MEng Computer Science with Distinction, University of Portsmouth
Professional and enthusiast AI Scientist, Maker, and Community Organiser.
Research interests include:
Natural Language Processing
Language Models
Recommender Systems
Automatic Speech Recognition
Fair & Trustworthy AI
Accessibility & Digital Equality
I've delivered several UKRI-funded research projects around AI for accessibility in higher education, worked on projects for the Parliamentary Office for Science and Technology, and published research on Recommender Systems.
In my free time I run and attend hackathons, make open source software, write, hike, and engage with various local communities and meetups (like PF Meet).
All projects listed on this site are entirely my own work except where stated otherwise.
I sometimes publish under the pseudonym David Ralph.
For correspondence, please use Sage Ralph (my real name) so I know you're (probably) not a robot.
All my research publications are open access. Source code and datasets are provided where possible.
Reproducibility, documentation, and open understandable code are important.
Where possible my AI/data science projects include complete source code, datasets, preprocessing, provenance, and evaluation tools, mostly written in Python.
Most of my other work is web based, generally REST APIs with vanilla-js and semantic HTML frontends.
Free and Open Source Software (FOSS) is an important resource for learning, transparency, community benefit;
the full source code for most of my projects can be found on GitHub.
Most of my projects are under the MIT License or similar open licenses.
My research assisted the UK Parliamentary Office of Science and Technology (POST) with identifying the key concerns
of experts regarding the COVID-19 pandemic in the United Kingdom, used to inform POST's Areas of Research Interest.
Abstract
The COVID-19 pandemic has wide reaching implications across many areas of society and experts from many
fields have offered their concerns and advice in response to the crisis. This paper presents an analysis of the
responses to the COVID-19 Expert Concerns survey conducted by the United Kingdom Parliamentary Office
of Science and Technology (POST).
We apply statistical, text-analytics, and visualisation techniques to this
new dataset to identify key areas of concern, overlapping areas of concern, and typical responses for each area
of concern. We contrast the human designed categorisation schemes produced before and after survey
responses were collected with categorisation schemes generated by our models through automated clustering.
The Text Insights Pipeline (TIP) is a research tool for visualising and interpreting collections of
unstructured text, such as survey responses, item descriptions, or short articles.
TIP combines various models and techniques to group similar items, identify naturally
occurring topics, generate names and key points for each topic, and visually present the items and
their topic groups for review by an analyst.
The tool provides a means for human analysts,
such as social scientists and policy advisers, to explore and navigate their data much more
efficiently than the traditional approaches of inspecting items in random or arbitrary order or by
use of constructed queries, which risk introducing bias or accidental omission.
This paper examines the challenging problem of new user cold starts in subset labelled and extremely
sparsely labelled big data. We introduce a new Isle of Wight Supply Chain (IWSC) dataset demonstrating
these characteristics. We also introduce a new technique addressing these challenges, the Transitive
Semantic Relationships (TSR) model, which infers potential relationships from user and item text content
and few labelled examples.
We perform both implicit and explicit evaluation of TSR as a recommender system and from new user cold
starts we achieve a hit-rate@10 of 77% on a collection of 630 items with only 376 supply-chain consumer
labels, and 67% with only 142 supply-chain supplier labels, demonstrating a high level of performance
even with extremely few labels in challenging cold-start scenarios.
TSR is suitable for any dataset featuring few labels and user and item content, where similarity of content
indicates similar relationship forming capability. TSR can be used as a standalone recommender system
or to complement existing high-performance recommender models that require more labels or do not support
cold starts.
Ralph, D., Li, Y., Wills, G. et al.
Recommendations from cold starts in big data. Computing (2020).
https://doi.org/10.1007/s00607-020-00792-y
BibTeX
@Article{Ralph2020,
author={Ralph, Sage
and Li, Yunjia
and Wills, Gary
and Green, Nicolas G.},
title={Recommendations from cold starts in big data},
journal={Computing},
year={2020},
issn={1436-5057},
doi={10.1007/s00607-020-00792-y},
url={https://doi.org/10.1007/s00607-020-00792-y}
}
RIS
TY - JOUR
AU - Ralph, Sage
AU - Li, Yunjia
AU - Wills, Gary
AU - Green, Nicolas G.
PY - 2020
DA - 2020/01/29
TI - Recommendations from cold starts in big data
JO - Computing
SN - 1436-5057
UR - https://doi.org/10.1007/s00607-020-00792-y
DO - 10.1007/s00607-020-00792-y
ID - Ralph2020
ER -
In this paper, we introduce Transitive Semantic Relationships (TSR), a new technique for ranking
recommendations from cold-starts in datasets with very sparse, partial labelling, by making use
of semantic embeddings of auxiliary information, in this case, textual item descriptions. We also
introduce a new dataset on the Isle of Wight Supply Chain (IWSC), which we use to demonstrate the
new technique.
We achieve a cold start hit rate @10 of 77% on a collection of 630 items with only
376 supply-chain supplier labels, and 67% with only 142 supply-chain consumer labels, demonstrating
a high level of performance even with extremely few labels in challenging cold-start scenarios.
The TSR technique is generalisable to any dataset where items with similar description text share
similar relationships and has applications in speculatively expanding the number of relationships
in partially labelled datasets and highlighting potential items of interest for human review. The
technique is also appropriate for use as a recommendation algorithm, either standalone or
supporting traditional recommender systems in difficult cold-start situations.
