Leveraging personality to predict consumption preferences

Share this post:

This post was contributed by Vibha Sinha, Haibin Liu, Jalal Mahmud and Kenneth Kuo. 

We are delighted to announce the availability of new consumption preferences associated with people’s personality characteristics as part of our IBM Watson Personality Insights service. Grouped into eight high-level categories, the more than 50 preferences indicate people’s inclinations when shopping for clothing (comfort versus fashion) and automobiles (cost versus safety), their tendencies for dining (dining out versus dining in), and their attitudes toward the environment and volunteering, among other things. The consumption preferences make the IBM Watson Personality Insights service even more actionable: businesses can use these preferences to design more personalized and targeted campaigns, products and services for their customers.

Developers may already be aware that IBM’s Watson Personality Insights service produces inferred personality characteristics from text that was authored by an individual. It does this by using machine learning models gleaned from people’s written text and their corresponding psychometric surveys. The service was made generally available in February 2015 and has since been used by various businesses to better understand their customers, to create enriched customer segments and to offer more personalized products and services. While the personality traits themselves are insightful, we knew that associating consumption preferences with these traits would make them even more readily actionable.

Earlier, we associated personality characteristics with consumption preferences in our documentation. While the characteristics are very useful, developers had to use the documentation to marry them with the associated consumption preferences to take action. We are very pleased now to offer a richer set of consumption preferences directly through the service’s API, preferences that we have tested through our own studies over the past year. Developers can now easily obtain a list of preferences that are associated with an individual’s dominant traits, and they can easily incorporate these results into analytical solutions that perform segmentation, campaign management and recommendations.

You can try out the new functionality in our demo. Read on to learn more about how we learned and developed these preferences.


The relationship between personality and purchasing behavior has been studied across a variety of products and services. For example, while testing individual’s preferences concerning organic foods, Chen [1] indicated that an individual’s personality traits play an important role in establishing personal food-choice criteria. Schlegelmilch et al [2] explored the relationship between personality variables and pro-environmental purchasing behavior; the authors showed that consumers’ overall environmental consciousness has a positive impact on green purchasing decisions. Hymbuagh et al [3] investigated the relationship between personality and skydiving and found that people who scored high in adventurousness and excitement-seeking generally indulge in skydiving. In fact, we have already summarized a lot of these existing works in our documentation.

Applying these known relations between consumption behaviors and personality is a challenge for two reasons: (1) most of these works used personality data derived from surveys, and (2) their models are not publicly available. Therefore, at IBM we decided to learn these consumption preference models ourselves. When training the models, we used personality scores returned from the IBM Watson Personality Insights service as features. As a result, when you apply these models to calculate a user’s personality traits by using our Personality Insights service, the predictions are likely to be more accurate.


From existing literature, we identified 104 consumption preferences that have proved to be correlated with personality. These include preferences related to shopping for clothes or automobiles, tastes in movies and music, and so on. We then created a psychometric survey to assess an individual’s inclination for each consumption behavior. The survey was taken by about 600 individuals for whom we also had Twitter data (more than 200 self-authored tweets). We used the tweets to calculate the personality of each individual with our Personality Insights service. We then built a classifier for each consumption preference, where the input feature set was the personality information.

Depending on the consumption preference, we have built either a binary or a three-class classifier. For example, one of the binary preferences includes whether a person will or will not be influenced by online ads when making purchasing decisions. For preferences such as inclination toward outdoor sports, we use a three-level classifier for which the results are likely, unlikely or neutral.

We selected only those consumption preferences for which we found that personality-based classification performed at least 9% better than random. Of the original 104 consumption preferences, 51 satisfied this criterion; these are the preferences we chose to release as part of our API. The bar chart that follows presents details about the improvement distribution of the personality-based model over random classification, which is used as the baseline for comparison. You can see the complete list of 51 preferences we released here.


We look forward to hearing your feedback on this newest capability of our Personality Insights service.

The technical team responsible for consumption preferences includes: Haibin Liu, Neil Boyette, Vibha Sinha, Jalal Mahmud and Rama Akkiraju. Kenneth Kuo is the offering manager.


  • Chen, Mei-Fang. “Consumer attitudes and purchase intentions in relation to organic foods in Taiwan: Moderating effects of food-related personality traits.” Food Quality and preference 18.7 (2007): 1008-1021.
  • Schlegelmilch, Bodo B., Greg M. Bohlen, and Adamantios Diamantopoulos. “The link between green purchasing decisions and measures of environmental consciousness.” European journal of marketing 30.5 (1996): 35-55.
  • Hymbaugh, Karen, and James Garrett. “Sensation seeking among skydivers.” Perceptual and motor skills (1974): p. 118.
More Developers stories

The How Behind Natural Language Processing (NLP)

November 4, 2019 | Discovery and Exploration, Trends and Profiles

In part one of the NLP blog series, Sean Sodha covered what Natural Language Processing (NLP) is and why industries are starting to implement NLP into their organizations. It's time now to take a deeper dive and understand the HOW behind NLP. more

IBM launches new market leading capabilities for customer service AI with Watson Assistant

October 21, 2019 | AI for the Enterprise, Conversational Services

Today, we’re happy to announce new innovations that further advance our Watson Anywhere approach to scale AI across any cloud helping you provide better customer service AI. more

A recap: here’s what you missed at this year’s BoxWorks

October 14, 2019 | AI for the Enterprise

At BoxWorks 2019, we were able to showcase the IBM and Box partnership, along with how it works and what’s in store for the future. more