Is this course right for me?

Target Audience: University of Strathclyde Academic/Research/KE Staff

We invite you to attend an in-person workshop on ML/AI on Monday, 15th December, from 2:00 to 3:00pm in TL423 (Teaching and Learning Building). 

Please register by 12noon on 12th December 2025.

For the first time, we have presenters from HaSS, the only faculty that has not yet presented in this series, which offers a great opportunity for others to learn about very different perspectives on ML/AI applications.

Prior knowledge of the topic is not required.

The Programme:

2:00 – 2:05   Welcome Professor Sergey Kitaev

2:05 – 2:20 From Events to Effects: Synthetic Counterfactuals for Volatile Social Phenomena

Dr Lorenzo Crippa, GPP

2:20 – 2:35 Networked Oligopsony and the Commodified Concentration of Attention: A Structural Analysis of Facebook Brand Partnerships Related to Climate Change

Dr Chamil Rathnayake, JMC

2:35 – 2:50 Abridged Debates? How competing AI algorithms revise and summarise parliamentary deliberation

Dr Z Greene, GPP

2:50 – 3:00 Q&A

Should you have any questions about the workshop, please contact Associate Dean (Research) Professor Sergey Kitaev.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

Abstracts

Speaker: Dr Lorenzo Crippa (HaSS)

Title: From Events to Effects: Synthetic Counterfactuals for Volatile Social Phenomena

Abstract: Social scientists often seek to assess the impact of events on highly volatile behaviors. Doing so requires a credible counterfactual, which is difficult to construct when the outcome is noisy and unpredictable. I present a design that combines machine learning and causal inference to address this challenge. I apply this approach to estimating the effect of politically relevant events on firms’ stock returns, an extremely volatile outcome. The design augments traditional event-study methods from corporate finance with machine learning to generate accurate counterfactual stock returns, which are then used to estimate causal effects. This effectively imputes a synthetic counterfactual for each firm, enabling the study of individual, aggregated, or heterogeneous impacts. I illustrate the approach with two applications in political science and international relations: (1) the direct and spillover effects of U.S. economic sanctions on a Chinese social-media firm, and (2) the consequences of President Trump’s February 2025 suspension of the U.S. anti-corruption policy.

---------------------------------

Speaker: Dr Chamil Rathnayake (HaSS)

Title: Networked Oligopsony and the Commodified Concentration of Attention: A Structural Analysis of Facebook Brand Partnerships Related to Climate Change 

Abstract: This study maps the structures of interaction related to climate change and climate action that emerge through brand partnerships on Facebook. The study especially focuses on examining whether the attention market operated through brand partnerships tends to display concentration or diffusion. A novel graph representation—strategic relation graphs—is introduced as an approach for mapping strategic relational choices made by brand partners, pages being sponsored, and the categories they represent. A sample of 8,139 branded Facebook posts was analysed with a multiplex network analytics approach using three competing Exponential Random Graph Models (ERGM): a baseline model, a concentrated market model, and a diffused market model. The analysis showed evidence of high concentration of activity within a significantly small group of partners, pages and segments. We identify such concentration as an attention oligopsony—a market where a core group consisting of a limited number of buyers (partners), pages and categories account for tie formation, dominating audience outreach. However, such concentration is not limited to commercial activity, since partners and pages representing different orientations such as non-profit and government organisations, and communities are present within the core group. We argue that, although the view that a handful of private sector organisations (mis)use platforms by exploiting monetised content promotion systems for greenwashing is an overgeneralisation, the critical challenge lies in the concentration of activity among a limited number of actors and segments (categories), which consequently restricts the ability of other actors and segments to mobilise audiences. 

---------------------------------

 Speaker: Dr Z Greene (HaSS)

Title: Abridged Debates? How competing AI algorithms revise and summarise parliamentary deliberation

Abstract: Political, business, and social groups place great optimism in the potential for large-language models and artificial intelligence to resolve major political challenges including government transparency and misinformation. Yet, preliminary analyses question this exuberance through the neutrality of the most popular Large Language Models (LLMs). The logic follows that LLMs cannot be relied upon to accurately relay information about the world when they themselves have been shown to reproduce societal and political biases (Gupta 2024, Gallegos 2024). Scholars have yet to evaluate the extent of these biases in the context of government transparency, however. If models are more narrowly fine-tuned to summarise the debates and deliberation within representative bodies such as the European Parliament, underlying social and political biases may be minimised.

Prerequisites