Dataflow Programming and Acceleration of Computationally-Intensive Algorithms

  • Abdul Qurashi

Student thesis: Doctoral Thesis

Abstract

The volume of unstructured textual information continues to grow due to recent technological advancements. This resulted in an exponential growth of information generated in various formats, including blogs, posts, social networking, and enterprise documents. Numerous Enterprise Architecture (EA) documents are also created daily, such as reports, contracts, agreements, frameworks, architecture requirements, designs, and operational guides. The processing and computation of this massive amount of unstructured information necessitate substantial computing capabilities and the implementation of new techniques.

It is critical to manage this unstructured information through a centralized knowledge management platform. Knowledge management is the process of managing information within an organization. This involves creating, collecting, organizing, and storing information in a way that makes it easily accessible and usable. The research involved the development textual knowledge management system, and two use cases were considered for extracting textual knowledge from documents.

The first case study focused on the safety-critical documents of a railway enterprise. Safety is of paramount importance in the railway industry. There are several EA documents including manuals, operational procedures, and technical guidelines that contain critical information. Digitalization of these documents is essential for analysing vast amounts of textual knowledge that exist in these documents to improve the safety and security of railway operations. A case study was conducted between the University of Huddersfield and the Railway Safety Standard Board (RSSB) to analyse EA safety documents using Natural language processing (NLP). A graphical user interface was developed that includes various document processing features such as semantic search, document mapping, text summarization, and visualization of key trends.

For the second case study, open-source data was utilized, and textual knowledge was extracted. Several features were also developed, including kernel distribution, analysis offkey trends, and sentiment analysis of words (such as unique, positive, and negative) within the documents. Additionally, a heterogeneous framework was designed using CPU/GPU and FPGAs to analyse the computational performance of document mapping.
Date of Award27 Nov 2024
Original languageEnglish
SupervisorAnju Johnson (Main Supervisor) & Colin Venters (Co-Supervisor)

Cite this

'