Greater Engagement of Mental Wellbeing Solutions Through Personalisation

Greater Engagement of Mental Wellbeing Solutions Through Personalisation

Mental health problems are a worldwide pandemic. Over 50-80% of New Zealanders will experience mental distress in their lifetime, costing an estimated $12B or 5% of GDP annually. While evidence indicates early detection and intervention is the key to improving outcomes, there is often a considerable delay from the onset of symptoms to the time help is sought.

Existing services are inadequate, particularly in the area of prevention and in supporting the majority of the population before they reach crisis point. Stigma and lack of access continue to be major barriers to people seeking help. We need new ways of increasing awareness and delivering assistance to sufferers, would-be sufferers and their carers.

This is where Mentemia comes in

Mentemia is a digital platform that helps people improve their mental wellbeing by incorporating science-based tools and expert content into approachable, ‘bite-sized’ content and interactive tools. The purpose of this study was to investigate if personalised suggestions of solutions based on the user’s Mentemia profile increase engagement with the Mentemia app. 

The involvement of data scientists in this project through PDH had two main purposes:

  1. To deliver an initial version of a machine learning-based recommender which enables the personalised daily content suggestion on the app’s main screen, as a replacement of the current static rule-based suggestion mechanism;
  2. To help build up the Mentemia team’s data science capabilities for future maintenance/improvement of the ML-based recommender.

The team reviewed the historical foundations that underpin recommender systems as well as the current state-of-the-art and decided to adopt a hybrid approach that combines collaborative filtering and content-based recommendation approaches as the starting point. An initial recommender covering the articles, videos and interactive tools has been trained. It has been deployed in a test environment, running a daily scheduled update to include new items and new users.

It employs not only the user-item interactions data but also the attributes of the users and items such as: 

  • Why they use the app.
  • The estimated length of time to consume the content.
  • The topic of an article and/or video. 

In doing so, the hybrid recommender unites the advantages of the collaborative and content-based recommenders. It can now provide not only personalised suggestions for users who have already consumed items in the app, but also more relevant recommendations for new users based on their user attributes before they interact with the app’s offerings. A code base to run automatic optimisations for Mentemia has been delivered at the same time to facilitate future recommender improvement. 

What are the next steps? 

We expect that Mentemia would develop further business logic based on the recommendations generated. Before moving the recommender and its surrounding logic to a public environment, some further testing would be required to answer questions such as: 

  • What if all the items have been seen by the user? 
  • What if the recommended content has already been consumed by the user? 

Mentemia will be monitoring the recommender’s effectiveness according to the relevant engagement metrics such as click-through rate (CTR, the proportion of recommendations that end up being clicked) once it becomes available for the users. Moreover, Mentemia will be regularly updating the recommender as new features and content are added to their database, such as taking into account additional users characteristics, and adding new content formats such as infographics and audio.


Key members of the research team:

Fiora Au, Mentemia – Project champion

Simon Hartley, Mentemia – Solution Lead

Sakthi Harish, Mentemia – Data Scientist

Dr Ning Hua, Orion Health – Advising Data Scientist