Posted on Wed 15 May 2019
The Final Honk
Last Friday as my residency drew to a close, I presented the Sentiment Honk to an audience as part of a lunchtime talk and got some excellent feedback. Here are some things that I’ve learnt in the making of this project.
Last Friday as my residency drew to a close, I presented the Sentiment Honk to an audience as part of a lunchtime talk and got some excellent feedback.
Here are some things that I’ve learnt in the making of this project.
Sentiment Analysis is being applied in a multitude of ways including interactive billboards, mental health assistants, call centre prompts, stock price predictions, mood-based sales and fraud detection.
Human hands are way more advanced than robots.
Google’s speech-to-text is bad with regional accents.
A report by the National Research Council Canada found that more than 75% of 219 sentiment analysis systems tended to mark sentences involving one race or gender with a higher intensity than the other (i.e ‘My Dad is laughing’ is deemed as a more positive sentence than ‘My Mum is laughing’)
Algorithms are judging me.
According to Google, the word ‘Tits’ is more positive than the word ‘dick’.
Every time the actress Anne Hathaway gets positive reviews for her films, stock prices of the firm Berkshire Hathaway go up.
People like swearing at robots.
It is important to question decisions made 'by machines'.
According to a sentiment analysis algorithm developed by Admiral, the car insurance company, good drivers write in short concrete sentences and use lists in their facebook posts while bad drivers use a lot of exclamation marks.
Sentiment Analysis can’t handle sarcasm.
The plan now is to do some tweaking with all the feedback, make more and exhibit! Meanwhile, I will be working on another AI related project, this time regarding facial recognition algorithms, as an Automation fellow for SWCTN https://swctn.org.uk/automation/fellows/rachel-smith/