Category: analytics and machine learning

Automatic Text Summarization: the plasticity of language

Automatic text summarization (ATS) techniques offer powerful solutions for generating accurate and informative summaries from textual data content. In our digital age, where an estimated 403 million terabytes of data are generated daily, it is vital that we are able to distil large amounts of textual content into focused summaries, containing just the salient details. In this article I provide an updated survey of state-of-the-art ATS methods, with a particular focus on how large language models models address the complexities and nuances of automated text summarization.

Topic Modelling with BERTopic and DataMapPlot

Topic modelling is a form of text analysis that uses unsupervised machine learning to identify patterns, themes, clusters, and groups across a collection of documents. In this article I discuss using the powerful BERTopic library alongside quantized large language models to identify themes and topics from a collection of research papers at the intersection of artificial intelligence and ophthalmology. Then, we'll use the DataMapPlot library to produce a publication ready visualization of the thematic structure contained within the abstracts of those research papers.

Loading data into Google Colab

Google Colab (or Colaboratory) is a complete, modern, cloud-based runtime environment. It gives individuals and teams the ability to work together on coding, data science and machine learning problems with shared access to data, state of the art GPU's and TPU's, and industry standard Python libraries. You are provided with an executable document, i.e. a notebook much like a Jupyter Notebook, that allows both code and markdown to be written and executed, and your results visualized. In this article I'll cover one of the most fundamental requirements all teams will face - how to get data into your Colab notebook in the first place!

Rapid Data Visualization and Interactivity with Streamlit

Streamlit is an open-source Python app framework that allows you to rapidly turn your data into an interactive data visualization - all in pure Python code. The apps you create are custom, fast, beautiful and very easy to share. What's more is that Streamlit seamlessly integrates with existing popular Python data visualization tools such as Matplotlib and Plotly Express. In this article we'll take some Twitter data and explore it in the context of building a Streamlit interactive data visualization.

Machine Learning for the Detection and Prevention of Credit Card Fraud

Financial institutions (including banks, insurance brokers, investment companies, etc.) are obliged to guarantee the highest level of security and protection against (an increasing) range of methods criminals use to commit financial crime. The UK Payment Services Regulations, for debit cards, state that the most you should have to pay is the first £35 of an unauthorised transaction if the bank has reason to believe you should've been aware that your payment details were lost or stolen. Your bank can only refuse to refund you if it has evidence that you acted fraudulently, or with 'gross negligence'. If a fraud occurs on your credit card or a credit facility, the UK Consumer Credit Act takes precedence. Thus, any accusation of 'gross negligence' on the part of the consumer doesn't arise in this act unless your card provider can demonstrate that you authorised the payment. Consequently it stands to reason that banks and financial institutions are especially interested in preventing bank fraud! In this article I'll look at the role machine learning can play in preventing credit card frauds - while also considering some of the specific challenges this domain presents.