Music Recommendation System for Old People with Dementia and

Other Age-related Conditions

Miriam Allalouf

, Avi Cohen

, Lea Cohen Sabban

, Ayelet Dassa

, Sagi Marciano

and Stella Melnitzer Beris

Department of Software Engineering, Azrieli College of Engineering, Jerusalem, Israel

Department of Music, Bar-Ilan University, Ramat-Gan, Israel

Keywords: Music Recommendation System, Music Metadata Mining, Music Information Retrieval, Digital Ageing,

Dementia, Music-based Intervention, Machine Learning Algorithms, Music Repository.

Abstract: The worldwide increase in life expectancy can be accompanied by age-related degenerative conditions such

as dementia. Dementia poses significant challenges for which music is a beneficial non-pharmacological

intervention. Based on research and clinical expertise we developed a web-based system, termed Tamaringa,

that builds and displays customized playlists. The recommendation mechanism incorporates an old person's

age, birthplace, and popular songs from their youth. That particular range is known as being most accessible

to seniors in terms of memory. Although there are a lot of repositories containing metadata and information

about music, there is no single repository that addresses all our requirements in terms of specific metadata,

range query application programming interfaces (API) capability and popularity information. This study

explores the APIs of several repositories in order to populate our internal database with suitable songs that

are required for accurate personalized recommendation. A preliminary promising pilot enabled twenty-four

residents in an assisted living facility in Israel to engage and enjoy the music recommendation system.

Personalized playlists were created using the system; the medical staff reports were positive. Further research

will help to develop our system and eventually to integrate its use both in assisted living facilities and at home.

1 INTRODUCTION

According to the World Health Organization (WHO),

life expectancy is increasing around the world. Aging

presents challenges such as isolation, loneliness, as

well as currently incurable diseases like dementia and

Parkinson’s. It is estimated that approximately 50

million people are affected worldwide by dementia

(WHO, 2017). The care of people with dementia has

become a focal point for policy makers, researchers

and healthcare providers, who recognize the need for

competent and effective services (Prince et al., 2016).

Studies show that music is an effective non-

pharmacological intervention for people with

dementia. It has an ameliorating effect on agitation

(Levingston et al., 2014; Ziv, Granot, Hai, Dassa &

Haimov, 2007), it improves well-being for the person

with dementia (Baird & Thompson, 2018), and

supports the caregiver in providing the best care

possible (Ray & Fitzsimmons, 2014; Särkämö, et al.,

2014).

The use of music can be crucial in the care of

people with dementia. Music-based intervention

mostly includes listening to familiar music. Usually a

customized playlist is gathered and recorded, either

manually or on a spreadsheet. Such music-based

interventions are not documented in digital

repositories and are thus unavailable for the purpose

of creating more accurate playlists for each person. In

this paper, we describe a system we have developed,

termed Tamaringa, that automatically builds

customized playlists and then stores them in a

scalable system that allows for fast playlist display on

demand.

Nowadays, music streaming systems, such as

Spotify (Eriksson et al., 2019) and YouTube (Airoldi

et al., 2016), that recommend customized playlists to

their listeners are popular. These systems are

unsuitable for people with age-related conditions in

several ways: The recommendation algorithm is not

adapted for cognitive difficulties; Spotify provides

general and calm playlists for old people without any

specific profiling or comprehension regarding the use

of music that taps into long-term preserved memory

Allalouf, M., Cohen, A., Sabban, L., Dassa, A., Marciano, S. and Beris, S.

Music Recommendation System for Old People with Dementia and Other Age-related Conditions.

DOI: 10.5220/0008959304290437

In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 5: HEALTHINF, pages 429-437

ISBN: 978-989-758-398-8; ISSN: 2184-4305

429

in the case of dementia and other related conditions;

Youtube recommendation algorithm is based on

averaging the preferences of people from different

ages and origins and not suitable for elderly people

that mostly prefer music from their past. In addition,

old people in general and those with dementia require

easier human-computer interface than the one that is

provided by the regular desktop or smartphone for

Spotify or YouTube (Finamore et al., 2011). Thus, it

is not practical to expect they will handle their own

playlist by themselves or ask their caregivers to

handle a unique playlist for each person.

