regular expressions with an added tolerance factor conceived experimentally. FRBS is applied to XML and text converted versions of the same input document to extract two streams. Afterwards, the WSD module synthesizes both streams and yields the outcome that is promising semantically as well as syntactically. The domain is significantly wide-ranging and comprises of articles from well-known publishing services like IEEE, ACM, Elsevier, Springer, and few others. It is observed from extensive experiments and contrasting with state-of-the-art techniques that the proposed scheme is robust to changes in format, extracts better information, and exhibits a significant precision, recall and F-score as 89.14%, 89.6% and 89%, respectively in testing phase. As an outcome, the extracted information can be stored in a digital library for the sake of archiving and retrieval by means of extract, transform and load (ETL) process.