In a computational analysis of the words used by more than 65,000 consenting Facebook users in some 10 million messages, it was discovered that women use language that is warmer and more agreeable than men. Additionally, algorithms of language use predicted one's gender on Facebook 90 percent of the time. The findings are published in PLOS ONE.
In the paper, titled "Women are Warmer but No Less Assertive Than Men: Gender and Language on Facebook," some of the most commonly cited topics, or automatically derived clusters of related words, used by women included words such as wonderful, happy, birthday, daughter, baby, excited, and thankful. Some of the words most commonly cited by men included freedom, liberty, win, lose, battle and enemy.
These language differences by gender on Facebook are being investigated by a team of researchers comprised of psychologists and computer scientists from Stony Brook University, the University of Pennsylvania and the University of Melbourne in Australia. Gender is a complex, multi-faceted and fluid concept. Their paper helps to illuminate some of that complexity through big data and computational analysis, and the findings suggest gender influences the way people express themselves on Facebook.
"Looking at language in social media offers a fresh perspective on understanding gender differences," said H. Andrew Schwartz, PhD, Assistant Professor of Computer Science in the College of Engineering and Applied Sciences at Stony Brook University and a co-author of the paper.
The analysis automatically identified differences in the types of words used by women and men. Women mentioned friends, family and social life more often, whereas men swore more, used angrier and argumentative language, and discussed objects more than people. On average, women used language that was characteristic of compassion and politeness while men were more hostile and impersonal.
Some findings, added Schwartz, illustrated nuances and differences in language by gender not previously revealed.
"We were able to explore the dimensions of warmness and assertiveness with a novel data-driven technique," explained Schwartz, citing one example. "While some previous work suggests men are generally more assertive, the language in Facebook did not reflect this, showing woman use slightly more assertive language than men."
In the analysis, the topics expressed via the Facebook language were rated for how affiliative (socially connected) and assertive they were. The authors built gender-linked language around an interpersonal circumplex. While most language appeared with both genders, other language use was clearly gender-linked.
Psychologist Dr. Margaret Kern, of the University of Melbourne and one of the study authors, noted that "in many ways, the topics most used by women versus men are not surprising -- they fit common gender stereotypes. The computational methods let us make visible what the human mind does to automatically categorize people and thanks that we encounter in our everyday life."
The study also demonstrates a method to test psychological theories at a large scale, with a way to visualize the results.
With such large-scale computational studies, generating thousands of statistical results, visualization is key.
"This is a good example of visualization helping us to see the bigger picture with complex data," said lead author Dr. Gregory Park, a psychologist from the University of Pennsylvania. "If we only look at individual topics in isolation, it's difficult to see patterns in the kind of topics that are used more by women or men. However, when we visualize them together, it's clear that many of the language differences break down along these interpersonal dimensions."