Data Mining Techniques Applied to Recommender Systems for

Outdoor Activities: A Systematic Literature Review

Pablo Arévalo

, John Calle

, Marcos Orellana

and Priscila Cedillo

Laboratorio de Investigación y Desarrollo en Informática - LIDI,

Universidad del Azuay, 24 de mayo y Hernán Malo, Cuenca, Ecuador

Keywords: Recommending Systems, Health, Air Quality, Data Mining, Air Pollutants.

Abstract: Currently, many pollutants are released into the air, representing a risk to the environment and human health.

There are significant volumes of data generated by the devices that monitor these pollutants. This information

can represent a relevant input that allows the construction of applications, techniques, and methodologies to

reach a prediction of the state of the air. On the other hand, recommender systems are present in numerous

data processing methods, supporting the decision-making and promoting the improvement of the quality of

service of solutions. Although several studies have been presented, no secondary studies have been proposed.

Therefore, this paper presents a systematic review of the literature, which aims to identify the knowledge

areas, tools, methods, and data mining approaches used in recommender systems for outdoor activities related

to atmospheric pollutants. The results obtained contribute to creating new ways of recommendation systems

based on the previous topics.

1 INTRODUCTION

Currently, everybody is exposed to high levels of

environmental pollution. This situation causes a

significant impact on people's health and increases

the risk of suffering from different types of diseases

directly related to pollution of the environment.

Moreover, poor air quality affects a more significant

proportion of the most vulnerable population, such as

children and the elderly (Singla, 2018) (An et al.,

2018). In addition, with the rise of technology use in

recent decades, solutions generate large volumes of

data; this situation allows the construction of models,

techniques, or methodologies (Singla, 2018). For this

reason, being aware of the current air quality when

carrying out any outdoor activity is of vital

importance to prevent possible damage to health.

Recommender systems today have become

essential when choosing a product or service.

Currently, there are recommender systems in various

fields and industries to be found in any system; these

solutions are of substantial help in decision-making

https://orcid.org/0000-0002-6085-541X

https://orcid.org/0000-0003-0299-9279

https://orcid.org/0000-0002-3671-9362

https://orcid.org/0000-0002-6787-0655

and a boost to the quality of service (Ricci et al.,

2011). However, lately, studies related to secondary

studies have not been reported that support

researchers in discovering findings related to these

topics. Therefore, the evidence is scattered and

difficult to find, being necessary to join the most

effective approaches summarized in a unique study.

Then, this paper presents a Systematic Literature

Review (SLR) delving into the issues related to air

quality and its impact on health and the

implementation of recommender systems. This SLR

follows the methodology presented by Kitchenham &

Charters (2007) and answers the following research

questions: RQ1. What kind of information is required

to develop an outdoor recommender? RQ2. What

methodologies are used to address the development

of an outdoor recommender? RQ3. How are outdoor

recommenders addressed in data science? And RQ4.

How has the research on recommender systems for

outdoor activities been carried out?

Finally, this paper is structured as follows:

Section 2 presents related work, Section 3 develops

228

Arévalo, P., Calle, J., Orellana, M. and Cedillo, P.

Data Mining Techniques Applied to Recommender Systems for Outdoor Activities: A Systematic Literature Review.

DOI: 10.5220/0011045400003188

In Proceedings of the 8th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2022), pages 228-235

ISBN: 978-989-758-566-1; ISSN: 2184-4984

the SLR methodology, Section 4 analyzes the results

of the SLR execution, Section 5 presents a discussion,

Section 6 analyzes the threads of validity, and Section

7 presents the conclusions and the next steps of the

research.

2 RELATED WORK

Air pollution places a heavy burden on human health,

and understanding the effects of pollutants is a

constant challenge for our society. People die due to

diseases induced by air pollution, such as ischemic

heart disease, lung cancer, among others (Singla,

2018) and (An et al., 2018). There are various sources

of environmental pollution. In urban areas, vehicular

traffic is the most predominant source of emissions,

mainly composed of exhaust emissions of carbon

monoxide, nitrogen oxides, and suspended particles

from vehicles in megacities (Suresh et al., 2015).

