Understanding the Influence of Politicians' Financial Interests using Natural Language Processing and Network Analysis

This project aims to examine the influence of financial interests on the actions of politicians. My central hypothesis is that politicians with significant interests in certain industries will support legislation that would be beneficial for these industries. Research questions I intend to explore include:
Can we predict political decisions based on the financial interests of Members of Parliament (MPs)?
How do politicians' interests affect their publicly voiced opinions? And to what extent do lobbyists and
corporations benefit from their financial relationships with legislators?

Whilst the UK ranks among the world's most open governments, McKay and Wozniak (2020) argue that the searchability and overall usability of published UK lobbying data can be categorized as low. This project will leverage recent advances in natural language processing (NLP) to convert unstructured textual information into a quantitative dataset regarding the financial interests of UK MPs. I will then use this new dataset-which will be made public-to assess the above topics and other important questions about the fairness and representativeness of outside interests' influence over the Members and policies of Parliament.

The project combines two distinct disciplines-political science and computer science. My proposed supervision team brings together an expert on lobbying, Prof Amy McKay (Politics), an expert on data science and network analysis, Prof Hywel Williams (Computer Science), and an expert on NLP as applied to politics, Dr Travis Coan (Politics and Q-Step), to help me to develop and apply a robust method for systematically evaluating lobbying and financial interests.

Despite increasing masses of information regarding potential threats to unbiased, representative policymaking, few scholars have leveraged NLP to evaluate the influence of financial interests over policymaking and public discourse. Exceptions are Kluver (2009), who demonstrates that a tool known as Wordfish (Slapin 2008) can be used to locate lobby groups on a policy spectrum, and Boussalis and Coan (2016), who use text-mining to evaluate signals from conservative political groups and determine that promulgation of climate scepticism is increasing, not decreasing. The potential of NLP to contribute meaningfully to the contradictory literature on the influence of financial interests in
policymaking is therefore considerable.

The proposed research would involve an application of Language Representation Models (e.g. the BERT architecture developed by Devlin et al. 2018) to UK parliamentary records, such as the Register of Members' Financial Interests, to generate structured data that can be quantitatively evaluated.

The resulting dataset of could be combined with other data, such as Tweets, to explore the effect of these interests. Further extensions could use network analysis to investigate relationships between politicians and companies (Porter et al. 2009). A bipartite network linking MPs and corporations could be developed to analyse policymaking communities in the UK, as has been done in the US (Porter et al. 2007; Ward et al. 2011).

To help answer the research questions I have identified various sources:
- Register of Members' Financial Interests: This register contains complex unstructured data from Parliament regarding payments made to MPs. The interests are categorized in ten topics including employment, donations and shareholdings. This will serve as my primary dataset.
- Votes in Parliament
- Cabinet Ministers' Meetings, Hospitality, Gifts and Overseas Travel
- The Electoral Commission's database of donations, election spending and party accounts
- Listing of Publishable Central Government Tender
- Twitter API / Parliament Hansard

I will secure approval by the College ethics committee for this project. Generated data should be validated before publication.