How might we use context-aware data to help personalise music therapy for people with dementia?
Thoughts on gathering and using context-aware music listening data in a healthcare setting
Last year, my grandfather was diagnosed with dementia with Lewy bodies (DLB), a type of dementia that shares symptoms with both Alzheimer’s disease and Parkinson’s disease. Having witnessed his experience of dementia and the effects that transitioning to a care home had on both himself and our family, I decided to use my knowledge of experience design and social science partnered with my passion for music and digital to explore how technology might assist the personalisation of dementia care, selecting this topic as the focus for my Industry Research Project, the final piece of my MA Digital Experience Design at Hyper Island UK. I have spent the past few months carrying out primary and secondary research in this area, attempting to explore whether or not this is a feasible approach. Having carried out in-depth desk research, interviewed six experts in music streaming, data and dementia care, and attended group music therapy sessions for people with dementia, the synthesis of my research has left me with two key opportunity areas for future innovation in this field. Please keep reading for an outline of these opportunity areas. All feedback will be carefully considered and greatly appreciated.
From the outset, this project focuses on three elements, exploring the intersection in the middle. The elements are experience design — the focus of my MA; a social need — personalised care for people with dementia (PWD); and a context — music therapy. Even if the focus of this project is on how we might use music streaming data to offer deeply personalised music therapy services for PWD, I am hopeful that the opportunity areas laid out below will also contribute to helping us to think more about the broader use of context-aware music listening data.
It has been proven and observed that listening to familiar music with positive personal associations aids relaxation in PWD, particularly when they are faced with aggression or behavioural issues. Familiar music also helps to evoke memories that would otherwise have been forgotten (watch the video above to see this in action), and can help to promote connection with family members and creative self-expression.
The music selection technique currently used by music therapists working with PWD is heavily reliant on trial and error as they test to see what resonates, as I found out in both my primary and secondary research. Often times, the starting point for this trial and error selection is based on a person’s culture and age. However, in this digital age of rapid globalisation, it is risky to assume that we can rely on culture as a reliable way to identify someone’s musical or popular culture preferences. Perhaps a trial and error approach is effective for the current majority of PWD, but it is questionable as to whether this can continue to work in the future as cultures blur and content is shared irrespective of geographical proximity. This practice also clashes with Kitwood’s (1997) focus on treating the person in care over their disease. Furthermore, the use of incorrect music, or music that has negative connotations for the patient, can be detrimental for PWD, causing them to become upset, agitated or confused.
This research project presents possible opportunities for future exploration in the area in the form of two opportunity areas for the industry.
Opportunity Area 1
Qualitative data signals how music might be personalised, but extracting this data from PWD is challenging and collection processes could be enriched using social media metadata
Using a life journey map such as the one created by Playlist for Life (available here) might help to extract life highlights and memorable moments from PWD with the assistance of their families and carers. However, it is difficult to extract verbal data from PWD because dementia destroys the brain’s cognitive functions (Proctor, 2001). There is a gap in the research around how we might carry out more precise ethnographic interviews with PWD and then transform these insights into a contextual study of a person’s life: for example, examining when they listened to specific music, and what the life context might have been around this. Songs and melodies tend to get lost in people’s memories, and it is only when these are replayed that connected memories are reignited.
Couldn’t this process be automated? One such opportunity could be the use of personal social media metadata. Social media is relatively ubiquitous in our online lives and using social media mining (SMM) techniques in retrieving music information is one option of how qualitative data could be collected (Schedl, 2013). This social media metadata is already being used to personalise user experience (this is known as context-relevant UX, or CRUX). Social media retrieval explores the use of metadata (think tags) collected from social media to gather timestamps and match these with users’ music listening events. Schedl (2015) outlines how this could be done using two social media sources — specifically Last.fm and Twitter — to acquire user and listening data. During this research, listeners were categorised according to age, gender, country and genre-preferences, as well as by more specific user features surrounding listening habits, allowing increasingly person-specific recommendation capabilities. Using SMM could be an interesting way to attempt to map a person’s life journey map, matching key life events with memorable tracks to create a memory-evoking musical experience for PWD.
