A plan to assess people’s personal characteristics from their Twitter-streams
I detect a laid-back personality
IN AMERICA alone, people spent $170 billion on “direct marketing”—junk mail of both the physical and electronic varieties—last year. Yet of those who received unsolicited adverts through the post, only 3% bought anything as a result. If the bumf arrived electronically, the take-up rate was 0.1%. And for online adverts the “conversion” into sales was a minuscule 0.01%. That means about $165 billion was spent not on drumming up business, but on annoying people, creating landfill and cluttering spam filters.
Which might, in the modern, privacy-free world of sliced and diced web-browsing analysis, come as something of a surprise. Marketing departments gather terabytes of data on potential customers, spend fortunes on software to analyse their spending habits and painstakingly “segment” the data to calibrate their campaigns to appeal to specific groups. And still they get it almost completely wrong.
A group of researchers at IBM’s Almaden Research Centre in San Jose, California, however, is here to help. According to Eben Haber, the group’s leader, the problem is that firms are trying to understand their customers by studying their “demographics” (age, sex, marital status, dwelling place, income and so on) and their existing buying habits. That approach, he believes, is flawed. What they really need is a way to discover the “deep psychological profiles” of their customers, including their personalities, values and needs. And he and his team think they can provide it.
Modern psychology recognises five dimensions of personality: extroversion, agreeableness, conscientiousness, neuroticism and openness to experience. Previous research has shown that people’s scores on these traits can, indeed, predict what they purchase. Extroverts are more likely to respond to an advert for a mobile phone that promises excitement than one that promises convenience or security. They also prefer Coca-Cola to Pepsi and Maybelline cosmetics to Max Factor. Agreeable people, though, tend to prefer Pepsi, and those open to experience prefer Max Factor.
People are, of course, unlikely to want to take personality tests so that marketing departments around the world can intrude even more on their lives than happens already. But Dr Haber thinks he can get around that—at least for users of Twitter. He and his team have developed software that takes streams of “tweets” from this social medium and searches them for words that indicate a tweeter’s personality, values and needs.
The personality-profiling part of the software is based on a study published in 2010 by Tal Yarkoni of the University of Colorado, Boulder. Dr Yarkoni recruited a group of bloggers and correlated the frequencies of certain words and categories of word that they used in their blogs with their personality traits, as established by questionnaire.
Some of the relations he found were commonsensical. Extroversion correlated with “bar”, “restaurant” and “crowd”. Neuroticism correlated with “awful”, “lazy” and “depressing”. But there were also unforeseen patterns. Trust (an important component of agreeableness), for example, correlated with “summer”, and co-operativeness (another element of agreeableness) with “unusual”.
Inspired by Dr Yarkoni’s findings, Dr Haber and his team are conducting research of their own, matching word use with two sets of traits not directly related to personality. These are people’s values (things they deem to be good, beneficial and important, such as loyalty, accuracy and self-enhancement) and their needs (things they feel they cannot live without, such as excitement, control or acceptance).
In a test of the new system, Dr Haber analysed three months’ worth of data from 90m users of Twitter. His software was able to parse someone’s presumptive personality reasonably well from just 50 tweets, and very well indeed from 200.
At the moment the system is being tested by a financial-services company. If all goes well, Dr Haber hopes to launch it commercially by the end of the year. He says the new software has the potential to serve people as individuals rather than “vague demographic blurs”. Whether they will actually wish to be “served” in this way, when the price of such service is having strangers build up intimate psychological profiles of them, remains to be seen.