Public-facing Autonomous Systems aim to increase productivity and enable more efficient and informed decision-making (Royal Academy of Engineering, 2017). Examples include Test and Trace, a Covid-19 contact-tracing application, and the Ofqual algorithm, used for automating Advanced Level results in 2020. These work without supervision and both had impact on United Kingdom citizens (Kretzschmar et al., 2020; Kelly, 2021). Investigating Autonomous Systems is beneficial as it can provide insight into how trustworthy they are, which they must by design and perception, according to the Trustworthy Autonomous System Hub (Devitt, 2018).
The collection of opinions here will act as validation or signpost the need for alternative exploration. To analyse the views expressed, a sentiment analysis – or opinion mining – tool may be deployed. This can be non-intrusive and cost-effective, as opposed to interviews or experiments. A typical source of data for sentiment analysis may be social media, notably Twitter, as many offer opinions on this public-access site, providing a large data set that Twitter’s API can analyse in real-time (Kumar et al., 2014).
However, current shortcomings of sentiment analysis, such as the lack of account for time lapsing or interaction with others affecting opinions (Liu, 2010), may impact the accuracy of the analysis into the views expressed about Autonomous Systems. Rectifying this may be crucial as there is an absence of interest in how trust in Autonomous Systems is examined over time. Additionally, the detection of sarcasm and irony, another limitation (Mohammad, 2017), could be useful when examining expressions on social media. A new approach may overcome these shortcomings so the discourse around public-facing Autonomous Systems can be understood in detail.
This project will investigate current sentiment and language analysis techniques by investigating the public discourse surrounding Autonomous Systems specifically. After this, improvements to the method will be made to gain more detail when examining this discourse. This will involve the use of primarily qualitative methods, in particular discourse analysis through corpus linguistic principles. This will involve amendments to current sentiment analysis techniques by applying emotion modelling principles within a Convolutional Neural Network to provide insight into the emotions expressed within online discourse. Finally, to detect sarcasm and irony, the use of emojis – and potentially other multi-modal features – will be factored in.
Devitt, S. K. (2018). Trustworthiness of autonomous systems. Springer, Cham.
Kelly, A. (2021). A tale of two algorithms: The appeal and repeal of calculated grades systems in England and Ireland in 2020. British Educational Research Journal.
Kretzschmar, M. E., Rozhnova, G., Bootsma, M. C., van Boven, M., van de Wijgert, J. H., & Bonten, M. J. (2020). Impact of delays on effectiveness of contact tracing strategies for COVID-19: a modelling study. The Lancet Public Health, 5, e452–e459.
Kumar, S., Morstatter, F., & Liu, H. (2014). Twitter data analytics. Springer.
Liu, B. (2010). Sentiment analysis and subjectivity. Handbook of natural language processing, 2, 627–666.
Mohammad, S. M. (2017). Challenges in sentiment analysis. Springer.
Royal Academy of Engineering. (2017). Algorithms in decision-making (pp. 1-6). Retrieved from https://www.raeng.org.uk/publications/responses/algorithms-in-decision-making
This author is supported by the Horizon Centre for Doctoral Training at the University of Nottingham (UKRI Grant No. EP/S023305/1).