SousChef: Mobile Meal Recommender System

for Older Adults

David Ribeiro

, João Machado

, Jorge Ribeiro

, Maria João M. Vasconcelos

Elsa F. Vieira

and Ana Correia de Barros

Fraunhofer Portugal AICOS, Rua Alfredo Allen 455/461, 4200-165 Porto, Portugal

REQUIMTE, Institute of Engineering of Porto, Polytechnic Institute of Porto, Porto, Portugal

CRPG - Gaia Vocational Rehabilitation Centre, Vila Nova de Gaia, Portugal

Keywords: Recommender System, Nutrition, Intelligent Companions, Mobile Health Monitoring Older Adults,

Personalized Interfaces.

Abstract: Nowadays, following a healthy diet is a challenge, either due to the large variety of food and ingredient

combination possibilities or due to the lack of knowledge required to make healthy choices. This problem is

even more patent amongst older adults. Although some recommender systems and applications have been

proposed with aim to monitor calorie consumption and/or to suggest healthy recipes to general consumers,

no similar solution was yet presented focused on older adults’ needs. In this work, a mobile meal

recommender system, SousChef, for this target group is presented. This system is capable of creating a

personalized meal plan based on the information provided by the user, including the anthropometric

measures, personal preferences and activity level. The nutritional recommendations and the application was

thought and designed for older adults, presenting friendly user interfaces and following the guidelines of a

nutritionist. Tests with users were conducted in order to ascertain recipe and nutritional plan suitability as

well as usability of the mobile application. Results showed that more than 70% of the older adult

participants were satisfied with the meal plan suggestions and with the simplicity of the SousChef

application.

1 INTRODUCTION

According to the literature improved nutrition is a

major cause of increased lifespan in the last two

centuries (Le Couteur et al., 2016; Bunker, 2001).

Poor diet, in contrast, is the main risk factor for

death and disabilities in developed nations (Murray

et al., 2013). Although the exact definition of elderly

age group is controversial, for high-resourced

countries the World Health Organization have

accepted the chronological age of 65 years as a

definition of 'elderly' or older person (World Health

Organization, 2010). An ageing population tends to

have a higher prevalence of chronic diseases,

physical disabilities, mental illnesses and other co-

morbidities (Saka et al., 2010). Thus, in order to

promote long-term biological effects, continuing

efforts to increase the relevance and effectiveness of

nutritional recommendations have been made

(World Health Organization, 2002). When looking

at older demographics, several studies refer that

older adults often struggle with making the right

decisions regarding meal preparation, healthy diets

or groceries shopping. Studies also suggest that

many older adults neglect nutrition and are more

inclined to do so if they happen to live alone (Ramic

et al., 2011). Furthermore, under financial

restrictions, which older adults often find themselves

in, balancing healthy eating habits with money

saving can become a complicated task (World

Health Organization, 2002; Ministry of Health,

2013).

Taking into account the previous facts and

acknowledging that nowadays technologies are

always present and may be used to assist people, in

this work we present SousChef, a mobile meal

recommender system to assist older adults by

providing a nutrition companion to guide them into

making wise decisions regarding food management

and healthy eating habits.

The meal recommender system was developed to

create personalized nutritional plans according to the

Ribeiro, D., Machado, J., Ribeiro, J., Vasconcelos, M., Vieira, E. and Barros, A.

SousChef: Mobile Meal Recommender System for Older Adults.

DOI: 10.5220/0006281900360045

In Proceedings of the 3rd International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2017), pages 36-45

ISBN: 978-989-758-251-6

information provided by the user, namely, the

personal preferences, activity level and the

anthropometric measurements (weight and height).

The system intelligence was built with the

professional knowledge and participation of a

nutritionist, who formulated the diet plans and

healthy meals suggestions, and took part in the final

system validation.

The present paper is organized as follows:

section 2 presents the related work; in section 3 the

system is described while in section 4 the mobile

application is presented; in section 5, the user testing

results are shown; finally, conclusions and future

work are drawn in section 6.