The MUSIC & MEMORY® (Thomas et al.,

2017) organization helps people in nursing homes and

other care organizations who struggle with a wide

range of cognitive and physical challenges to find

renewed meaning and connection in their lives

through the gift of personalized music. They train

care professionals how to set up personalized music

playlists, delivered on digital devices, for those in

their care. But they do not provide an automatic

system such as the one presented here. The

Tamaringa platform identifies personal musical

preferences related to preserved memories and builds

an appropriate playlist accordingly.

Figure 1: Web Application guided by an instructor.

Figure 2: Suggested playlist with buttons to rate the song.

2 MUSIC REVIVES YOUR SOUL

Until now, customized playlists were put together by

a close family member, hand-written on a sheet of

paper, and given to a caregiver or to nursing home

staff to play for the person with dementia. As the

music sessions were played at the discretion of the

caregivers and were not monitored, they could not be

modified to generate more precise playlists, and

behavioural or other well-being measurements were

at best mediocre.

The Tamaringa platform automatically builds

customized playlists, based on where and when

people were born, where they grew up, where they

went to school and university, their ethnicity,

religious and social background, and their taste in

music, among other things. Suggested playlists are

kept in storage and played on demand in a simple

interface that can be operated by the end-user (Figure

2). The results of patients' preferences and reactions

to the chosen playlist (such as verbal comments,

singing, or reduction of agitation) are saved to the

internal database by an instructor or family member

in order to create better playlists and hone baseline

target profiles.

In this work we followed the Music Information

Retrieval (MIR) modelling as surveyed in Schedl et

al., (2014) that categorizes two main classes: user

profiling and context, and music content and context.

In particular their survey focused on user-centric

music retrieval as well as how to recommend music

based on the user's profile and preferences. User

profiling is described in Section 2.1. The music

recommendation algorithms that are described in

Section 2.3 are based on having the relevant music

information and metadata in the system (described in

Section 2.2).

2.1 User Profiling

A preliminary pilot was conducted in a nursing home

in Israel. Israeli society is culturally diverse and only

twenty-two percent of the elderly were born in Israel

(Myers-JDC-Brookdale, 2018). Old people may have

grown up in other countries and may speak various

languages. The elderly participants in the nursing

home were from different cultures, spoke different

languages and were at different stages of dementia.

Lack of memory was evident, as well as other neuro-

psychological symptoms such as agitation,

depression, verbal and physical aggression and

refusal of medical treatment.

People between the ages of 75-95 in Israel have

usually immigrated from a variety of countries over

HEALTHINF 2020 - 13th International Conference on Health Informatics

430

the world, at different stages of their lives. We

identified and split the residents into several groups

with regard to their origins. For example, 90-year-

olds who arrived from Russia would probably like

war songs such as the Red Army choir Smuglianka

https://www.youtube.com/watch?v=n7hHlh2IusY or

Dark is the Night by Mark Bernes from 1943

https://www.youtube.com/watch?v=Wdx7fkss6kU.

Younger people around 75-years-old would prefer

music from Soviet films like Kidnapping, Caucasian

Style, a Soviet comedy film from 1967

(https://en.wikipedia.org/wiki/Kidnapping,_Caucasia

n_Style).

As an example, 85-year-olds who arrived from

Arabic-speaking countries (Iraq, Syria, Egypt,

Morocco, Tunis and Algeria), and who speak Arabic,

would prefer Arabic songs e.g. Umm Kulthum and

Farid al-Atrash. The 70-85-year-olds who came from

North African countries and spoke French in addition

to Arabic, had been influenced by French culture and

songs by French performers like Edit Piaf, Jacques

Brel and Nana Mouskouri, and they would probably

elicit positive reactions. Old people who were in

Israel in their twenties would like the Israeli and

Western songs that were played during the years

1940-1970. Thus, in order to recommend suitable

music, we had to define their profile by their age, the

period corresponding to their twenties, their first

language, place of birth and other attributes.

The nursing home where we applied our system

established computer rooms with instructors who

could offer help, where the residents could come and

watch video content and listen to music. As depicted

in Figure 1, each person has a pair of headphones and

his/her own computer. The participant received

personalized content that fitted his or her profile,

operated by an instructor in the room. It was evident

that the residents at different stages of dementia were

eager to come to the computer room. We realized that

it was important to accompany music with a video

rather than only playing the audio. Usually, the role

of the instructor was to suggest a suitable content to

the person and follow the resident's reactions. We

learned that music that they liked in their twenties

pleased them very much and increased their

motivation to return to the computer room. This is in

accordance with previous research which shows that

the majority of seniors’ accessible songs are from

earlier decades of life (Cohen, Baily, & Nilsson,

2002).

