Understanding Successful Social Interactions
Research Question: How do conversations lead to social connection?
Data: Over 1,800 pairs of strangers conversed with one another for 25 minutes via our custom online conversation platform, which automatically records video and audio data from each conversation. After the conversation, participants reported their experience with the conversation and their impressions of their conversation partners.
Methods: We have automated tools to extract tons of information from the audio and video data of the conversation. We can extract the mood of each individual from their vocal tone or their facial expressions. We can examine their language to learn what they talked about, how much they expressed empathy, or how much they disclosed personal information about themselves. We can examine the video data to see the extent to which they were paying attention to the screen or nodding along to their partner. We can look at synchrony between the two individuals to test how aligned they were with each other. With all of these features, we can then run machine learning analyses to test how each conversation feature or combination of features contributes to the dyad’s social connection, closeness, and enjoyment at the end of the conversation.
Challenges: Extracting features that we’re interested in from conversation data can be tricky. For instance, how do we define an objective measure of self-disclosure? A coarse method is to count the proportion of turns that include first-person pronouns. A more nuanced method is to train a linguistic embedding model to learn when a sentence expresses a real, meaningful disclosure. We are currently developing these kinds of advanced linguistic methods to better capture our constructs of interest.
Findings: Preliminary analyses reveal that the more people self-disclose, the closer they feel to their conversation partner, but it doesn’t affect how close their partner feels toward them. We are currently examining different hypotheses for why we see this asymmetry.