How Social Media Can be Mined to Predict our Behaviour: New Advances in Social Networks Analysis
Updated: Apr 13, 2020
Last Wednesday, I learned how scientist could confidently predict my voter preference using only information from my public Facebook profile. Read about their research and other new advances in social networks analysis.
Attending the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM) at the Marriott Hotel in Downtown Vancouver was a real treat. As a practitioner of natural language processing (NLP), I was able to learn new ideas and network with brilliant researchers from around the world. The three sessions I’d like to share with everyone today were particularly fascinating to me because they showed me:
the effects of social media influence on adoption behavior;
a novel way to parse large amounts of social media data to understand public opinion; and
how text analysis can predict your voting preference even if share very little public information
Links to the research papers and the lead researcher's bio included when available
1. Impact of Social Influence on Adoption Behavior: An Online Controlled Experimental Evaluation
Soumajyoti Sarkar, Ashkan Aleali, Paulo Shakarian, Mika Armenta, Danielle Sanchez and Kiran Lakkaraju
Market research and survey design are passions of mine. After all it’s what we I do! This presentation was such a delight for their added twist to the popular discrete choice experiment. In their research, Soumajyoti and his team wanted to observe if social influence will change a user’s behavior from a technology adoption standpoint.
To test their hypothesis, they built an online game which took place over a period of 18 rounds divided into two parts. In the first part, participants were given incentives to make the best choice for a cybersecurity solution based on what they can learn from the different product offerings. In the second part, the same participants were introduced to social influence by being allowed to see their peers’ choices after every round. To test their hypothesis, they further varied the degree of social influence over time. Their findings found evidence that individuals can deviate from their original choices as a result of social influence.
For me, the most interesting aspect of this research is all the real-world applications. Especially in areas like promoting healthy choices, testing the effects of brand loyalty in the face of competing messages, or understanding the social effects of peer pressure... just to name a few.
2. Surveying public opinion using label prediction on social media data
Marija Stanojevic, Jumanah Alshehri and Zoran Obradovic
The study by Marija et al explored effective ways to gauge public opinion by mining social media data as opposed to more traditional methods of contacting people. Mining social media data is no trivial effort, as comments from online social media platforms are very messy –think emojis, slangs, incomplete sentences, etc. A huge amount of data is often required and labeling every message is very expensive.
To tackle these challenges, Marija et al proposed a method called, “Semi-supervised label prediction” (SLP). SLP algorithms basically selects and labels unlabeled data in an iterative process until you reach a threshold where the “machine” is no longer sure if its right or wrong. Labeled data is fed back into the training and the process is repeated until your pool of unlabeled data is minimized.
One of the most creative aspect of presentation by Marija et al is where they sought their training data. For example to understand the opinion around guns rights vs. control, they pulled information from influential twitter accounts from both sides of the argument. Through this approach, they fed two bird with one slice of bread (my less violent proverb) by leveraging a data set that contained both relevant hashtags (to be used for labeling) and posts that were consistent (thus less messy). Ultimately, their model showed accuracy upwards of 95%.
3. Show me your friends, and I will tell you whom you vote for: Predicting voting behavior in social networks
Lihi Idan and Joan Feigenbaum
While understanding voter preference using social media data is not new, research by Lihi and Dr. Feigenbaum was unique because they didn’t need information on what you've posted nor your voluntarily disclosed political preferences. Instead, by using only the publicly available social-demographic information from your Facebook profile, they were able to build a model which could predict with over 70% accuracy who you'd vote for in the 2016 U.S. Presidential election. They could even predict if you didn't vote!
For me, I learned a new vocabulary from the session, “Homophily”, which according to the first results from Google is, “the tendency for people to have ties with people who are similar to themselves in socially significant ways… which isn’t it just that old proverb, “birds of a feather flock together"?
Jokes aside, I don't know about you, but when I got home I had to take a second pass at my own Facebook profile! What about you? When's the last time you took a look at your Facebook information? Comment below to share your thoughts!
Thanks for reading. If you're interested in learning how to leverage some of these method can benefit your business, check out our work at: www.kaianalytics.com/case-studies