Hence, the instructors suggested music that fits

the background of each person. In this sense it is

similar to a regular customized MIR (Music

Information Retrieval) application that recommends

music for the user based on the user’s cultural

background, interests, musical knowledge, and usage

intention, among other factors. But Tamaringa also

learns the person’s preferences given their tendency

to remember and react more positively to music from

their past rather than unfamiliar or recent songs.

2.2 Music Metadata Retrieval

Our recommendation system contains relevant

information about music content that is kept in an

internal database that is used for the recommendation

pool. To populate the database, we pulled the

information automatically from external repositories

using software tools. The information about songs

that we looked for included detailed information such

as the name of the performing person or band, the

name and the language of the song, creation or

recording date, the year of release, the country of

origin, the tags that characterize the song and

popularity ratings for a song at a relevant period. In

order to retrieve songs from several years, we

required the repository Applicative Programming

Interface (API) to allow year range query.

The need for this diverse information became

problematic due to several reasons. We encountered

difficulty in gathering this music metadata. Although

there are a lot of repositories containing metadata and

information about music, there is no single repository

that addresses all our requirements in terms of

specific metadata, range query API capability and

popularity information.

In addition, we found that information about

songs from the mid-1950s, for example, that was

taken from several repositories, contained

contradictions and was vague. Therefore, to build the

database that would serve as the basis for the music

recommendation system, we examined several

repositories, each with different programming APIs.

2.2.1 Youtube

Study of the YouTube API showed that the metadata

of songs in YouTube usually include the title of the

song, the performer’s name and the comment of the

user who uploaded it

(https://developers.google.com/youtube/v3/getting-

started) . These features did not help in characterizing

the songs in the way we needed. Still, we used

YouTube to display the songs, as will be described

later.

The popularity rating of each song is also a

difficult parameter to meter. We can find today’s

popularity rating of a song from the 1930s. But there

Music Recommendation System for Old People with Dementia and Other Age-related Conditions

431

Figure 3: Tamaringa Workflow.

is no way to evaluate its popularity back in the 1930s.

The Billboard-Hot-100 shows the best 100 songs in

USA (only) every week since 1901 to today

(https://www.billboard.com/charts/hot-100). There

is no way to query Billboard about the popularity of a

certain song since it does not provide any API or

access to its internal database. Thus, we assumed that

a song that was then popular will have a large number

of plays / likes on YouTube today and relied on this

information for the popularity rating a of a song.

Another requirement of our project was the ability

to view and play a song while remaining in the

context of our application without being transferred

to another application such as YouTube. It is clear

that YouTube is the main platform for this, because it

is the largest system, with extensive content.

At the beginning of the project it was not possible

to get a link to the song by its name, since the songs

are characterized by a unique ID for YouTube.

YouTube considerably improved the documentation

and client API for many languages, including

Javascript, on the 27

of October, 2017,

(https://developers.google.com/youtube/v3/revision_

history#april-27,-2017) and we could obtain a link to

the video by the name of the song and the artist. And

thus, we overcame the difficulty and were able to play

any song that would appear in the playlist.

2.2.2 Musicbrainz

MusicBrainz's extensive research shows that it

contains a lot of songs’ metadata but after a long

investigation in MusicBrainz API

(https://musicbrainz.org/doc/Indexed_Search_Synta

x#Overview) there were many difficulties in

searching for specific information such as the first

publishing date of a song by year and by its original

geographical area. A song's metadata on MusicBrainz

contains the year of release and in which country,

based on the year and location the album was

released. But there could be more than one album

containing the same song, each album with a different

publishing date or location. So, according to many

tests, and verification of Google data on songs, we

realized that the best way to get the most accurate

information was to take the first year of publication

listed.

MusicBrainz does not contain a popular rating,

which is very important to our recommendation tool;

the popularity data is meanwhile taken from the

YouTube using the Video ID that was fetched from

MusicBrainz. We found that MusicBranz API does

not fit such massive search and retrieval. One

connection session allows, before terminating, not

more than 16 requests to the MusicBranz server; each

request can fetch up to 100 songs, which adds up to

the retrieval of 1600 songs' metadata in one

connection.

In order to get the metadata of several hundred

thousand songs published in a certain and

geographical location, we downloaded and stored all

the information about the songs from 1880 to 2018

(121,348 Records). The data was saved in JSON files

containing 100 songs in each file, and stored in our

MongoDB server with the information that was

relevant to us. Additional information like the song's

rating or views, and the ability to play it, was fetched

and stored from other repositories.

