How uses AI to match users to their compatible partners?

We can agree that life has indeed become much more convenient since the mass press coverage of artificial intelligence (AI) in early 2000. The world is moving faster, driven by technology and innovation. From self-driving cars to e-commerce, facial recognition, music and so forth, we have recommendations for everything.

Despite the advancement in technology, partner search industry still remains unsolved for potentially 1000 years or more. Have you ever wondered why?

We discovered that while the industry talks about matchmaking through one model or the other on matching people — which is the last step of partner search journey — they don’t talk about the entire search process, which is way more volatile, uncertain and frustrating and anxiety driven.

At, we aspire to transform uncertain partner search journey to certain, timely and delightful for 500M people globally through an AI-based partner prediction engine. Betterhalf’s AI engine starts learning about a user’s personality as soon as the user starts the on-boarding process. It uses AI in the five different stages:

  1. During registration of users: We capture users’ personality in six different relationship personality dimensions namely Emotional, Social, Intellectual, Physical, Relationship and Values by asking a series of sixteen Likert types questions. While we are able to estimate one’s initial personality and background information through these questions with a reliable accuracy, to begin with, we use in-product gamification, pre-match and post-match activities of the user/ feedback about the users to get more information.
  2. During pre-chat/conversation stage: While a user is interacting with the product we capture her behavioural information such as click-map, scroll-map, time spent on different sections of their matches’ profile etc to learn more about the user. For example, if a user has visited 10 matches and 5 have mentioned that they like to travel, now if the user spends more time with these profiles then the system learns that this particular user is interested in matches who actually like travelling. As another example, if a user spends comparatively more time on their matches profiles only for the matches who are in the healthcare profession and ignores other profiles than this implies that user is more inclined toward marrying a healthcare professional.
  3. During product gamification: Through product gamification, we capture more personality information from users. In a gamified way, we ask more personality based questions from users to learn more about their personality. As we get this additional data over time as users spend more time on the product, it helps us rectify any personal biases cropping in users’ mind due to bad experience which user might have faced that day. Asking questions over a period of time also helps us capture the information in different states of mind and thereby helping us evaluate the exact personality of a user. We use an AI-based algorithm to probabilistically update and correct initial personality representation.
  4. During post chat stage: After a chat with their match or at post-chat/conversation state: we also take timely private star rating feedback about the user from their matches once they have interacted/chatted with them. The feedback covers a variety of topics like the authenticity of the profile, intent to marry, timely response, compatibility with a match to various others likes/dislikes about their matches. We again apply AI to refine user’s profile and personality representation based on these feedbacks from others users. This helps our system create a more accurate version of “user” and “their personality”. This helps us give matches which are truly compatible with you.
  5. Removal of any outliers/biased data using Machine Learning: Final step is to remove/rectify any outlier data by using an ML algorithm. Suppose if on a short-tempered dimension 99.9% user rates themselves between 1 to 4 and if there is a user who has marked herself as 7 then we apply ML-based rectification to improve this data.

Our Compatibility Estimation and Matching formulae which is derived from research work going on in Cambridge University [1], [2], [3] and [4], [5], [6], [7], [8], [9], and [10], utilise this personality representation to predict best compatible matches for users. Now we use user’s basic partner preferences (Age range, height, caste, religion, location, education, salary etc.) to filter and rank those matches. This is how users receive matches and as they interact with more number of users, their matches improve over time, thereby leading them to find a compatible partner sooner with whom they are likely to be happier.


  1. Donnellan, M.B., Oswald, F.L., Baird, B.M., & Lucas, R.E. (2006). The mini-IPIP scales: Tiny-yet-effective measures of the Big Five factors of personality. Psychological Assessment, 18, 192–203
  4. Daniel Nettle. Personality: What Makes You the Way You Are (Oxford Landmark Science), ISBN: 9780199211432
  7. Henley, N.; Meng, K.; O’Brien, D.; McCarthy, W.; Sockloskie, R. (1998). “Developing a Scale to Measure the Diversity of Feminist Attitudes”. Psychology of Women Quarterly, 22(2), 317–348.
  8. Hirschfeld, Gerrit, Ruth von Brachel, and Meinald T. Thielsch. “Selecting items for Big Five questionnaires: At what sample size do factor loadings stabilize?.” Journal of Research in Personality (2014).

Authored by: Rahul Namdev (CTO/CoFounder

Leave a Reply

Your email address will not be published. Required fields are marked *