2 RELATED WORK

Regarding mobile technologies related to food and

nutrition, several studies may be found in the

literature addressing issues such as: recommender

systems (Aberg, 2006; Mika, 2011; Sezgin and

Ozkan, 2013; Hazman and Idrees, 2015; Espín et al.,

2015), social interaction (Terrenghi et al., 2007),

menu generation (Kuo et al., 2012) or cooking

assistance for users with specific impairments, such

as in memory (Tran et al., 2005) or language (Tee, et

al., 2005). As this work presents a recommender

system, we will briefly detail the related work in this

topic.

In (Adomavicius and Tuzhilin, 2005), the authors

performed a survey of the State-of-the-Art on

recommender systems and identified three main

types of systems based on the employed

methodologies: content-based, collaborative and

hybrid recommender systems. The challenges

related to the design and implementation of

nutritional recommender systems are discussed in

(Mika, 2011). This author identifies the uncertainty

of nutritional information of recipes or foods, or the

missing or incorrect data from food recording

measurements as the main challenges and suggests

ways to tackle them. In (Sezgin and Ozkan, 2013) a

literature review on Health Recommender Systems

(HRS) is presented: studies have demonstrated that

HRS have branched out in different fields of health

industry and HRS applications have been

increasingly embedded in the health service systems.

Challenges and opportunities in HRS are also

addressed in this paper.

In (Åberg, 2006) the author has focused

specifically on older adults and their nutritional

needs, designing a recommender system with user

interfaces designed to consider the specific needs of

the user group. The recommendations generated by

this system are based on parameters that go beyond

nutritional needs, such as taste or available food

items at the person’s home. However, the

description of the system says nothing about recipes

prepared specifically by dieticians, it did not allow

multiple users and it was not designed to

accommodate specific medical conditions.

Moreover, the user interface, given the time of that

work, does not consider current mobile contexts of

use.

In (Hazman and Idrees, 2015), a prototype for a

healthy nutrition expert system for children is

proposed that considers all stages of the child, their

growth stage, gender and health status. A case study

is presented and a web application was developed

however, the validation of the knowledge for the

proposed system is still needed.

Recently, (Espín et al., 2015) presented

NutElcare, a semantic recommender system that

provides healthy diet plans for the older adults. It

claims to retrieve reliable and complete information

from expert sources as nutritionists, gerontologists

as well as knowledge from information systems and

nutritional databases and with that information aim

to assist older people to take advantage of these tips

and make their own diet plans.

Regarding commercially available ICT

applications designed for non-professionals, there

are different options, such as EatThisMuch

MyFitnessPall

, Lose It!

, Lifesum

, or Nutrino

Among other features, in general all of these

applications offer the ability to monitor calories

consumption based on the manual input of food

from a database. Other features can be provided,

such as step-by-step guidance to prepare meals, a

shopping list or the use of social interactions through

social networks and gamification. Among these

applications, only Nutrino and EatThisMuch are

capable of creating personalized meals plan.

The mobile meal recommender system that we

are presenting could be a good solution for older

people, providing them healthy and personalized

dietary plans, which are suitable to their individual

requirements (based on age, sex, activity level,

weight and height) and dietary preferences.

SousChef system is specifically addressed to this

https://www.eatthismuch.com/

https://www.myfitnesspal.com/

https://www.loseit.com/

https://lifesum.com/

https://nutrino.co/

SousChef: Mobile Meal Recommender System for Older Adults

target group because consider the nutrient intake

recommendations for older persons (age > 65)

regarding (1) Energy: 1.4-1.8 multiples of the basal

metabolic rate to maintain body weight at different

levels of physical activity; (2) Protein intake of 0.9-

1.1 g/kg per day, the equivalent of 15-20% of the

daily energy; (3) Lipids intake of 30% of the daily

energy in sedentary older persons and 35% for

active older persons; consumption of saturated fats

should be minimized and not exceed 8% of energy,

and (4) Carbohydrates intake of 55 to 60% of the

daily energy (World Health Organization, 2002).

Moreover, dietary recommendations using the

“food-based dietary guideline approach” (Wahlqvist,

2002) were also taken into account. SousChef

system can be a remarkable advantageous system

compared to other applications available in the

market.

3 SYSTEM OVERVIEW

The SousChef system is composed of a central cloud

server and a mobile application which is the user

interface for the system. The cloud server is

responsible for centrally storing the information of

the system and making it accessible through web

service application programming interfaces (APIs).