With this problem, systematic reviews have been

carried out that have been interested in air quality in

different areas. For example, seeking to determine the

best statistical model based on machine learning

techniques to capture the non-linear relationship

between the concentration of air pollutants and their

emission and dispersion sources (Rybarczyk &

Zalakeviciute, 2018). In another case, the scientific

evidence linking air pollution to physical activity in

China was systematically reviewed, showing that it

affects behaviours related to daily physical activity in

residents (An et al., 2019). In another example, the

objective is to investigate the applications of deep

learning in the forecast of air quality in time series,

demonstrating that the accuracy of the results is

greater than that of individual models (Zaini et al.,

2021). In the same way, another study provides a

valuable synthesis of the relevant literature on smart

cities by analyzing and discussing the key findings in

creating sustainable cities and communities

considering environmental control and air quality

(Ismagiloiva et al., 2019).

Additionally, considering the particulate matter,

related studies have been evaluated in order to assess

the impact on health in terms of possible reductions

in premature deaths due to the reduction of this

pollutant (Jahn et al., 2011).

On the other hand, recommender systems are

techniques or software tools that provide suggestions

for elements, which can be useful to a user. These

suggestions or recommendations are related to

decision-making processes normally focused on a

specific topic (Ricci et al., 2011). Therefore, these

recommenders represent a substantial aid in decision-

making and an impetus to improve the quality of

services. For example, recommender systems have

been developed focused on educational data mining

in order to predict student performance (Thai-Nghe et

al., 2010) or focused on text mining (Betancourt &

Ilarri, 2020). Therefore, having adequate information

at the right time, especially when the individual is

exposed to a lesser amount of pollution, is a

substantial issue to protect their health and achieve a

better quality of life.

3 RESEARCH METHOD

A systematic review is a research method for

obtaining, evaluating, and interpreting information

related to a specific research question or area of

interest. Its objective is to facilitate an objective

evaluation of a research topic in a reliable, rigorous,

and methodological way. For the process, the study

carried out by Kitchenham, who provides a

methodology to carry out systematic reviews, it was

considered as reference (Kitchenham & Charters,

2007) due to it is mainly focused on three stages:

1) Planning the review: The review needs to be

identified, the research questions are specified, and

the review protocol is defined.

2) Conducting the review: the primary studies are

selected, the quality assessment used to include the

studies is defined, the data is extracted and monitored,

and the data are synthesized.

3) Reporting the review: the dissemination

mechanisms are specified, and the review report is

presented.

According to the research carried out on

recommender systems focused on the domain of the

human health, in physical activity specifically, there

is evidence of intervention of these systems in 2012

(Knoch et al., 2012). Therefore, this study considers

articles retrieved from digital databases between 2012

and 2021.

3.1 Research Questions

Research questions are part of the support that

manage the entire research process as it allows the

relevant data to be determined and transformed into a

research contribution. They should be formulated in

four sections: population, intervention, comparison,

and result (Kitchenham & Charters, 2007); this to

carry out a complete examination of the variation in

the study factors and between populations, for which

it was necessary to relate three variables, the use of

recommender systems, development of the physical

Data Mining Techniques Applied to Recommender Systems for Outdoor Activities: A Systematic Literature Review

229

activity, and the effect that it causes of the quality of

air when carrying out any type of activity. The

research questions that were asked are:

 RQ1. What kind of information is required to

develop an outdoor recommender?

 RQ2. What methodologies are used to address

the development of an outdoor recommender?

 RQ3. How are outdoor recommenders

addressed in data science?

 RQ4. How has the investigation of

recommender systems for outdoor activities

developed?

