What is stance detection?

Let’s start by defining the word “stance” which simply means the position of an individual towards an issue or entity. Therefore, stance detection is the inference of stand or position towards an object of evaluation in an automated manner . 

Why is it important?

Stance detection has multiple applications which are as follows: 

  1. Provide better recommendations to people:
  • Similar people will tend to have similar preferences regardless of their ideologies allowing for recommendations based on the information present
  1. Triangulate the views of people
  • Useful in areas of marketing and polling
  1. Sub-group populations into segments
  • Allows for the discovery of intersecting communities and crafting tailored messages for such groups

How is stance detection different from sentiment analysis?

People tend to mistake one for the other and consider them to be the same. However, stance detection is NOT sentiment analysis. To put in simpler terms, sentiment analysis reflects the emotional state of a user whereas stance detection reflects the position towards a target. 

Let’s take a look at this short phrase, “I feel bad he lost”. The sentiment here is negative due to the conveyed emotional state. However, the stance here is positive due to the position the speaker has taken towards the target, “he”, who has lost. This individual is empathetic towards the person who lost leading to a positive stance. 

Another example, “I am so glad that he bought the farm (died)”. The sentiment is positive as the person feels a positive emotion. The stance is negative due to the position it takes against the target as it is speaking about the individual passing. 

Thus, stance detection and sentiment analysis are detecting different things.

Stance detection in social media

It is important to understand that stance can be expressed in many different ways:

  • Phrases, like the ones we discussed in our previous examples
  • Hashtags, such as #ClimateAction vs. #ClimateGate showcasing opposing views on one topic of interest (in this case, climate change). 
  • Images/emojis as well that display a certain stance against the subject at hand
  • Type of Interactions, such as constantly liking, retweeting certain type of content on social channels
  • User profile information on their social accounts.

Types of stance detection

  1. Text stance detection: What is the stance in a “text snippet”?

Ex: “Economy is tanking”

–  Pro’s: fine-grained

–  Con’s: stance is often implicit and difficult to determine, ignores context

  1. User stance detection: What is the stance of a person/user?

Ex: user x retweets “@MSNBC: 66% disapprove of Pres. Trump’s handling…new ABC News/Ipsos poll”

–  Pro’s: uses lots of context, potentially very accurate

–  Con’s: Coarse grained, based on the assumption that the user has a durable stance over time.

  1. Stance prediction: What might be the stance of a person/user? 

This is difficult as it is trying to figure out a person’s stance on a subject without them saying anything about it.

–  Pro’s: useful, allows for predictions based in evidence

–  Con’s: difficult to be 100% sure

How does social psychology inform stance detection?

Social psychology tells us about two social phenomena.  First is “homophily”, which means that individuals who share common characteristics  tend to congregate with each other. The second is social influence and pressure, where ideas tends to spread in homophilous groups, leading to wide adoption within the group. 

These factors progressively result in so-called echo chambers. This leads individuals in  homophilous communities to share socially confirming content from similar sources to maintain their standing within the group. This gets amplified on social media because users can easily choose their social groups.

Conclusion

Stance detection on social media can be performed in different ways by observing and listening to users’ activities. 

As discussed above, users with similar stances are more likely to

  • Retweet the same accounts
  • Listen to the same media
  • Use similar hashtags
  • Share similar items
  • Use similar expressions
  • Have similar preferences
  • Live next to each other

It is a valuable area of study that helps understand and possibly predict public opinion and behavior, especially for social and political issues.

Pin It on Pinterest