Although still emerging and constantly evolving, digital methods already play an important role in ODID’s work. At a roundtable in February, researchers showcased four projects centred on digital analysis, showing how these approaches can help us understand and shape political, economic and behavioural trends – now and in the future. 

Using artificial intelligence to value technological innovation  

Xiaolan Fu, Professor of Technology and International Development, explained that to date, the world has lacked objective, accurate and affordable methods to calculate the value of new intellectual property (IP). Current valuation methods, such as discounted cash flow or a market approach via comparison of similar cases, are subjective, slow and costly, and often rely on confidential data. This makes it hard for industry to raise the funds needed for innovation based on new IP, hindering economic growth and knowledge transfer. It also leaves innovators in developing countries at a disadvantage when negotiating sale of their IP to multinational companies. Better valuation methods would enable start-ups, innovators, and small and medium-sized enterprises to increase their credit value, yet policymakers and researchers overlook this – despite the huge global market for intellectual property.  

To improve on traditional valuation methods, which are 21% accurate on average and 67% in the best record in the literature, Professor Fu and her colleagues developed large industry-specific databases and tried different artificial intelligence (AI) models – machine learning, deep learning and large language models – for predicting the value of IP. The machine learning model proved consistently strong, achieving over 90% accuracy when tested on information and communication technology (ICT) start-ups in California which had raised finance in the past ten years. This was validated using data from different sectors and regions, such as biopharmaceuticals, medical devices, renewable energy and clean technology in the EU, the United States, China and Japan – even accounting for different institutional factors that affect value.  

Fully automated, the model is very fast and is affordable, at around 10% of the cost of traditional private-sector analysis. It also uses objective indicators, although there remains a need for vigilance over biases in the data, such as gender. In California, the model showed it was harder for female company founders to raise finance, affecting their start-ups’ valuation.  

Find out more: OxValue.AI.

Profiling India’s digital election campaigns  

Research into India’s election campaigns over the past two decades shows digital media playing an increasingly important role. The 2014 general election that brought Narendra Modi and the Bharatiya Janata Party (BJP) to power was the first to feature digital media significantly, explained Amogh Dhar Sharma, Departmental Lecturer in Development Studies. The 2019 election was dubbed India’s “WhatsApp election”, while 2024’s upcoming vote is being seen as the “Big Data election”. Parties in India frequently hold digital rallies and place considerable importance on communicating with voters through social media. Analysing this digital imprint reveals underlying social and political changes. 

Internet data evolves constantly, making websites hard to research, so Dr Sharma used the “Wayback machine”, a tool which captures screengrabs from websites over time, to assess how parties like the BJP have positioned themselves to the electorate. In the mid-1990s, when access to the emergent internet was restricted to India’s urban, English-speaking citizens, the party’s website looked untidy, used English and carefully skirted its association with Hindu nationalism. By the early 2000s, technological possibilities in website design had improved and the party’s self-fashioning was also discernibly different. Now, the website can be accessed in both Hindi and English, offering informative profiles of party officials and detailed statements of the BJP’s ideological commitment to a range of socio-cultural and economic issues. 

However, digital content creation during election campaigns is no longer within parties’ total control. Researching a major party’s campaign team in Punjab’s 2022 elections, Dr Sharma found that a number of professionals, ranging from data scientists to Instagram influencers, were quickly emerging as the real drivers of a political campaign. Exploring the experiences, motivations and frustrations of these actors provides deeper understanding of the intersection of democratic politics with new media technology.   

Find out more:

Amogh Dhar Sharma (2024 forthcoming) The Backstage of Democracy: India's Election Campaigns and the People Who Manage Them, Cambridge University Press

Fear and anger on social media: the politics of emotions 

Digital analysis offers valuable insights into the emotions driving, and fuelled by, political trends. Many people think populists thrive on fear, explained Associate Professor of Politics Jörg Friedrichs, but machine analysis of political messaging shows the reverse. Fear is a governmental – or mainstream – emotion, while populism thrives on anger. Using dictionary tools, which assess a text for particular words, Professor Friedrichs’ team analysed thousands of Brexit campaign tweets, constructing a time series to identify fluctuations in fear and anger among Leave and Remain supporters. They found the governmental Remain camp was driven by fear – the establishment’s response when populism increases. This approach can be used to analyse any type of text for emotions or tone. In their paper on fear-anger contests, Professor Friedrichs and co-authors specifically apply it to tweets, adopting the Colneric and Demsar emotion recognition model.  