2.2.3 Last.fm

Like MusicBrainz, Last.fm (https://www.last.fm)

displays information about songs from different

periods. Unlike MusicBrainz, the Last.fm repository

HEALTHINF 2020 - 13th International Conference on Health Informatics

432

Figure 4: System Architecture.

shows a range of songs using the API for a certain

period of time, but a manual check for the parade of

songs we received found that the songs are out of date

and do not show the hits for that period. A thorough

examination of various forums brought to our

attention that the database has not been updated since

2015. It was decided to abandon it and continue

research.

2.2.4

Discogs

We conducted another study in the Discogs

(https://www.discogs.com) repository. Discogs

provides a lot of information about songs, such as

year of publication and information about various

record companies. Discogs did not fit our need

because of several reasons that disqualified its use.

The site is mainly focused on electronic music and

does not fit the songs we were seeking for elderly

people. The information in Discogs is categorized by

complete albums rather than specific songs. Finally,

the information is not categorized according to

geographical location, which is very critical to this

work.

2.3 Music Recommendation

A prototype of the recommender application is

depicted in Figure 3. The system is composed of a

web-based and mobile application that displays a

recommended playlist that is customized to the

profile and needs of the end-user. The person’s

reactions to this playlist will be monitored, stored and

used to improve it. The application is illustrated in the

upper left-hand box of Figure 3. The system provides

monitoring of the person by the medical staff filling

out a report regarding their psycho-social behaviour

before and after the music session. A scalable

ecosystem (illustrated in the lower box of Figure 3),

learns to match a list of songs, filtered by tags, to a

certain profile of an elderly person.

The recommendation filter determines the

makeup of the playlist created as a result of this

matching process. The system has to build a new

dataset for each type of recommendation algorithm

and continuously analyse the data for more updates.

Storage of the vast amounts of data around the world

warrants the use of big data technology for scalable

and reliable systems. As stated in 2.1, an old person’s

profile is characterized by their details such as: year

of birth, country of origin and language. The system

calculates the years when the person was 15-25 years

old. The recommendation algorithms and playlist

composition work in two stages as follows:

Stage 1:

The population (for example all the residents of a

long-term care facility) are clustered together into one

group by country of origin, language and year of

birth. That is, all participants who were born in

Arabic-speaking countries in 1940 and whose

language is Arabic (in addition to Hebrew) were

under the same group. The cosine similarity distance

is used in order to set similar groups.

Each group of residents having similar parameters

is assigned initially to the same playlist. A playlist of

25 songs with the highest rating by year, country of

origin and the song's languages will be extracted from

the database. Currently, our main load of data was

retrieved from MusicBrainz and contains information

on each song including year of release, country of

origin and language. The play popularity rating is

determined by relevant information on YouTube.

Stage 2:

The first time a person logs into the system, the

recommender filter (Figure 3 on the right) randomly

displays 10 default video songs from the playlist

(containing 25 songs) using the YouTube ID that is

kept in the internal database. It is important to note

that the video is displayed within the framework of

our application and is not transferred to the YouTube

environment. Thus, we have full control over the

content that is displayed to the old person. In addition

to the video of a song, the song’s name and the

Music Recommendation System for Old People with Dementia and Other Age-related Conditions

433

performer's name will be displayed. The software

uses YouTube API to fetch the video by its YouTube

ID and display it.

The old person himself (or the instructor

according to the person's reactions) is able to rate the

songs on a scale of 1-5. 5 - songs he liked / knew, 1 -

didn't like / didn't know. The rating is kept

in the

internal database to be used the next time this person

enters the system, in order to improve his playlist.

When the user ranks the list of songs presented to

him, the recommender filter uses the collaborative

filtering with the cosine similarity as the distance in

order to calculate the similarity between this person’s

preferences and the rest of the group. The next time

this person logs in the system the playlist is refreshed

and displays songs that he liked in the past along with

songs that others in his cluster group liked.

2.3.1 Example of 1970 Playlist for People

from the UK

The following example presents the recommendation

process for three 70-year-old people, Emma, John and

Sam, who speak English and came from the UK. (The

presented names are pseudonyms). All of them were

assigned the same playlist with 25 songs; it is listed

in column 1 and 2 (Table 1). This playlist was created

using our clustering mechanism as explained before.

