Leveraging free text data written by users of a mental eHealth application
Type of assignment
Graduation project (MSc, 30EC)
Internship (MSc, 20EC)
Own data collection or existing data?
Type of research
Machine learning / Natural language processing
Roessingh Research and Development (RRD) is an internationally recognized impact lab for personalized health technology with expertise in user-oriented health technology focused on rehabilitation, sports and health management. Together with eight international partners in Switzerland, Portugal, and The Netherlands, RRD has been developing an online service for mourning older adults that features a virtual agent. This service is called LEAVES. LEAVES has a modular structure and consists of 10 content modules about grief, including a number of therapeutic exercises to support the users’ healing process. The exercises are usually answered in free text format by the user in the application.
At RRD, you have the opportunity to explore the potential of free texts written by its users for personalization and monitoring purposes. Ultimately, everyone takes a different road through the valley of grief, but nobody should walk it alone – and automatic text analysis can be a good alternative to more obtrusive questioning methods to enable personalizing an application such as LEAVES to its users.
The text data in this assignment stems from an evaluation study of LEAVES conducted between February and November 2022 in the Netherlands. The data set contains text data written by 21 users distributed over 43 exercises, with a total of ca. 30,000 words (excluding punctuation, but including stop words). All text data is written in Dutch.
We are looking for enthusiastic students with a background in computer science and/or human-computer interaction, an interest in (mental) eHealth, and experience with natural language processing (NLP) techniques, theoretically and programatically. Speaking Dutch is very desirable because all text data is in Dutch.
Exemplary assignment topics are not limited to, but could involve:
- Exploring how well a pre-trained sentiment analysis model can distinguish between different levels of suffering between users as indicated by their initial grief scores.
- Predicting user characteristics when they start using the application, such as how long ago the loss occured (based on, for example, sentiments as expressed in free text or the order in which people answered which exercise).
LEAVES project website
van Velsen, L., Cabrita, M., Op den Akker, H., Brandl, L., Isaac, J., Suárez, M., ... & Canhão, H. (2020). LEAVES (optimizing the mentaL health and resiliencE of older Adults that haVe lost thEir spouSe via blended, online therapy): Proposal for an Online Service Development and Evaluation. JMIR research protocols, 9(9), e19344.
Brodbeck, J., Jacinto, S., Gouveia, A., Mendonça, N., Madörin, S., Brandl, L., ... & van Velsen, L. (2022). A Web-Based Self-help Intervention for Coping With the Loss of a Partner: Protocol for Randomized Controlled Trials in 3 Countries. JMIR research protocols, 11(11), e37827.