Artificial intelligence (AI) has begun to permeate many facets of the human experience. AI will not be just a knowledge evaluation tool – it’s changing the best way we communicate, work and live. From ChatGP to AI video generators, the lines between technology and parts of our lives are increasingly blurring.
But do these technological advances mean AI can discover our feelings online?
In our latest research We investigated whether AI could detect human emotions in posts on X (formerly Twitter).
Our research focused on how emotions expressed in user posts about specific nonprofit organizations can influence actions, comparable to the choice to make a donation to them at a later date.
Using emotions to elicit a response
Traditionally, researchers depend on sentiment evaluation, which categorizes messages as positive, negative or neutral. Although this method is straightforward and intuitive, it has limitations.
Human emotions are much more nuanced. For example, anger and disappointment are each negative emotions, but they will elicit very different reactions. In a business context, offended customers may react far more strongly than upset ones.
To address these limitations, we applied an AI model that would recognize specific emotions expressed in tweets – comparable to joy, anger, sadness and disgust.
Our research found that the emotions expressed on These feelings had a direct influence on donation behavior.
Recognize emotions
We used the “Transformer transfer learning” model for recognizing emotions in texts. Transformers are sophisticated AI algorithms pre-trained on massive data sets from firms like Google and Facebook and are great at understanding natural languages (languages which have evolved naturally, versus computer languages or code).
We refined the model using a mix of 4 self-reported emotion datasets (over 3.6 million sentences) and 7 additional datasets (over 60,000 sentences). This allowed us to represent a big selection of emotions expressed online.
For example, when reading an X post, the model would recognize “joy” because the dominant emotion:
Starting the morning at college is the perfect! Everyone smiles at #Purpose #Children.
Conversely, the model would express the sadness in a tweet, saying:
I feel like I've lost a component of myself. I lost my mother over a month ago and my father 13 years ago. I'm lost and scared.
The model achieved a powerful 84% accuracy in recognizing emotions from text, a remarkable achievement in the sector of AI.
We then checked out tweets about two New Zealand-based organizations – the Fred Hollows Foundation and the University of Auckland. We found Tweets expressing sadness were more more likely to result in donations to the Fred Hollows Foundation, while anger was related to a rise in donations to the University of Auckland.
BlackJack3D/Getty Images
Ethical questions in the middle of the further development of AI
Identifying specific emotions has significant implications for areas comparable to marketing, education and healthcare.
The ability to discover people's emotional responses in specific contexts online may also help decision makers reply to their individual customers or their broader market. Each specific emotion expressed in online social media posts requires a distinct response from an organization or organization.
Our research shows that different emotions result in different results when donating.
Awareness of sadness in marketing messages can increase donations to nonprofits and enable more practical, emotional campaigns. Anger can motivate people to answer perceived injustice.
While the Transformer Transfer Learning model excels at recognizing emotions in text, the subsequent big breakthrough will come through integration with other data sources comparable to vocal tone or facial expressions to create a more complete emotional profile.
Imagine an AI that not only understands what you write, but in addition how you’re feeling. Of course, such advances include ethical challenges.
If AI can read our emotions, how will we ensure this ability is used responsibly? How will we protect privacy? These are critical questions that have to be addressed because the technology continues to evolve.
