eduankotze / bitm

Natural Language Processing

Business Intelligence and Text Mining Research Group

This is an online repository for articles and resources related to the UFS-BITM Research Group.

The research focus is on Digital Text Analysis where we combine natural language processing (practical applications on textual data), data science (data analysis and practical data visualization from large document collections) as well as machine and deep learning.

There are currently 647 state-of-the-art tasks (or topics of interest) for Natural Language Processing research.

As a research group, applications are mainly centered around the following few topics of interest:

• text classification / document classification / topic modeling
• hate speech detection / abusive language detection
• few-shot text classification / few-shot learning / zero-shot classification
• large language models (i.e. Mistral, LLama2, ChatGPT)
• question answering
• chatbot (or conversational AI) [PhD only]

Research Team

Principal Investigators

• Prof Eduan Kotzé

Research Fellows

• Dr Burgert Senekal
• Dr Femi Oriola

Postdoctoral students

• Dr Adewuyi Adegbite (2024 - 2025) - Generative AI Authorship Verification / Detection
• Dr Aliyu Olubunmi (2021- 2022) - Text Generation for Conversational Agents

Postgraduate students

• Jaco Marais (PhD student) - Near Real-Time Monitoring of Civil Unrest in South Africa utilising Bilingual Data Sources
• Gavin Dollman (PhD student) - Exploring Predictive Machine and Deep Learning Models for Prospecting Fossil Sites within the Elliot Formation of South Africa
• Janet Agunbiade (PhD student) - Topic still to be decided
• Fezile Mfengwana (MSc student) - Neural Machine Translation for Low-Resourced South African Language isiXhosa to English: Using Transformer-based Large Language Models
• Zama Dhladhla (MSc student) - Predicting Meteorological Drought Categories Using Graph Neural Networks
• Molise Makafane (MSc student) - Topic still to be decided
• Nhlanhla Baloyi (MSc student) - Topic still to be decided
• Ernest Mohlabane (MSc student) - Topic still to be decided
• Takudzwa Machida (MSc student) - Topic still to be decided

How is Business Intelligence related to Text Mining?

Business intelligence (BI) and text mining are two distinct but interconnected fields that can complement each other in various ways. Let's explore their relationship with some examples:

Data Extraction and Analysis:

• BI involves the collection, processing, and analysis of structured data, typically from databases, to provide insights into business performance.
• Text mining, on the other hand, deals with unstructured textual data, extracting valuable information from sources like emails, social media, and documents.

Integration of Unstructured Data:

• Text mining can contribute to BI by processing and analyzing unstructured data, converting it into a structured format that BI tools can use. This integration allows organizations to gain insights from both structured and unstructured data sources.

Predictive Analytics:

• Text mining can be used to extract features and patterns from textual data that can contribute to predictive analytics. By combining this with traditional BI data, organizations can make more accurate predictions about future trends and events.

Is Text Mining and Natural Language Processing the same?

Text mining (or text analytics) and natural language processing (NLP) are related fields that deal with processing and analyzing textual data but they have distinct focuses and goals. Text mining is more oriented towards discovering patterns and insights from text, while NLP is concerned with a deeper understanding of language to enable machines to interact with it more intelligently. For example, NLP is rooted in computational linguistics and involves tasks such as text categorization, language understanding, translation, and language generation. These tasks utilize either rule-based or machine/deep learning systems which is a subfield of AI. Text mining applications “extract" information and insight from text using AI and NLP techniques. These techniques turn unstructured data into structured data to make it easier for data scientists and analysts to do their jobs. In short, you can have NLP without text mining, but it would be difficult to do text mining without NLP. In our research group, we use both fields.

Why our interest in Text Mining and Natural Language Processing?

We believe that these related fields are the most important technologies of the information age. Applications of both are everywhere because people communicate almost everything in language (mostly text): web search, advertising, emails, customer service, language translation, virtual agents, medical reports, and now even ChatGPT. It is important to learn the necessary skills to perform natural language processing and text mining.

Table of Contents

Modules

• CSIA7915 Natural Language Processing

Books

• AI in and for Africa: A Humanist Perspective (2023)
• Reinventing the Social Scientist and Humanist in the Era of Big Data (2019)

Articles

• 2024

  • ACDSA 2024 - Afrikaans Literacy Genre Recognition using Embeddings and Pre-Trained Multilingual Language Models
  • ACDSA 2024 - The Automatic Detection of Abusiv Language in Dota 2 Chat Messages
  • TALIP - Improving Detection of Multilingual South African Abusive Language using Skip-gram and Joint Multilevel Domain Adaptation
• 2023

  • Stilnet - Pyplyn vir die versameling, ontsluiting en beskikbaarstelling van materiaal vir die Nasionale Afrikaanse Letterkundige Museum en -Navorsingsentrum (NALN) se knipselversameling
  • CCIS - A Feature Selection Method Based on Rough Set Attribute Reduction and Classical Filter-Based Feature Selection for Categorical Data Classification
  • JSASA - Machine Learning for Document Classification in an Archive of the National Afrikaans Literacy Museum and Research Centre
• 2022

  • ICCSA 2022 - Exploring Neural Embeddings and Transformers for Isolation of Offensive and Hate Speech in South African Social Media Space
  • IST-Africa 2022 - Stacked Language Models for an Optimized Next Word Generation
• 2021

  • Nomina-Africana - Names and ethnic heritage: a study of ethnic diversity in the company director network on the Johannesburg Stock Exchange
• 2020

  • TD-SA - Not just a language with white faces: Analysing #taalmonument on Instagram using machine learning
  • SACJ - Improved semi-supervised learning technique for automatic detection of South African abusive language on Twitter
  • SAJS - Automatic classification of social media reports on violent incidents in South Africa using machine learning
  • Prasa/RobTech - Exploring the Classification of Security Events using Sparse and Dense Representation of Text
  • IEEE Access - Evaluating Machine Learning Techniques for Detecting Offensive and Hate Speech in South African Tweets
• 2019

  • FAIR - Automatic detection of abusive South African tweets using a semi-supervised learning approach
  • ISCMI - Automatic Detection of Toxic South African Tweets Using Support Vector Machines with N-Gram Features
  • African Security Review - Open source intelligence (OSINT) for conflict monitoring in contemporary South Africa - Challenges and opportunities in a big data context
  • SATNT - Open Source Intelligence (OSINT) for security purposes: Developing a data analysis pipeline to analyse relevant WhatsApp messages {Afrikaans paper}
• 2018

  • TD-SA - Employing sentiment analysis for gauging perceptions of minorities in multicultural societies: An analysis of Twitter feeds on the Afrikaner community of Orania in South Africa
  • CLiPS - Multilingual Cross-domain Perspectives on Online Hate Speech
  • JNGS - Design and Evaluation of an Artefact for Real-Time Twitter Sentiment Analysis

Completed Theses and Dissertations

• Ian van der Linde (2020) - A comparison of sentiment analysis techniques in a parallel and distributed NoSQL environment