Its easy accessibility also facilitates the integration

of information from other sources, which is

demonstrated through the integration with Fitbit

cloud services to retrieve users’ activity

measurements measured by Fitbit

devices.

Given the computational demands for the

creation of meal recommendation, the superior

processing capability of the server when compared

to mobile devices also makes it a more suitable

option. The generation of recommendations can be

triggered by the mobile devices through web service

APIs.

The recommendations are created considering

information from different sources: personal

information provided using the user interface,

activity data through Fitbit devices and nutritional

information from the food composition database.

Work performed on top of the database will be

described in the following subsection.

http://www.fitbit.com/eu/home

3.1 Food and Recipe Database

The development of the SousChef system was based

on the Portuguese Food Composition database

elaborated by INSA (Ministério da Saúde, 2006).

This database contains the nutritional composition of

over a thousand products.

For the purpose of the system, the ingredients

that are not suitable for direct consumption (e.g. raw

chicken meat) and the specific ingredients (e.g. salty

chips) which are not considered healthy for the

target group were selectively removed by the

nutritionist. Moreover, combinations of recipes and

ingredients were also created by the nutritionist to

ensure that SousChef’s recommendations are not

only suitable for every meal of the day and for the

target population, but also culturally acceptable. In

this system context, a recipe does not refer to the

instructions to prepare the meal, but rather to a

combination of ingredients and respective quantities.

Multiple combinations were created for each meal of

the day: 40 combinations for breakfast, mid-morning

and mid- afternoon snacks; 20 combinations for

supper and 340 combinations for main dishes. Each

combination for a meal comprises the list of

ingredients and respective quantities based on a

daily intake of 2000 kcal.

3.2 Recommender System

The implemented recommender system is a content-

based recommender system and mainly employs

information retrieval techniques (Adomavicius and

Tuzhilin, 2005). In order to create a personalized

weekly meal plan, there are three main steps:

calculation of nutritional requirements, selection of

food items for each meal and scaling the meals to

match the user's caloric needs (Figure 1). The

algorithm was designed in collaboration with a

nutritionist.

The Estimated Energy Requirement (EER) is

calculated from the predictive Harris-Benedict

equations (Long et al., 1979) equations used to

estimate the Basal Metabolic Rate (BMR) by

multiplying with the physical activity level (PAL)

(National Research Council, 1989). BMR is

calculated based on a user's energy expenditure, age,

sex, weight, height and PAL is calculated based on

the Total Energy Expenditure measured using Fitbit

devices. The daily caloric need is then distributed

across meals as follows: 15% for each breakfast,

middle morning and middle afternoon snack meals;

25% for each lunch and dinner and 5% for supper

(Candeias et al., 2005).

ICT4AWE 2017 - 3rd International Conference on Information and Communication Technologies for Ageing Well and e-Health

Figure 1: Overview of Meal Plan Recommender System.

After estimating the user’s caloric needs, the next

step consists in selecting the most appropriate

ingredient combinations for each meal considering a

given context. The meal planning context consists of

the user’s personal information, including nutritional

needs and food preferences, as well as meal planning

history (in order to enable dietary diversity in the

resulting plan).

Considering the previously created

combinations as candidates, for each meal of the

week being planned, a two phase selection process

will determine the most suitable candidate for that

meal. The first phase consists of applying restriction

rules, which filters candidates that are not suitable

for a given meal and context. Multiple rules have

been implemented for removing candidates: limiting

repeated recipes in the same week by removing

candidates which have been used in planning more

than twice; removing any candidates that include

comfort food; filtering candidates that include

ingredients users are allergic to.

The next phase consists in selecting the most

suitable candidate from the ones that were not

filtered. This is performed by calculating a score S

for each candidate c according to different criteria

measured by different heuristic functions H

. Each

heuristic function evaluates the meal planning

context, which meal is being planned and a single

candidate and assigns a suitability score between 0.0

and 1.0 to each candidate. Different heuristic

functions have been implemented to reflect the

criteria identified by the nutritionist. One of them

consists in favouring main dish combinations with

the same soup in four consecutive meals (about one

litre). The idea behind this criteria is allowing users

to cook soup for several days. The heuristic function

checks if the series of one litre was met. If so the

same value (0.5) is assigned to every candidate.

