Highlights of recent summer scholarships
Our summer scholarship projects have been a great success. Find some of the highlights below.
Automatically Verifying Statements of Fact
Supervisor: Dr Ben Hachey
Natural Language Processing is used to automate the search for verifying facts in legal documents. The “Verify” web application has been created for this purpose. It is to be used by stakeholders in the document verification process who are legally required to sign off on statements.
The Verify system allows the user to quickly verify given facts within a pool of source documents. Such a system can feasibly reduce the time and complexity of searching for verification of facts. It can also be used to easily allow company representatives to sign off on the verifications.
Image credit: Wroclaw University Library digitizing rare archival texts by j_cadmus via CC licence.
Fostering Long-Term Motivation and Engagement: What can Public Health Interventions Learn from Social Media Celebrities?
Supervisors: Prof Margaret Allman-Farinelli, Prof Judy Kay, A/Prof Bob Kummerfeld, Lie Ming Tang
This project looked at mHealth and studied Kayla Itsines, a social media fitness celebrity. Some lessons to take away:
- Create communities where health behaviours are praised by other people
- Attempt to foster people’s identity as a healthy person or a health-conscious person
- Capitalise on already meaningful rewards systems, such as likes on social media platforms
- Encourage people to be mindful of the immediate benefits of heart behaviours
- Present a diverse range of possible benefits from health behaviours, such as energy levels, strength, endurance, mobility, happiness, etc.
- Empower people to use this information to make the best decisions for their personal health according to what makes them feel their best
- Do not focus on weight or appearance
- Do not focus on the contents of food, such as calories. Encourage people to focus on the food itself and how it makes them feel.
How to train your PA: Learning to rank entity news
Supervisor: Dr Ben Hachey
We address the task of creating a personalised brief for a user. We do so by selecting news articles that provide a concise overview of a person’s professional activity. A brief should include three to five documents that are relevant and diverse. We propose and compare features of good articles.
At the touch of a button, get informed and relevant business information which will help users prepare for events and meetings. An essential part of this process is automatically finding several relevant news articles.
We find that random forests are an incredibly effective tool for identifying relevant articles. We can get an excellent prediction of whether or not an article is relevant to a brief. The model incorporates a diverse range of features. This is an exciting and important finding with broad applications.
Individual Differences in Detecting Lies: Standard Deviation and Variance Estimations and Predictions
Supervisors: Dr Simon A Jackson, Dr Sabina Kleitman
People often think that they can detect lies, but can they really? This research examines whether or not people have different capacities for detecting lies as well as whether the ability to deceive others.
Our research supports Bond and DePaulo (2008)’s findings. Even when presented in a language that wasn’t understood, there is little individual differences in lie detection capacities. This means that regular people cannot detect lies well! Much greater individual differences can be found in how well people lie and how trustworthy they appear. This is the direction that future research should be targeting. Research in lie detection is limited by measures with low reliability: Thus, future research should examine better lie detection measures.
Large Interactive Surfaces and Gesture Interaction
Supervisors: Prof Judy Kay, A/Prof Bob Kummerfeld, A/Prof Martin Tomitsch
When people use the gesture-interaction large public display called Media Ribbon, 48% engaged in playful behaviours. This project used a range of methods to automatically detect playful interaction. This is important for improving the recognition of gesture commands. This opens the possibility of creating a surprising fun response to users who play and better gesture recognition for serious users.
Physiological signals: Measurement during long haul flight
Supervisors: A/Prof Corinne Caillaud, Dr Peter Jones, Dr Alistair McEwan
There are three phases for measurement, i.e. pre-flight and post-flight. Heart rate (HR), blood oxygen saturation (Spo2) and activities can be measured via wearable devices. The devices are Onyx Vantage (HR and SpO2), Fit bit (HR and activities) and Gene active (activities).
Data from devices can be synchronised to solve the problems of the differences and frequency discrepancies.
For better accuracy and coherency of data, three devices are used to measure three different physiological signals. That is, Onyx Vantage for SpO2, Fit Bit for heart rate and Gene active for activities.
Rekindle: Usability Study on a system providing Psychosexual support
Supervisors: Dr Haryana Dhillon, Prof Judy Kay
User data analysis and usability highlight strength of Rekindle and potential improvements. Online intervention enable a far broader population of people to access counselling that would not otherwise be available.
Image credit: rekindle
Revealing the Development of the Immune Response through Dynamic Time Series Clustering
Supervisors: Thomas Ashhurst, Prof Nicholas King, Dr Mark Read, A/Prof Uwe Rohem
Flow cytometry is used in pathology as a way to find and identify populations of cells in infected patients. The analysis and clustering of flow cytometry data has been primarily explored through the technique of gating where boundaries are placed over chosen characteristics to observe the relationship of sub populations. However the analysis of flow cytometry data over time intervals has been left relatively unexplored so far. By using an evolutionary clustering technique we observed the changes of the immune response as it reacts to the West Nile virus and developed visual frameworks of these changes.