Narrative Analytics: Telling the Story behind the data

Lead Research Organisation: University of Bristol
Department Name: Computer Science

Abstract

The project aims at developing methods and a tool capable of generating Narrative Analysis
of spatio-temporal data: a succinct natural language description of a selected time-space
window of the data of interest. In healthcare and smart house scenarios, such techniques
can be combined with activity recognition models to generate comprehensible descriptions
of monitored activities with terminology and level of detail tuned to a particular recipient. In
the context of the SPHERE IRC led by the University of Bristol and funded by EPSRC these
techniques would facilitate making sense from acquired data for healthcare applications in a
more transparent way than is possible with black-box approaches.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/N509619/1 01/10/2016 30/09/2021
1729919 Studentship EP/N509619/1 01/04/2016 30/09/2020 Kacper Sokol
 
Description The major contribution of the project is a theoretical framework and a practical algorithmic approach to explain opaque predictive (machine learning and artificial intelligence) models. In addition to publishing a broad range of academic articles, we released open-source software that allows to apply our findings in practice. Our research took an interdisciplinary perspective -- combining insights from computer and social sciences -- to ensure its real-life applicability. Explainability of predictive models allows to compose a description of their behaviour that is understandable by a desired target audience, however the prose generation module was not implemented in the project.
Exploitation Route The main outcomes relate to interpreting and explaining opaque predictive (machine learning and artificial intelligence) models. During the award we have made numerous theoretical (academic publications) and practical (open-source software) contributions, which can be easily used by others to build new tools and/or continue the line of our research. The breadth and modularity of our approach allows for future work that includes further investigation of bespoke explainers, extending the published software or deploying it in a desired use-case, among many others.
Sectors Digital/Communication/Information Technologies (including Software)
 
Title FAT Forensics 
Description FAT Forensics is a Python toolbox for evaluating fairness, accountability and transparency of predictive systems. It is built on top of SciPy and NumPy, and is distributed under the 3-Clause BSD license (new BSD). 
Type Of Technology Software 
Year Produced 2019 
Open Source License? Yes  
Impact It appeals to both lay and technical users given its modular implementation. It makes all of the project deliverables reproducible. 
URL https://fat-forensics.org/
 
Description Hands-on tutorial organised at an international machine learning conference 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact In this hands-on tutorial we:
* introduced popular interpretability, explainability and transparency techniques for tabular data;
* discussed their strengths and weaknesses;
* illustrated how to identify their interoperable algorithmic components;
* showed how to decompose them into these atomic functional blocks; and
* demonstrated how to use these components to (re)build robust explainers with well-known failure points.
Year(s) Of Engagement Activity 2020
URL https://events.fat-forensics.org/2020_ecml-pkdd