FEATURE16 March 2012
FEATURE16 March 2012
Research by Tom De Ruyck and Stephan Ludwig finds that shared ‘function word’ usage among MROC members is correlated to the amount of community activity.
In mining online text-based conversations for insights, the market research industry has been primarily concerned with what people say rather than how they say it.
The focus of text analytics tools has been on extracting content words – verbs, adjectives, nouns and adverbs – to meet the requirement of companies to know what is being said about their brands, products and services and the sentiments being expressed.
But in doing so, we ignore the function words (prepositions, pronouns) which contour how people express their ideas and views. While there are only roughly 500 function words in the English language, they make up 55% of our daily word usage.
“The greater the percentage of overlap between community members’ function word usage, the greater the average amount of posts obtained per community member per day”
James W Pennebaker, a psychology professor at the University of Texas at Austin, has written multiple studies with colleagues to show that these function words, rather than being clutter to be discarded, are key to “understanding relationships between speakers, objects, and other people”.
In a study on speed dating, for example, Pennebaker et al showed that dates who more closely mirror each other’s function word usage were more likely to start dating and have a longer-lasting relationship.
This intrigued us, so we set about trying to understand whether the relationship-forming attributes of function word usage applied in a market research online community (MROC) context.
Our aim was twofold: we wanted to see whether we could increase the degree to which MROC members mirror each other’s function word usage by hosting introductory chat sessions for participants, and we wanted to investigate whether shared function word usage related to increased activity levels in an MROC.
We compared 30 MROCs with a shared set-up structure and a shared community management team. We text-mined the function word usage from members’ community posts and computed similarities between members’ usage styles as suggested by Pennebaker et al.
First, we found that the greater the percentage of overlap between community members’ function word usage, the greater the average amount of posts obtained per community member per day ( 0.485, p < 0.01 ). In trying to find out if our introductory chat sessions could help build stronger community affiliations, we measured average attendance percentages for each MROC individually and then analysed the relationship between attendance percentages and the overall degree of function word matching in the community. We found a significant positive relationship between the two ( 0.120, p < 0.01 ).
Note that chat session attendance percentages also had a direct effect on the amount of posts per member ( 0.306, p < 0.05 ).
In order to check for the robustness of our results, we controlled for the average interest community members had in participating in the particular MROC by sending out an initial survey (with an average response rate of 78%) prior to the start of each MROC.
We found that average topic interest level had a significant impact on the amount of posts generated by communities ( 0.522, p < 0.01 ). Yet the level of topic interest did not significantly impact the degree to which members’ function word usage overlapped. This shows that the affiliation between members is unrelated to their average topic interest, though it is a strong predictor of their participation.
Mining for people’s particular use of function words may open up a treasure trove of insights on their personal backgrounds, emotional state, personality, perceptions and relationships. We have illustrated how a strong shared language relates to the amount of posts MROCs can accumulate, and how kick-off sessions help in establishing stronger shared language – yet that is just the tip of the iceberg.
Analysing the particular usage of function words could reveal the profiles of customers who are either recommending or complaining about a company’s products and services online. Moreover, given the insights into peoples’ emotional states that are retrievable through function word usage, it is recommended that text-based sentiment analysis start incorporating these words.
Tom De Ruyck is head of research communities at InSites Consulting and Stephen Ludwig is a researcher at Maastricht University.
Ireland M., Slatcher R., Eastwick P., Scissors, L., Finkel E., Pennebaker, J. ( 2011 ) Language Style Matching Predicts relationship Initiation and Stability, Psychological Science, 22( 1 ) 39-44.
Pennebaker, J. W. ( 2011 ). Your Use of Pronouns Reveals Your Personality, Harvard Business School Publication Corp., 32-33.