When Emma logged into the system for her first

time, 10 random songs were presented to her, one

after the other, and Emma ranked them according to

her taste. The order of the presented songs to Emma

with her rank score is presented, from the top

downwards, in the third column of Table 1. Columns

4 and 5 present the top downwards order of songs

presented to John and Sam respectively and the ranks

they allotted to each song.

The preference vectors of each one of the users

were updated with the given ranks. The similarity

between each pair of users was calculated using the

collaborative filtering with similarity distance; The

similarity distance we have used takes into account

that one person may consider 4 as a very high rate and

another considers 5 as a high score. It was found that

preferences of Emma and John are more similar than

the others. Thus, the second time Emma logged into

the system, her playlist was updated as can be seen

below (Table 2). The four first songs are those that

she liked more than the others. The next four songs

are those that John liked and the last two were new

songs from the initial playlist.

Table 1: Playlist display for Emma, John and Sam at their

HEALTHINF 2020 - 13th International Conference on Health Informatics

434

Table 2: Emma's updated playlist presented to her on her

3 SYSTEM PROTOYTPE

ARCHITECTURE

The architecture of the system is depicted in Figure 4.

The client side was developed in Angular. The server

functionalities were developed in NodeJS and include

the recommendation mechanisms and the interaction

with repositories APIs and database servers. The

internal storage for the metadata of the songs is

MongoDB. The songs’ metadata was downloaded

and updated from MusicBrainz and YouTube, as

explained in Section 2.2.2, into MongoDB in JSON

format which is the most prevalent format for

serializing content (Humphrey et al., 2014).

MongoDB was selected because of its fitness with

JSON in terms of document creation and indexing.

MongoDB has replication capabilities and works well

for scalable big data storage, as is required in our

system.

4 PRELIMINARY PILOT USING

TAMARINGA

A preliminary pilot enabled 24 residents in an assisted

living facility to try and use the system described

above over a period of three months. The activity was

suggested to the residents as part of other activities

offered in their assisted living facility. Residents were

free to access the computer room and were also

invited by staff members to try the system. The

instructor helped each resident to operate and

formulate their customized content. The residents

who participated included those with dementia, those

who are wheelchair-bound with/without cognitive

decline, and those who have other age-related

conditions, such as stroke survivors. Some residents

had a weekly meeting in the computer room, some

asked to attend few times a week, and most of the

residents with dementia, when asked, expressed their

wish to attend again.

For three months, the resident entered the

computer room regularly on certain days at certain

hours or by his/her request and was presented with the

customized playlist (as described in Section 2.2).

During the sample period, the professional staff

observed the behaviour of the residents and gave their

impressions regarding changes in the residents'

psycho-social behaviour. Though no systematic data

was gathered.

According to the staff report, it seemed that all the

residents shared their enjoyment with others after the

experience. They told their friends, staff members,

and family about the content they watched. Some

residents continued singing their preferred song

afterwards, some stated that watching their preferred

songs helped them to forget their worries. The

instructor in the computer room documented

moments of joy, laughter, and singing. Pivotal

moments were with residents who had previously

refused to participate in any suggested activity, and

after realizing they could watch their preferred

content, kept asking to attend. Further investigation

using validated assessment tools will help to

understand the impact of using customized content

via the suggested system.

5 RELATED WORK

In addressing a disease that destroys memory,

preserved musical memory serves as an important

tool for enhancing quality of life (Baird & Samson,

2015). Individualized music evokes memories,

despite memory loss, particularly by means of songs

associated with the early decades of seniors’ lives

(Dassa & Amir, 2014). The theory-based intervention

of individualized music has been evaluated clinically

and empirically and was found to be beneficial in

reducing anxiety and agitation, eliciting memories

Music Recommendation System for Old People with Dementia and Other Age-related Conditions

435

and in promoting communication among people with

dementia (Gerdner, 2012).

A communication barrier is most evident in the

case of dementia. In an exploratory research case

study, a personalized database was formed with the

help of spouses who visit their partners with dementia

in a long-term care facility. The database included

preferred music and personal photos and helped to

evoke a reaction and facilitate communication

between the couple. Although the process of

preparing the data was very powerful for the spouses,

it demanded extensive preparation. The main

challenge is to create a feasible procedure that will

allow caregivers to have the use of a personalized

database (Dassa, 2018).