Otherwise, 0 is assigned to candidates with a

different soup and 1 to the others. Another criteria is

to favour meat dishes for lunch and fish for dinner in

order to have lighter dinners. For main dishes, the

candidates meeting the criteria are assigned the score

1 and 0 to the others. All candidates for other meals

are assigned 0.5. Finally, another heuristic takes the

food preferences of users into consideration. Using

the application, users are able to provide ratings to

ingredients from 0 to 4, which are normalized into a

value between 0 and 1. The preference for a

candidate is calculated by combining the preference

for each of its composing ingredients whenever

available or the value 0.5 instead. The final score, S,

for a candidate is also calculated as the average of

the values calculated by each heuristic function.

The chosen candidate for each meal being

planned is the one with highest suitability score S.

The approach followed by this algorithm benefits the

scalability of the system, since it facilitates the

inclusion of new restrictions and heuristics to

consider new criteria and data to provide users with

better recommendations. It also enables in the future

to use different weights to each heuristic function for

each user, personalizing the recommendations even

further.

Once the ingredient combination has been

chosen for each meal, the final step of the planning

process consists in scaling the ingredient quantities

in order to match the energy requirements of that

particular individual. The previously calculated

caloric requirements for a given meal are compared

with the total energy of the ingredients in the chosen

candidate’s plan. If the difference is higher than an

acceptable deviation, the quantity of the ingredients

will be scaled to suit the requirements. To ensure

that daily nutrient requirements are maintained, only

ingredients from specific categories will be scaled:

cereals, fruit and legumes, dairy products, meat, fish,

eggs and oil. The new quantity for each scalable

ingredient is then calculated using a weighted

average so the quantity of more caloric ingredients

change more than others, therefore reducing the

changed amount in grams of the overall

combination.

4 MOBILE APPLICATION

To make the interaction between the food

recommender system and the end-user possible, a

mobile application was designed. The current

version of the mobile application is organized in

three different components: the meal plan, the

grocery list, and the activity monitoring.

Supporting these different components, there is a

profile that aggregates the various settings and

preferences used to tune the system. The profile

contains anthropometric data, food-related

SousChef: Mobile Meal Recommender System for Older Adults

preferences, and activity profile. Regarding the

food-related preferences, two levels of control are

provided: food restrictions, and dietary

considerations. Food restrictions are hard constraints

that remove food from the recommendation engine.

That is, some food added to the food restrictions list

will not be returned in the meal recommendation.

Dietary considerations work with a similar principle,

but on a group level, therefore, restricting an entire

set of products, and consist of a predefined list of

diets that can be followed by the user (e.g.

vegetarian, lactose free). For instance, if the lactose

free diet is selected all the products with lactose will

be removed.

4.1 Meal Plan

The Meal Plan section is the centre of the entire

system from a technical perspective but also from

the point of view of the user. The Meal Plan fulfils

two main goals: to generate new meal plans, and to

track the user’s food intake. In the application, the

Meal Plan section is responsible for presenting all

the information related to nutrition.

The Meal Plan (Figure 2) is the place to record

everything that the user eats throughout the day.

There are two different ways to log information into

the food diary: 1) Automatic food recommendations;

2) Manual input.

Automatic food recommendation: The

fundamental feature of this system is the ability to

generate personalized meals plan according to the

needs of the user.

Manual input: Although the system is in theory

capable of generating an entire meal plan, in

practice, the recommendations generated by the

system might not be enough to cover all the different

scenarios involved in a typical diet. To address this

issue, it is possible to manually record additional

information in the diary. New entries can be added

based on a search query on the food database that

also powers the food recommender engine.

To complement the food diary, which provides a

quick overview of the user’s diet, each recipe or

ingredient has a dedicated view to display additional

information and actions (Figure 3). The nutritional

information of a product is one of the aspects that

are provided in this user interface. Besides the

energy value in calories, there is also the macro and

micro nutrients information. Basic mechanisms for

editing the product and the diary (for example,

removing products form the diary) are also provided.