Publication Details
In Proceedings of the 4th International Conference on Internet of Things, Big Data and Security (IoTBDS 2019), pages 185-194
Ralph, D.; Li, Y.; Wills, G. and Green, N. (2019).
Recommendations from Cold Starts in Big Data.
In Proceedings of the 4th International Conference on Internet of Things, Big Data and Security - Volume 1: IoTBDS,
ISBN 978-989-758-369-8, pages 185-194. DOI: 10.5220/0007798801850194
Bibtex
@conference{iotbds19,
author={Sage Ralph. and Yunjia Li. and Gary Wills. and Nicolas G. Green.},
title={Recommendations from Cold Starts in Big Data},
booktitle={Proceedings of the 4th International Conference on Internet of Things, Big Data and Security - Volume 1: IoTBDS,},
year={2019},
pages={185-194},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0007798801850194},
isbn={978-989-758-369-8},
}
EndNote
TY - CONF
JO - Proceedings of the 4th International Conference on Internet of Things, Big Data and Security - Volume 1: IoTBDS,
TI - Recommendations from Cold Starts in Big Data
SN - 978-989-758-369-8
AU - Ralph, D.
AU - Li, Y.
AU - Wills, G.
AU - Green, N.
PY - 2019
SP - 185
EP - 194
DO - 10.5220/0007798801850194
A short paper comparing motion classifications algorithms, written during my undergraduate studies.
Abstract
Motion and gesture recognition is an active area of
research in computer vision, and holds promise for
applications in a number of areas, including
human-computer-interaction, medicine, security,
and robotics, as well as many others. Extensive research
has taken place into means of extracting information
from video for the purpose of analysis and training of
models to enable accurate prediction and classification
of the actions taking place.
Computer vision software libraries and frameworks,
such as OpenCV, commonly include a number of these
algorithms for use in research and software development.
This project will compare the performance and accuracy
of several combinations of the available algorithms and
provide analysis on the possible interactions from
combining descriptor extractors with varying feature
detectors. To evaluate these performance metrics,
machine learning models will be trained using the output
descriptors and tested by classifying human motion in
videos from multiple datasets.
A paper on high-precision location estimation, written during my undergraduate studies.
Abstract
A number of different methods and technologies exist for determining the location of mobile devices.
Depending on the desired usage, a different subset of these may be suitable for a given project.
One of the key issues in location detection is the degree of precision to which these tools remain accurate
and additionally the reliability of that data. For some usages, such as markerless augmented reality,
requiring accuracy to within a few meters, typical solutions such as GPS may not be sufficient
and alternatives must be considered.
This paper will evaluate and compare technologies and tools for mobile devices,
such as smartphones and tablets, to accurately determine their location to within 5 meters or less with consideration
for different environments and conditions in which they may need to operate given different use cases.
Techniques for ensuring operability in challenging environments will be discussed.
Hack Pompey is a social hack weekend on the south coast. We see people from all backgrounds and disciplines join together to learn something new, work together, and build something awesome!
As of 2024, I've helped organise and run seven events, some with over 100 attendees, with themes including sustainability, smart homes, and wearable tech.
I also built and maintained the old website hackpompey.co.uk, written in Gatsby (React).
ARMAS is an Augmented Reality Mobile Asbestos Survey tool
created in collaboration with Hampshire Scientific Services.
This was the dissertation project for my undergraduate degree at the University of Portsmouth.
The app is built for Android mobile devices, and is optimised for use on both phones and tablets.
It utilises Wikitude to provide an augmented reality interface for highlighting known asbestos locations.
The app was demonstrated at the university's student conference March 2016,
and has since been taken forward by a new group of students.
A custom implementation of a genetic algorithm simulator in Java.
The program allows for custom fitness functions and generates a detailed output log of the simulation,
making it an ideal learning tool.
The GUI allows for input of a custom fitness function (currently 1 dimensional only),
as well as selection of population size (population is always constant between generations),
elitism, mutation chance, and optional sigma scaling.
A number of stopping conditions are supported, including 'known best' and max-generations.
The program utilises a thread pool with an optional number of workers.
This Java application uses APARAPI to perform molecular dynamics simulation with hardware acceleration.
The GPU is utilised to calculate Lennard-Jones potential for particles with one GPU thread per particle.
Originally an attempt to recreate Angry Birds in Python.
The game models projectile motion using SUVAT and features procedurally generated levels.
While originally created as a hobby-project in 2013,
it was later re-purposed for use as a teaching resource and example program for the
IT Society Python Games Development workshop.
Floop is an information managment website, to help people organise their food storage,
which movies they want to watch, and contact details.
The site is intended to streamline many of the common functions required in shared accommodation.
Floop is a frameworkless RESTful PHP and HTML5/JavaScript webapp, with an extensive REST API.
The site has been extensively used in practice (the first launch being in 2014).
A demo version is linked below.
Logo and advice on visual design contributed by Ryan Thickett.
This project is a RESTful PHP and HTML5/JavaScript online shop,
with separate CMS (content management), Customer, and Admin pages, as well as an extensive REST API.
The site contains no functional purchase / transaction controls and is intended as a demo only.
This Java EE web application can be used to calculate the degree classification
that would be awarded to an MEng student at the University of Portsmouth based on a set of unit grades.
I created this as while grade calculators exist for most other courses, there was not yet one for MEng courses.
An attempt to make a simple Rogue-like game using VB Windows forms.
The game features procedural building and creature placement, an item system,
and a detailed tile description system.
Note: This is a very old project form my first year of college and as such is quite poorly implemented.
The project is not a serious attempt at making a functional game, more a proof of concept.