Old people are having a lot of difficulties

accessing and operating modern applications

(Zajicek, 2006; Sayago & Blat, 2009) and our vision

is to create a suitable interface for the elderly, yet not

to exclude communication with another person who

will guide and accompany this process. The design of

Tamaringa application follows this vision.

Music Recommendation aims to suggest suitable

music to users by inferring their music preferences.

Different kinds of ancillary information have been

applied to boost recommendation accuracy. A typical

example is to add temporal dynamics and music

taxonomy bias (i.e., artist, album, and genre)

(Koenigstein et al., 2011). Another research study

explored how the usage of the user’s demographic

information in collaborative filtering and additional

user's characteristics improve the tailoring of a more

suitable playlist (Schedl et al., 2015; Schedl &

Hauger, 2015).

In recent years, more research efforts have been

devoted to explore user-related contexts in music

recommendation. Music popularity trends and user’s

current location context were taken into consideration

to facilitate personalized music recommendation

(Cheng & Shen, 2014). Several works (Koenigstein

& Shavitt, 2009

; Schedl et al., 2010; Hauger &

Schedl, 2012)

considered the popularity of a

performer or a music piece highly relevant,

specifically in order to estimate the popularity of

music releases and promising artists. For this purpose,

different data sources have been investigated: search

engine page counts, microblogging activity, query

logs and shared folders of peer-to-peer networks, and

play counts of Last.fm users.

6 CONCLUSIONS

Conventional medical treatment does not counteract

the progression of the course of deterioration in the

case of dementia, nor does it help recall those

memories thought to be lost. Pharmacological

solutions also have a limited effect on the symptoms.

Personalized music sessions can help memories re-

emerge and revitalize the spirit of those with

dementia. Music was proved to reduce behavioural

symptoms and elicit physical activity. Our Tamaringa

system prepares customized media and music

playlists in order to improve the daily lives of people

with dementia, those with other age-related

conditions, and their caregivers. For this purpose, we

explored several music repositories for their

suitability to retrieve relevant metadata into an

internal database of songs suitable for

recommendation.

The music recommendation process, and the

architecture of our system, is different than the work

that was done so far in several aspects. The starting

point for the recommendation is based on the

characteristics of the old person, such as birth year

and place, followed by the music search and filter.

The information about the songs cannot be found in

an online social media and should be retrieved from

static repositories. Finally, the popularity back in the

years the songs were released cannot be retrieved by

the same mechanism we have today. Hence, in order

to be able to recommend correctly the suitable songs

for the old person we had to include the metadata of

these songs in our internal database. We explored

several repositories and decided to use the

MusicBrainz data and API.

Our next step will be to conduct a full assessment

scale research that will explore the benefits and

impact of our system on specific socio-behavioural

symptoms among people with dementia and other

age-related conditions. We also aim to further

develop our system to match the necessary

requirements in the field of gerontology. We believe

that adding the use of personalized music using this

system could benefit well-being and promote

communication between people with dementia and

their caregivers in assisted living facilities, and at

home.

REFERENCES

Airoldi, M., Beraldo, D. & Gandini, A. (2016). Follow the

algorithm: An exploratory investigation of music on

YouTube. Poetics 57: 1–13.

HEALTHINF 2020 - 13th International Conference on Health Informatics

436

Baird, A., & Thompson, W. F. (2018). The impact of music

on the self in dementia. Journal of Alzheimer’s Disease,

61(3), 827-841. DOI: 10.3233/JAD-170737

Baird A., & Samson, S. (2009). Music and dementia.

Progress in Brain Research, 217:207-235. doi:

10.1016/bs.pbr.2014.11.028

Brodsky, J., Shnoor, Y., Beer, S., & Neville, A. (2018). The

65+ population in Israel. Myers-JDC-Brookdale,

Jerusalem: Israel.

Cheng, Z., Shen, J., & Mei, T. (2014). Just-for-Me: An

adaptive personalization system for location-aware

social music recommendation. In SIGIR 2014.

Cohen, A., Bailey, B., & Nilsson. T. (2002). The

importance of music to seniors. Psychomusicology, 18,

89–102.

Dassa, A. & Amir, D. (2014). The role of singing familiar

songs in encouraging conversation among people with

middle to late stage Alzheimer's disease. Journal of

Music Therapy, 51(2), 131-153.

Eriksson, M., Fleischer, R., Johansson, A., Snickars, P., &

Vonderau, P. (2019). Spotify Teardown: Inside the

black box of streaming music. The MIT Press.