The user is able to control the recommender engine

by removing ingredients from the list of approved

products. In this case, products added to that list are

entirely ignored by the system. However, the control

level is rather limited since users are only able to

make binary decisions. In order to provide the user

with additional control of the system, another feature

particularly important in terms of personalization is

the ability to personally rate the products both

recipes and ingredients. Whereas in the former case

the system works with hard constraints that remove

products from the recommender engine entirely, in

the latter case the user input is used to influence the

weight of a product in the engine. The rating system

used in this evaluation contains three different

values, which are used to adjust the weight of the

product when a new plan is generated. From the

perspective of the user, this translates to (Figure 3,

right): 1) “I don’t like it that much”, 2) “I like it”,

Figure 2: Meal Plan.

Figure 3: Ingredient view.

and 3) “I love it”. For instance, a product rated with

the first score will not stop being recommended to

ICT4AWE 2017 - 3rd International Conference on Information and Communication Technologies for Ageing Well and e-Health

the user, but since it has a lower weight, it will be

recommended less often.

4.2 Grocery List

Although the Grocery List is not the focus of the

SousChef system, it can nevertheless be a valuable

tool in the above mentioned goal to assist food

management. The goal of the Grocery List

component is to help the user with shopping related

activities. This component takes on a physical

grocery list, but augments it with new capabilities

(Figure 4). One of the main features is the

integration with the Meal Plan component and the

ability to automatically add products to the list based

on the ingredients of a recipe. There is also the

option to manually add new products to the list

based on a database search.

Since the goal of the Grocery List is to assist

users in their shopping activities, there was the

concern to design a user interface that would be easy

to use in the wild, e.g. in the supermarket. Therefore,

one characteristic of the system is the ability to work

offline, so that the user does not need an Internet

connection to access his or her grocery list while in

the supermarket. Users are also able to mark

products as bought as they go, and that information

will be synchronized the next time there is an

Internet connection. Moreover, products are

organized by aisles to make the shopping process in

the store more efficient.

Figure 4: Grocery List.

4.3 Activity Monitoring

Along with nutrition, an important part of a healthy

lifestyle is a person’s activity level. The Activity

Monitoring used in the SousChef system presents

the users with metrics related to their activity

(Figure 5). The system is able to collect and display

calories burned, steps and active time. For each of

these metrics, daily objectives can be set in order to

raise awareness to their healthy physical activity

habits.

Given that one of the inputs required by the

system in order to generate new meals is the level of

activity of the user, the Activity Monitoring

component is responsible for collecting user activity

data and feeding those data to the recommender

engine. In order to do that the system can be

connected to an activity monitoring device such as a

Fitbit bracelet. The information gathered from such

devices is then converted to a scale that classifies the

level of activity as belonging to one of five levels,

ranging from “sedentary” to “extremely active”. The

advantage of using a wearable device such as Fitbit

is that the user activity information can be collected

seamlessly, without direct input, with a reasonable

level of accuracy, and updated automatically to our

system. However, in order to free the system from

dependence on external devices or systems, it is also

possible to manually insert this information in the

user profile. In such scenario, the charts with the

user activity information would not be used.

Figure 5: Activity monitoring.

5 USER TESTING

5.1 Algorithm Validation

In this section the results of the tests performed with

older adults regarding the recommender system and

the mobile application itself are presented and

discussed. For both tests, the project was explained

to the subjects and informed consents were obtained.

SousChef: Mobile Meal Recommender System for Older Adults

The questionnaires and protocols were designed to

be applied to older people.

In order to ascertain user’s quality perception of

the meal recommender system, a first set of tests

were performed with 16 subjects, 8 men and 8

women, with a mean age of 70 ± 4.2 years, mean

height of 1.64 ± 0.07 m and a mean weight of 76 ±

12 Kg. Besides the anthropometric measures to

insert in the system, participants were also asked

about their preferences concerning ingredients and

their activity level. This information was then

submitted to the system in order to generate a

personalized weekly plan composed of six meals per

day. Afterwards, participants were asked questions

regarding the recommended meals, daily plans and

the weekly plan itself, which they were asked to

evaluate on a 4 point Likert scale, ranging from 1

(Strongly disagree) to 4 (Strongly agree).

For each meal, participants were asked whether

they could eat the suggested recipe and if they

considered the recipe to be adequate for that meal.

