BiasLenz Govhack

Govhack Project biaslenz.qrl.nz Questionable-Research-Labs/govhack-23 Presentation Video

Have you ever seen a truly awful photo of a politician in a newspaper? Next to the confident stance of the opposing party? The photography used in our media is the first thing we see, and can bias our opinions before we’ve even read the article. BiasLenz scans photography from New Zealand’s biggest media organisations, looking for a political stance in their portrait photography.

Data Scrapper

BiasLenz get’s it’s data from two places:

• RSS Feeds from each organisation, scraping the headline photo from the page
• Google Images search for images hosted on each news site with the politician

Once it has the dataset of images, it cleans up and labels the dataset:

• Deduplicating using a crop-resistant hash
• Identifying all faces in the pictures, and cropping to them
• Matching those cropped faces to the headshots on the parliament website

Sentiment Analysis

Using the schibsted/facial_expression_classifier AI model, BiasLenz determines the sentiment of the photo. Here’s the most positive, and most negative photos it found in the dataset:

Hugging Face schibsted/facial_expression_classifier