A machine learning approach yielded even deeper insights when used to assess UK governmental and populist framing of the Covid pandemic. The team applied their model to 5,000 tweets to code whether they were governmental (promoting risk-averse behaviour) or populist (dismissing risk), and then built a machine learning classifier which they applied to 1 million tweets. The resulting time series show, once again, that fear correlates with governmentality, and populism with fear. They also show causality – how one time series can predict others. The analysis demonstrates how a shock in one impacts another – for example, increased fear expressed by ruling Conservative MPs was a response to fear expressed by opposition Labour supporters. During the first 2020 lockdown, there were around four times more governmental than populist tweets, but by the end of mask-wearing in 2021, the gap had almost closed. The approach can be adjusted to explore many aspects of political rhetoric, such as left vs. right, or pro- vs. anti-war, providing a powerful tool for exploring diverse research questions. 

Find out more:

Jörg Friedrichs, Niklas Stoehr and Giuliano Formisano (2022) 'Fear-anger contests: Governmental and populist politics of emotion'. Online Social Networks and Media 

Using AI to interpret diplomatic signalling: lessons from the Ukraine war 

In times of crisis, AI modelling can help ministries of foreign affairs cope with the “fog of war” by adjusting the impact of the factors that contribute to reducing, as opposed to increasing, uncertainty in decision making. In the Ukraine war, for example, many politicians worldwide support for Ukraine, yet may vacillate in terms of actions that support their statements, explained Corneliu Bjola, Associate Professor of Diplomatic Studies. AI modelling can capture and analyse these signals. Professor Bjola’s pilot analysed real-time data from tweets in July 2022, around the time of the Ukraine grain deal, to see which world leaders were talking the most, and what was being said. The model allows diplomats to trace in real time which international actors are most active and confident in terms of signalling, how these signals coalesce or diverge from each other, and to what extent these signals are consistent and predictable.  

The analysis revealed that the EU, through the signals sent by its Presidents, Ursula von der Leyen and Charles Michel, was committed to supporting the long-term reconstruction of Ukraine and to demonstrating solidarity with other countries that might be threatened by Russia, such as Moldova. The NATO Secretary General, Jens Stoltenberg, and US Secretary of State Antony Blinken insisted that Russian aggression required stronger military preparation, collective deterrence and coordinated support for Ukraine, while UN Secretary General, António Guterres, highlighted the severe humanitarian costs of the war. 

The predictive capacity of the model could be improved by adjusting it to assess data on subsequent actions. This would enable it to demonstrate the alignment between political rhetoric and policy action, providing valuable insights for diplomatic responses. For instance, if Antony Blinken pledges military support for Ukraine, the model could analyse whether there are shipments of arms in the following weeks. This capability allows for a more informed approach, for example by determining whether Ukrainian leaders should express concern and dispatch representatives to the United States to secure resources. 

Find out more:

Corneliu Bjola (2022) 'Artificial Intelligence and Diplomatic Crisis Management: Addressing the ‘Fog of War’
Problem', DigDiploROx Working Paper No 6

An essential, multi-faceted research tool 

In all these cases, digital methods offer potential drawbacks, including the time complex approaches can require, possible data biases, the need for thorough validation of findings, and the dependence on big-tech companies for data – as shown when Elon Musk shut down the Twitter/X application programming interface, denying access to large tranches of data. Questions also arose over ensuring data veracity and building trustworthiness into digital systems. But whether in the form of immersive virtual reality experiences to measure human responses, or combined with traditional methods such as econometrics or qualitative research to determine causality, digital methods emerge as a far-reaching asset to research. 

AI is likely to be a game-changer for social science research, offering new tools and capabilities for data analysis, modelling and prediction. These will enable researchers to automate processes, gain deeper insights, and ultimately address complex societal problems more effectively. 

This post draws on an ODID Research Roundtable on digital methods held in Hilary Term. ODID Research Roundtables are intended to create productive conversations around shared intellectual interests, methods, and practices in the department. Each roundtable seeks to cut across the department in terms of the seniority of speakers, disciplines, geographical regions, and the location of participants in degree programmes and research groups.