Regarding the daily plans, participants were asked if

they felt the daily plan was adequate for them.

Finally, for the weekly plan, participants were asked:

if they would follow the generated weekly plan; if

they would use the application to create a meal plan,

should this application be available to them; to rate

the sentence “I do not like the system, therefore I

would not use it”. It should be noted that the last two

questions are actually the inverse of each other. This

was done in order to minimize biases in the subjects’

personal perception of the questions. In addition to

these questions, the subjects were asked, by

answering yes/no, if the generated weekly plan was

good for them. Furthermore, throughout the

questionnaire, the subjects’ comments and

suggestions were also annotated.

Regarding the results for meals, the subjects

were, in general, agreeable with the recommended

meals. Concerning the case if they could eat the

suggested meal 60.86% of the subjects strongly

agreed, 26.04% agreed, 8.48% disagreed and 4.62%

strongly disagreed. To the question about recipe

adequacy to the meal 52.98% of the subjects

strongly agreed, 22.02% agreed, 20.69% disagreed

and 4.31% strongly disagreed. In Figures 6 and 7,

the answers are grouped by meal type.

Considering Figure 6, it can be seen that

participants felt that the recipes for mid-afternoon

and mid-morning meals were not as conforming to

their dietary restriction as were the recipes allocated

to the remaining meals (in this case, it was due to the

recommendation of too many recipes with pork

sausage). When it comes to Figure 7, it can be seen

that the participants overwhelmingly felt that the

recipes allocated to the mid-morning meal were not

a good fit. Through the observation of the

participants’ comments on the suggested recipes for

the mid-morning meal, the cause for this misfit is

due to the fact that they either felt that the recipes

consisted of too much food or that this meal was not

necessary. It should also be noted that some subjects

felt that the recipes for dinner consisted of too much

food.

Figure 6: Scores for per meal regarding if the subject

could eat the suggested meal.

Figure 7: Scores regarding recipes adequacy for the meal.

Considering the results about the adequacy of the

daily plans, Figure 8 shows that the most disliked

daily plans where those for Sundays. Taking into

account the participants’ comments throughout the

tests, the cause of this was due to the fact that in this

day the participants usually eat recipes that are not

eaten in other days of the week or that they do not

have as many as meals in this day as they do in the

other days of the week. Regarding all evaluations of

the judged adequacy of each daily plan, 6.25% were

graded with 2 (disagree), 41.07% with 3 (agree) and

52.68% with 4 (strongly agree).

Concerning the weekly personalized plan, as

depicted in Figure 9, it is possible to conclude that

the majority of the participants would follow such a

plan (more than 70%) and also use an application

like this one, with 20% of the subjects disagreeing.

Furthermore, 50% of the participants felt that the

ICT4AWE 2017 - 3rd International Conference on Information and Communication Technologies for Ageing Well and e-Health

weekly plan was good for them. Figure 8 shows that,

in general, the participants had a positive experience

with the system.

Figure 8: Scores for daily plan adequacy.

Figure 9: Meal plan and SousChef acceptance scores.

5.2 User Interface Validation

In parallel to the validation of the outputs of the

recommender engine, we were also evaluating the

user interface. The evaluation of the SousChef

application was executed through the assistance of

usability tests with a low-fidelity prototype of the

user interface. The primary objectives of these tests

were to: 1) identify navigation and flow issues; 2)

evaluate the rating system used for the recipes; 3)

explore different data visualization methods with

elderly users.

Usability test sessions took place at a local day-

care centre, and five participants over 65 years were

recruited to participate. All of the participants had

taken part in previous usability tests with

smartphones or tablets. For the testing material, a

Motorola Moto G with a 5-inch screen and running

Android was used. An interactive prototype

developed with InVision was used in order to

simulate behaviours and flows.

Participants were required to complete a total of

seven small tasks. Task success, errors, deviations

and assistances were collected in order to evaluate

the performance of the users. Standard usability tests

were not used at this point because, rather than

general levels of usability and satisfaction, the

authors were seeking to identify specific problems

with the user interface, information architecture and

the mechanics of interaction, which could be

improved in future iterations. Task analysis was a

better fit for the purpose. On a user level, three

participants completed six out of the seven tasks

successfully; one performed five tasks successfully;

and the other participants only completed four tasks.