A possible limitation to this approach might be the discrepancy between our true selves and the personalities we portray in public (Goffman, 1959). Social media is infamous for allowing us to create filtered and idealised versions of our actual lives, leaving us to question how real the information we are posting online actually is and how accurate a journey map could be construed from our social media presence. However, could our tendency to filter life’s less desirable moments out of our social media personas mean that music with the ability to evoke negative memories could be filtered out too?
Opportunity Area 2
The capacity of technology to accurately contextualise music listening data is currently limited and can be further developed
As I learned both from my primary and secondary research, the ability to capture context-aware data is currently limited to a few areas including location, speed and time of day. With Google Play Music having recently announced the launch of new features involving the use of machine learning and contextual anchors like behaviour, location and activity to recommend new music to users however, it is clear this technology is something that is currently in development and moving quite quickly.
In the area of music therapy and dementia care, the use of context-aware data is a future-oriented solution to the personalisation of care. It is almost certain that a majority of people who now have dementia, or those who will develop dementia imminently, have not been using music streaming platforms and therefore have not built up the relevant data. In the UK, for example, a majority of PWD fall into the 65+ age bracket while 61% of Spotify users are under the age of 29. It is most likely that the use of context-aware data in choosing music for use in therapy with PWD could come into play in fifteen to twenty years, when the first generation to really use music streaming services and social media begin to enter the age bracket at most risk of developing dementia. Should musical context be able to be gathered through the use of SMM as discussed in opportunity area 1, it could see the alleviation of music therapists’ overwhelming workloads, enabling music therapy to take place in care homes on a deeply personalised scale, relying on digital automation to match music and mood.
An interesting way to think about this might be to consider how we could collect more detailed context-aware data in a manual or semi-automatic way. During my research process, I queried the use of chatbot technology to collect data from people in a semi-automatic way. The creation of such datasets are arguably beneficial to the user. Could a chatbot be developed using Spotify’s API, for example, to collect information by asking users simple questions about the songs they have recently been listening to? (for example: why are you listening to this song?) Though I did not have the time to prototype this concept in the time period permitted by this project, it would be interesting to consider what data points might be worth looking at in relation to contextual data collection, and how these might be automated in the future. A next step would be to manually prototype the idea of a chatbot using the concierge MVP over a longer period of time, perhaps 1+ years. Prototyping over an extended time period would allow space for things to occur in a person’s life, letting the MVP capture memories and seeing if it could later use the relevant tracks and surrounding data to evoke these. Below you can see a quick service design blueprint I created to illustrate this idea.
Collecting contextual data in an automatic way is reliant on the use of ubiquitous computing. The ethical issues associated with the use of ubiquitous computing (also known as ‘pervasive computing’) are important to consider in this area, and this has been a topical issue since the emergence of ubiquitous computing in the 1970s. The use of a chatbot would allow an opt-in way of collecting data while issues of consent and ethical ubiquity are ironed out.
To reiterate, this article presented two opportunity areas for the use of future technology in the area of music therapy for PWD:
- Qualitative data can show how music might be best personalised, but due to the cognitive impairments they are faced with, it is difficult to gather this data from people with dementia. Collection processes could be automated using social media mining to gather metadata which could be matched this with music listening habits.
- Context-aware data capture is currently limited, and there are ways that we could collect this in a manual way, pre-automation.
This article obviously skips over the details of the in-depth research that was carried out prior to reaching these conclusions. These opportunity areas attempt to explore how we might create and use context-aware music listening data, with a view to these processes eventually reaching automation. While the full research report will be available online in the near future, the aim of this piece was to gather feedback to help iterate on these opportunity areas. Any feedback would be truly beneficial to my work, please do comment below with your thoughts.
Nice one,
Laura
Twitter | Personal Website | LinkedIn
For a full list of sources referenced in this article, click here.
