Project

Affective Network

Belen Saldias 

Try Affective Network!

Emotional contagion in online social networks has been of great interest over the past years. Previous studies have mainly focused on finding evidence of affection contagion in homophilic atmospheres. However, these studies have overlooked users' awareness of the sentiments they share and consume online. In this work, we present an experiment with Twitter users that aims to help them better understand which emotions they experience on this social network. We introduce Affective Network (Aff-Net), a Google Chrome extension that enables Twitter users to filter and make explicit (through colored visual marks) the emotional content in their news feed.

The extension is powered by machine learning algorithms that classify tweets into different sentiment categories: positive posts tend to use happy or surprising language; negative posts tend to use sad, angry, or disgusting language; and posts without strong emotional language are classified as neutral.

Affective Network aims to help social media users better understand which emotions they tend to consume on social media, and how these emotions c… View full description

Try Affective Network!

Emotional contagion in online social networks has been of great interest over the past years. Previous studies have mainly focused on finding evidence of affection contagion in homophilic atmospheres. However, these studies have overlooked users' awareness of the sentiments they share and consume online. In this work, we present an experiment with Twitter users that aims to help them better understand which emotions they experience on this social network. We introduce Affective Network (Aff-Net), a Google Chrome extension that enables Twitter users to filter and make explicit (through colored visual marks) the emotional content in their news feed.

The extension is powered by machine learning algorithms that classify tweets into different sentiment categories: positive posts tend to use happy or surprising language; negative posts tend to use sad, angry, or disgusting language; and posts without strong emotional language are classified as neutral.

Affective Network aims to help social media users better understand which emotions they tend to consume on social media, and how these emotions can spread through their social networks. It was built by researchers at the Laboratory for Social Machines and the Affective Computing group at the MIT Media Lab.

Note that Affective Network does not necessarily reflect the official position of the MIT Media Lab regarding the benefits and drawbacks of filtering out specific emotional content.

Try Affective Network!

How it works

Set up

Install the Affective Network extension available in the Google Chrome Extension Store. Make sure you enable the extension by visiting your browser settings page. The extension will never make posts on your behalf.

Activate

Once you’ve enabled the extension, select the buttons below your profile picture to explore the emotional sentiment (positive, negative, or neutral) of posts in your Twitter feed. We use machine learning to classify each post into a different sentiment category.

Reflect

Are you confused? Surprised? Concerned? Take a moment to think about what you see and what it means for you. To better understand your experience, we will present you two surveys: one during the activation and one to two weeks after activation.

Frequently asked questions

  1. What about protected Tweets?
  2. What is the emotions classifier that Affective Network uses?
  1. What about protected Tweets?

    Some Twitter users may have decided to make their Tweets only visible to their Twitter followers. Hence, even though the Twitter's Privacy Policy would allow us, Affective Network does not read or evaluate emotions of protected Tweets. Find more details at https://affectivenetwork.media.mit.edu/privacy-policy.

  2. What is the emotions classifier that Affective Network uses?

    Aiming to deliver an efficient application, as well as predictions of an emotions classifier that is well known and trained for tweets (see [1] for a survey), we use the model by Go. et al. [2]. They use Twitter APIs to generate training and validation datasets. To train the model, they strip out all emojis in the tweets, to later try to predict these emojis as emotions (positive, neutral, or negative) labels. There are several emojis that can be classified as positive (e.g., :), :-), and :D), and negative (e.g., :(, :'(, and :@). The full list of emojis can be found in [2]. We use this pre-trained classifier through its available API.

    Like most machine learning systems, this classifier can make mistakes. Its authors reported accuracy score levels above 80 percent. Unintended "accidents'' can occur and be harmful for people [3], and we will address this in future versions of Affective Network by allowing participants to rectify the emotional classification of what they observe.

    [1] Giachanou, A., & Crestani, F. (2016). Like it or not: A survey of twitter sentiment analysis methods. ACM Computing Surveys (CSUR), 49(2), 28.

    [2] Amodei, D., Olah, C., Steinhardt, J., Christiano, P., Schulman, J., & Mané, D. (2016). Concrete problems in AI safety. arXiv preprint arXiv:1606.06565.

    [3] Go, A., Bhayani, R., & Huang, L. (2009). Twitter sentiment classification using distant supervision. CS224N Project Report, Stanford, 1(12).