3.2 Search Process

The digital libraries and indexers selected for this

study were chosen because they cover a large number

of articles related to recommender systems, which are

detailed below: Digital library ACM, ScienceDirect,

SpringerLink, Scopus, IEEEXplore, Taylor and

Francis, EBSCO, Web of Science and Hinari (OARE)

The search string was then developed using

concise words, relevant terms, and alternative terms

that emerged from the research questions. Moreover,

the relevant terms defined for this study were:

“Recommender”, “Data mining”, “Activities”,

“Weather”, “Pollution”, and “Air”. Also, a set of

alternative terms was developed consisting of: “Big

data”, “Data Science”, “Physical activities”, “Sport”,

“Fitness”, “healthy”, “Air Pollution” and “Air

quality” to complement the relevant terms.

By using connectors “AND” and “OR” various

attempts were made with combinations of these

terms, adapting the search string to each library or

index, the better results were obtained with the

following combination:

Abstract: recommend* AND (data mining OR big

data OR data science) AND All Metadata: (activities

OR physical activi* OR sport OR fitness OR health*)

OR (weather OR pollu* OR air*).

3.3 Exclusion and Inclusion Criteria

The protocol, inclusion, and exclusion criteria must

be developed. This strategy reduces the number of

selected primary studies. In this study, the exclusion

criteria help eliminate studies that matched at least

one of the following:

 Duplicate publications that have reference to

the same study in several digital libraries.

 Short publications of less than five pages.

 Publications that are not in English.

 Gray literature (They do not have a digital

object identifier, also called DOI for its

acronym in English).

 Publications made before 2012.

The inclusion criteria to select works to meet the

following items:

 Publications that address issues related to air

quality.

 Publications that address issues related to

recommender systems.

 Publications that relate health, air quality, and

outdoor activities.

 Publications that implement methodologies to

develop recommender systems.

 Scientific articles, conferences, books.

 Recommenders that integrate data mining

techniques.

3.4 Quality Assessment

It is necessary to provide individual quality control to

each study. This information should be included

when answering the research questions posed. Table

1 shows a list of quality criteria. Additionally, each

question was answered, dividing them into sub-

questions identified by the prefix EC ##. The studies

that passed the inclusion and exclusion criteria were

identified by a code that has the following format: [A

+ sequential number Author Title].

The information collected per study had the

following characteristics: name of the library, the title

of the article, author(s), DOI, year of publication,

number of pages, and the number of times it was

cited, the latter using the search tool of articles

provided by Google Scholar

The most important characteristics were obtained

through the support of data mining experts and an

analysis of the literature, which will allow to answer

the research sub-questions. In the full reading stage of

the articles, each of them will be scored with a zero

or one, based on the presence or absence of that

characteristic within the article. This strategy ensures

that the same data extraction criteria are applied to

each article.

To carry out this process, a matrix was designed

in which both the score of each article – based on the

sub-questions – and the evaluation of its quality were

recorded. Subsequently, the bubble diagrams used in

the reporting stage were obtained based on this same

matrix. Each research question with its respective

extraction criteria, the options available to each, and

a summary of the data collection process are in the

following url: https://bit.ly/3HRujm5.

ICT4AWE 2022 - 8th International Conference on Information and Communication Technologies for Ageing Well and e-Health

230

4 RESULTS REPORT

It is necessary to externalize the results obtained

when conducting this study. At this stage, primary

studies were identified, selected, and evaluated using

the exclusion, inclusion, and quality criteria defined

above.

The list of selected articles is in the following url:

https://bit.ly/3HRujm5. The procedure was divided

into the following stages:

 Systematic search: The search string was

adapted according to each library or index in

this activity. Subsequently, the searches and

downloads of the resulting articles were carried

out, obtaining 3,417 studies.

 First selection: Since certain libraries and

indexers have a high number of resulting

works, a reading of the title of each of the

articles was previously carried out to discard

works that do not have a relationship to the

topic and scope of this research. In this step,

2,337 articles were discarded.