Finamore, A., Mellia, M., Munafò, M., Torres, R. & Rao,

S. G. (2011). YouTube everywhere: Impact of device

and infrastructure synergies on user experience.

In IMC '11 2011.

Gerdner L. A. (2012). Individualized music for dementia:

Evolution and application of evidence-based protocol .

World Journal of Psychiatry, 2 26-32(2)

Hauger, D. & Schedl, M. (2012). Exploring geospatial

music listening patterns in microblog data. In

Proceedings of the 10th International Workshop on

Adaptive Multimedia Retrieval AMR 2012.

Humphrey, H. J., Salamon, J., Nieto, O., Forsyth, J.,

Bittner, R. B., & Bello, J. P. (2014). JAMS: A JSON

Annotated Music Specification for Reproducible MIR

Research. In Proceedings of the 15th International

Society for Music Information Retrieval Conference

(pp. 591–596).

Koenigstein, N. & Shavitt, Y. (2009) Song Ranking Based

on Piracy in Peer-to-Peer Networks. In ISMIR 2009.

Koenigstein, N., Dror, G. & Koren, Y. Yahoo! music

recommendations: modeling music ratings with

temporal dynamics and item taxonomy. In ACM

RecSys, 2011.

Levingston, G., Kelly, L., Lewis-Holmes, E., Baio, G.,

Morris, S., Patel, N. et al.(2014). Non-pharmacological

interventions for agitation in dementia: Systematic

review of randomised controlled trials. The British

Journal of Psychiatry: The Journal of Mental Science,

205(6), 436-442.

Prince, M., Gemma-Claire, A., Guerchet, M., Prina, M. A.,

Albanese, E., & Wu, Y. T. (2016). Recent global trends

in the prevalence and incidence of dementia, and

survival with dementia. Alzheimer's Research &

Therapy 8:23 DOI 10.1186/s13195-016-0188-8

Ray K., & Fitzsimmons, S. (2014). Music-assisted bathing:

Making shower time easier for people with dementia.

Journal of Gerontological Nursing, 40(2), 9-13.

Särkämö, T., Tervaniemi, M., Laitinen, S., Numminen, A.,

Kurki, M., Johnson, J. K., & Rantanen (2014).

Cognitive, emotional, and social benefits of regular

musical activities in early dementia: Randomized

controlled study. The Gerontologist, 54(4), 634-650

421 doi:10.1093/geront/gnt100.

Sayago, S. & Blat, J. 2009. About the relevance of

accessibility barriers in the everyday interactions of

older people with the web. In Proceedings of the 2009

International Cross-Disciplinary Conference on Web

Accessibility, W4A.

Schedl, M., Pohle, T., Koenigstein N. & Knees, P. (2010)

What’s Hot? Estimating Country-Specific Artist

Popularity. In Proceedings of the 11th International

Society for Music Information Retrieval

Conference,ISMIR 2010.

Schedl, M., G´omez, E., & Urbano, J. (2014) Music

information retrieval: Recent developments and

applications. Foundations and Trends in Information

Retrieval, 8(2–3):127–261.

Schedl, M. & Hauger, D. Farrahi, K., & Tkalciˇ c, M.

(2015). On the influence of user characteristics on

music recommendation algorithms. In ECIR, 2015.

Schedl, M. & Hauger, D. (2015). Tailoring Music

Recommendations to Users by Considering Diversity,

Mainstreaminess, and Novelty. In SIGIR '15 2015.

Thomas, K. S., Baier, R., Kosar, C., Ogarek, J., Trepman,

A., & Mor, V. (2017). Individualized music program is

associated with improved outcomes for U.S. nursing

home residents with dementia. American Journal of

Geriatric Psychiatry, 25(9), 931-938.

doi.org/10.1016/j.jagp.2017.04.008

World Health Association (2017). Global action plan on

the public health response to dementia 2017-2025.

Retrieved from https://www.who.int/mental_

health/neurology/dementia/action_plan_2017_2025/en

Zajicek, M., Aspects of HCI research for older people.

Universal Access in Information Society, 2006. 5: p.

279- 286.

Ziv, N., Granot, A., Hai, S., Dassa, A, & Haimov, I. (2007).

The effect of background stimulative music on

behaviour in Alzheimer's patients. Journal of Music

Therapy 44(4), 329-343.

Music Recommendation System for Old People with Dementia and Other Age-related Conditions

437