The main problems were identified in Task 3, which

required participants go back to the previous page by

using the back button. And on Task 6, which

required participants to view the meal plan for the

entire week. This was a critical navigation issue that

resulted from the lack of affordances of the button to

change the week.

In terms of data visualization, there were a few

issues with the charts used. First, the measurement

unit was not clear; the labels were also not readable

enough; and the daily objective plotted on the chart

was not self-explanatory.

All of these issues will be addressed in the next

iteration of the application.

6 CONCLUSIONS AND FUTURE

WORK

In this work a mobile meal recommender system,

named SousChef, was presented having as target

audience older adults. SousChef is intended to act as

a nutrition companion that guides older adult users

into making wise decisions regarding food

management and healthy eating.

Although in the literature, recommender systems

were presented and different ICT applications are

available, we have found no similar mobile-based

solution designed with a focus on older adults’

needs.

Tests to the recommender system and mobile

application were made with older adults, to infer, on

the one hand, the adequacy and quality of the meal

plans suggested and, on the other hand, to test the

usability of the mobile application itself for this

specific audience. The user testing results were very

satisfactory with more that 70% of the participants

considering following the meal plan suggestions and

use an application like the one presented. On the

usability side, results were also satisfactory for the

first prototype and we will keep iterating the current

solution and testing new features.

SousChef: Mobile Meal Recommender System for Older Adults

As future work, new features are to be

implemented, such as choosing the number of meals

to be planned per day and setting recurrent meals.

Another interesting feature to implement in the

system would be to take into consideration the

profile of more than one person. Also, a study with

more participants will allow a more detailed analysis

of the quality of the recommendations in terms of

the nutritional requirements of the participants.

This paper reports the first version of the system

and respective user mobile application, which brings

some limitations. Due to its stage of development, it

has not yet been possible to conduct a full-fledged

evaluation of the system. Because people have many

deeply rooted habits and beliefs related to food, we

are aware that a thorough evaluation on the usability,

usefulness and acceptance of the system may only

be done in a longitudinal study in users’ real context.

The authors are aware of this current limitation.

However, in this regard, the strength of the work

relies on it having involved older adult users from

onset, which allows to identify and eliminate

potential barriers to acceptance and use of SousChef.

ACKNOWLEDGEMENTS

We would like to acknowledge the financial support

from North Portugal Regional Operational

Programme (NORTE 2020), Portugal 2020 and the

European Regional Development Fund (ERDF)

from European Union through the project ‘Deus ex

Machina: Symbiotic Technology for Societal

Efficiency Gains’, NORTE-01-0145-FEDER-

000026.

REFERENCES

Aberg, J. (2006). Dealing with Malnutrition: A Meal

Planning System for Elderly. In AAAI Spring

Symposium: Argumentation for Consumers of

Healthcare, pages. 1-7.

Adomavicius, G., and Tuzhilin, A. (2005). Toward the

next generation of recommender systems: A survey of

the state-of-the-art and possible extensions. IEEE

transactions on knowledge and data engineering

17(6), 734-749.

Bunker, J. P. (2001). The role of medical care in

contributing to health improvements within societies.

International Journal of Epidemiology, 30(6), 1260-

1263.

Candeias, V., Nunes, E., Morais, C., Cabral, M. and Silva,

P. R. (2005). Princípios para uma Alimentação

Saudável. Lisbon: Direcção Geral de Saúde, page

255.

Espín, V., Hurtado, M. V., and Noguera, M. (2015).

Nutrition for Elder Care: a nutritional semantic

recommender system for the elderly. Expert Systems.

Hazman, M. and Idrees, A. M. (2015). A healthy nutrition

expert system for children. In E-Health and

Bioengineering Conference (EHB), 2015, pages 1-4.

IEEE.

Kuo, F. F., Li, C. T., Shan, M. K., and Lee, S. Y. (2012).

Intelligent menu planning: Recommending set of

recipes by ingredients. In Proceedings of the ACM

multimedia 2012 workshop on Multimedia for cooking

and eating activities, pages 1-6. ACM.

