Last updated 9 July 2024
YouGov will be publishing regular reports of the news story that is cutting through the most among the British public, offering in-depth, fast and innovative political analytical insights.
We have developed a tracker using an in-house large language AI model which allows us to capture and model people’s written responses to the question: ‘Thinking back over the past week, which news story did you hear about the most?’.
This tool uses artificial intelligence to identify which single recent news story has resonated the most for individual respondents, which we can then aggregate into a nationwide ‘cut through’ report.
What is topic modelling?
Our cut-through tracker uses topic modelling to analyse our open text responses. Topic modelling is a text analysis tool that systematically identifies recurring themes in a collection of texts. Commonly used in natural language processing, topic models typically employ complex probabilistic models or matrix factorisation to analyse the relationships between words. More recently though, a new type of topic model has emerged, one that taps into neural networks and machine learning (otherwise known as AI).
These new methods are particularly useful, as they identify thematic patterns in texts based on the relation of words within sentences, providing context to their meaning and therefore improving the analysis of their salience. To function, these models tap into libraries of sentence-based transformer models, which are pre-trained on vast amounts of existing texts, allowing them to ‘understand’ contextual references.
This is particularly relevant when it comes to political text analysis. For example, we know ‘right-leaning’ refers to an ideological preference towards socio-economic and cultural policies of the right-wing, rather than someone physically leaning on the right.
What does it have to do with polling?
In polling research, most questions are designed to be ‘closed’ rather than ‘open’ – that is to say, respondents are given a list of specific answer options to choose from. While this allows us to swiftly measure the general public stance on an issue, it does not provide any depth of insight into those views, nor does it pick up public opinions not in the pre-selected list of answer options.
Previously, open text responses have been analysed manually, but this is a vastly time-consuming task, and can create inconsistencies in the interpretation of people’s responses. Topic modelling streamlines the process of identifying themes and classifying responses. Additionally, it makes the process replicable and transparent, where anyone can re-run the model and obtain the same results.
Why use it now?
Artificial intelligence can offer new insights into information that was already available, but has not been analysed systematically, until now. The analytical power it provides to different industries is a popular and heated subject of discussion. We are aware the use of this technology on political surveys is still nascent, and therefore know that it is paramount we develop our model methodically. While we use a unique fine-tuned algorithm to analyse open-response data, there is human supervision at each step of the process. Most importantly, we revise our results meticulously and run internal checks before publishing any findings.
YouGov are continuing to offer innovations and new products in the AI space - reach out to us if you think our AI tracker could be of use to you in your future projects.
See the full results for each survey below
- 5-6 June
- 6-7 June
- 7-8 June
- 10-11 June
- 11-12 June
- 12-13 June
- 13-14 June
- 14-17 June
- 17-18 June
- 18-19 June
- 19-20 June
- 20-21 June
- 21-22 June
- 22-23 June
- 24-25 June
- 25-26 June
- 26-27 June
- 27-28 June
- 28-30 June
- 1-2 July
- 2-3 July
- 3-4 July
Photo: Getty