 Second selection: At this stage, of each work

obtained, the title, abstract, and keywords were

read, and later it was graded with "Accepted",

"Undecided", and “Rejected” based on the

inclusion and exclusion criteria. Only articles

that in their entirety had a rating of “Accepted”

by all investigators were considered for the

next stage, resulting in 108 studies.

 Third selection: In this step, the disagreements

and doubts in the selection of certain articles

collected in the previous step were resolved, by

means of a consensus among all the researchers

after the total reading of each article; the same

ones that were identified using coding,

mentioned before. At this stage, the primary

studies were reduced to 56.

 Quality evaluation: As a final step, from the

resulting articles, those that met the quality

criteria were selected, resulting in 44 primary

studies.

The quality of the studies was evaluated by

applying the criteria presented above. The results are

in the following url: https://bit.ly/3HRujm5, where

the percentage of studies that answered each research

question is shown. Articles EC03, EC08, EC10, and

EC15 have the highest scores.

5 DISCUSSION

The intention of this study is to show the areas of

interest, tools, and data mining methods that have

been used in recommender systems for outdoor

activities based on atmospheric pollutants. After the

systematic review, and based on the articles analyzed,

it can be observed that the main area of study of the

recommender systems lies in health, with the main

goal of improving the quality of life of people

regardless of the input data. These include, for

example, IoT events [A02], patient data [A04] or

various groups of contaminants [A33].

For this, they have different methods in data

mining to issue recommendations, focusing mainly

on classification techniques [A08] and clustering

[A16]. Also, they consider the users’ knowledge,

whether this is the content that the user develops

when using the system (i.e., source of content

knowledge) or generalized information from all users

(i.e., source of social knowledge).

5.1 Information Required to Develop

an Outdoor Activities

Recommender

The results obtained after the review indicate that the

required information is based on physical activities

and human health [A04, A05, A07, A08, A14, A21,

A28, A29, A30, A31, A32, A35, A39] [ A01, A02,

A04, A05, A06, A07, A08, A09, A10, A11, A12,

A13, A14, A15, A16, A18, A19, A23, A24, A25,

A26, A28, A29, A30, A31, A32, A33, A35, A36,

A37, A38, A39, A41].

Here, the final objective of the recommendations

seeks to reduce the possible damage to health and an

improvement in the quality of life. The data sources

of those solutions are databases on diabetes to predict

this disease [A10]. Information collected on diseases

and their respective symptoms to implement a

personalized system of recommendation of

prevention techniques [A11]; data regarding lifestyle

habits and patterns [A41] or sensory data from IoT

devices to recommend interventions to promote an

active lifestyle [A29].

Similarly, the articles related to health are directly

related to environmental pollution [A03, A05, A16,

A17, A28, A33, A37, A40]. Although the input data

comprises various pollution particles (e.g., pollutants

per particle, ozone, carbon monoxide, sulphur oxides,

nitrogen oxides), many studies share a central goal: to

prevent complications to human well-being. For

example, they are the vehicular mitigation pollution

Data Mining Techniques Applied to Recommender Systems for Outdoor Activities: A Systematic Literature Review

231

[A16] or recommended pedestrian routes that

minimize the time of exposure to allergens [A28].

Consequently, before developing a

recommendation system based on pollutants, it is

necessary to consider the benefit or positive impact

that can be provided to human well-being. A

recommender to determine the best times for outdoor

activities fully covers this objective to improve the

quality of life.

Figure 1: Study areas with the various approaches

It should be emphasized that it is necessary to

know the main approaches (algorithms) involved in

health issues when trying building a recommender

system. In Figure. 1, the influence of the health area

on collaborative and content-based filtering

approaches can be seen as they are the most required.

[A01, A06, A08, A14, A15, A18, A29, A35, A36,

A38] [A01, A15, A18, A24, A35, A38].