Le Couteur, D. G., Solon-Biet, S., Wahl, D., Cogger, V.

C., Willcox, B. J., Willcox, D. C., Raubenheimer, D.,

and Simpson, S. J. (2016). New Horizons: Dietary

protein, ageing and the Okinawan ratio. Age and

ageing, page afw069.

Long C. L., Schaffel N., Geiger J..W., Schiller W. R. and

Blakemore W. S. (1979). Metabolic response to injury

and illness: estimation of energy and protein needs

from indirect calorimetry and nitrogen balance.

Journal of Parenteral and Enteral Nutrition, 3(6):452-

456.

Mika, S. (2011). Challenges for Nutrition Recommender

Systems. In Proceedings of the 2nd Workshop on

Context Aware Intel. Assistance, Berlin, Germany,

pages 25-33. Citeseer.

Ministério da Saúde (2006). Tabela da composição de

alimentos. Instituto Nacional de Saúde Dr Ricardo

Jorge. Centro de Segurança Alimentar e Nutrição. 1a

Edição. Compil. Ilda Martins; concep., elab. e coord.

da edição Andreia Porto e Luísa Oliveira. INSA:

Lisboa, p 355. ISBN: 978-972-8643-19-5.

Ministry of Health. (2013). Food and Nutrition Guidelines

for Healthy Older People: A background paper.

Wellington: Ministry of Health. Report, ISBN: 978-0-

478-39399-6.

Murray, C. J., Abraham, J., Ali, M. K., Alvarado, M.,

Atkinson, C., Baddour, L. M., Bartels, D. H.,

Benjamin, E. J., Bhalla, K., Birbeck, G., et al. (2013).

The state of us health, 1990-2010: burden of diseases,

injuries, and risk factors. Jama, 310(6):591–606.

National Research Council (1989). Recommended Dietary

Allowances. 10 ed. Washington, D.C., National

Academy Press.

Ramic, E., Pranjic, N., Batic-Mujanovic, O., Karic, E.,

Alibasic, E., and Alic, A. (2011). The effect of

loneliness on malnutrition in elderly population.

Medical Archives, 65(2), 92.

Saka, B., Kaya, O., Ozturk, G. B., Erten, N., and Karan,

M. A. (2010). Malnutrition in the elderly and its

relationship with other geriatric syndromes. Clinical

Nutrition, 29, 745-748.

Sezgin, E., and Özkan, S. (2013). A systematic literature

review on Health Recommender Systems. In E-Health

and Bioengineering Conference (EHB), 2013, pages 1-

4. IEEE.

ICT4AWE 2017 - 3rd International Conference on Information and Communication Technologies for Ageing Well and e-Health

Tee, K., Moffatt, K., Findlater, L., MacGregor, E.,

McGrenere, J., Purves, B., and Fels, S. S. (2005). A

visual recipe book for persons with language

impairments. In Proceedings of the SIGCHI

conference on Human factors in computing systems,

pages 501-510. ACM.

Terrenghi, L., Hilliges, O., and Butz, A. (2007). Kitchen

stories: sharing recipes with the Living Cookbook.

Personal and Ubiquitous Computing, 11(5), 409-414.

Tran, Q. T., Calcaterra, G., and Mynatt, E. D. (2005).

Cook’s collage. In Home-Oriented Informatics and

Telematics, pages 15-32. Springer.

Wahlqvist, M. L., Kouris-Blazos, A. and Savige, S.

(2002). Food-based dietary guidelines for older adults:

healthy ageing and prevention of chronic

noncommunicable diseases. In Keep fit for life:

Meeting the nutritional needs of older persons, pages

81-111. Geneva Switzerland: World Health

Organisation 2002. ISBN 92 4 156210 2.

World Health Organization (2002). Keep fit for life:

meeting the nutritional needs of older persons. In Keep

fit for life: Meeting the nutritional needs of older

persons. World Health Organisation 2002. ISBN 92 4

156210 2.

World Health Organization (2010). Definition of an older

or elderly person. Geneva, Switzerland: WHO; 2010.

http://www.who.int/healthinfo/survey/ageingdefnolder

/en/index.html [acessed 20/03/2017].

SousChef: Mobile Meal Recommender System for Older Adults