A new research finding highlights the need for stronger linguistic indicators to be developed for use in online personality prediction tools, which are being rapidly adopted by companies to improve digital marketing strategies.
Today, marketing companies use predictive algorithms to help them forecast what consumers want based on their online behaviours. Companies are also keen to leverage data and machine learning to understand the psychological aspects of consumer behaviour, which cannot be observed directly but can provide valuable insights about how to improve targeted advertising.
For example, an 'extravert consumer' might be attracted to marketing messages that match their personality, and retail brands could then choose to target such consumers by using more extraverted and creative language to advertise their products.
However, personality prediction tools available today that are used by marketing firms are not entirely accurate due to a lack of theoretically sound designs.
Current machine learning algorithms for personality prediction can seem like a black box -- there are many linguistic indicators that can be included in their design, but many of them are dependent on the type of computer application used. This may lead to biases and overfitting, an error affecting the performance of the machine learning algorithms. This begs the question -- how should we create robust and accurate personality predictions?"
The study found a correlation between extraverts and their tendency to use certain categories of words. The results showed a small strength of the relationship between extraversion and the use of "positive emotion words" and "social process words."
Positive emotion words are defined by psychologists -- using text analysis tools -- as words that describe a pleasant emotional state, such as 'love', 'happy', or 'blessed', or that indicate positivity or optimism, such as 'beautiful' or 'nice'. Social process words include words containing personal pronouns except 'I', and words showing social intentions, such as 'meet', 'share' and 'talk'.
To establish the effectiveness of such linguistic predictors, the team reviewed 37 studies looking at the same topic to conduct a meta-analysis. Extraversion was determined using internationally recognised personality type questionnaires.
Moving forward, the research team will investigate the relationship between extraversion and other word categories.
While machine learning and predictive analytics can provide companies and marketers with an added advantage in their business strategies, more thought must be put into the design of such analytical models.
Source: Nanyang Technological University