A recommender with collaborative filtering looks

for the system to issue recommendations taking into

account the users’ interactions in the past. For

example, consider that if users shared similar

preferences long ago, they might have equivalent

tastes (Yu, 2018). Then, through various methods,

citing some: frequency, weighting, or similarity of the

cosine, it is possible to obtain a measure that indicates

this similarity among users, generally based on the

available ratings (Yu, 2018).

On the other hand, in a content-based approach,

the system learns to recommend elements similar to

those that interested the user long ago. The similarity

of these elements is calculated based on the

characteristics associated when comparing them. In

other words, each element has a table with its main

attributes. When issuing a recommendation, a search

is done for similar attributes between the profile of

the user and the items they liked in the past (Ricci et

al., 2011) (Yu, 2018).

In collaborative filtering, other additional factors

such as demographics can also be considered by

taking additional attributes (e.g., gender, age,

location) to issue recommendations demonstrating its

usefulness in health by extracting more information

from patients.

5.2 Methodologies Used in Addressing

the Development of an Outdoor

Activities’ Recommender

The extraction criteria ranging from EC05 to EC08

are taken into account to answer the RQ2 research

question. As mentioned above, collaborative filtering

and content filtering approaches are the most

frequently considered algorithms that are most

frequently considered in the development of

recommenders. However, it is also necessary to

consider the source and type of knowledge used in the

system.

The results indicate that a large part of the primary

studies uses different sources of knowledge when

addressing the construction of recommender systems.

[A01, A02, A03, A04, A05, A06, A07, A08, A09,

A10, A11, A12, A13, A15, A16, A17, A18, A19,

A20, A21, A22, A23, A24, A25, A26, A28, A29,

A30, A31, A33, A34, A35, A36, A37, A39, A40,

A41, A42, A43, A44].

The sources of knowledge include the

understanding that one has about the target user to

issue personalized recommendations. The knowledge

can be divided into three sections: the knowledge that

is available from the target user (Individual), the

knowledge about the characteristics of the article to

recommend (Content), and knowledge about the

broader community, including the target user (Social)

(Ricci et al., 2011).

Figure 2: Sources of knowledge used in the various

approaches.

As shown in Figure 2, a collaborative filtering

approach uses social knowledge represented by the

opinions or behaviour of a user community. In other

ICT4AWE 2022 - 8th International Conference on Information and Communication Technologies for Ageing Well and e-Health

232

cases, it is possible that there is only knowledge about

the characteristics, uses, or domains of the elements

that are recommended, which allows us to infer what

particularities attract users. In this case, a focus on

content is a better option.

From a perspective focused on knowledge

sources, providing a hybrid recommendation is

mixing a social source with a content source by

adapting the necessary algorithms to accept a

knowledge source more typically associated with

another type (Ricci et al., 2011). Being clear about the

base content of the system helps to determine the

algorithms that best adapt to the sources, considering

the domain in which the recommendations will be

issued.

5.3 Data Mining Techniques Used in

Developing an Outdoor

Recommender

The use of data mining techniques, to aid

recommendation systems in their goal to learn the

correct user profiles, is an active area of research

(Alabdulrahman et al., 2018). The primary studies

analyzed focus on using neural networks as described

in Figure 3, where particulate pollutants such as

carbon monoxide are targeted. These pollutants are

the most common in most studies as the main

component for generating predictions or

recommendations on air quality.

The data found in the literature analysis are

primarily concentrated on neural network techniques

since it is identified as one of the most efficient

processing of the collected data [A09]. Thus, many

primary studies address particular pollutants, like

carbon monoxide, due to their more significant

impact on people's health and greater environmental

concentration.

Figure 3: Classification variables in air quality.

The studies dealing with Artificial Neural

Network (ANN) and the criteria regarding air quality

as pollutants per particle and carbon monoxide are:

[A03, A05, A06, A08, A09, A11, A12, A16, A17,

A21, A27, A28, A33, A34, A37, A38, A40, A43].

Next, Figure 4 shows a high concentration of

primary studies that focus on Bayesian networks as

the main method to use in collaborative and content

filtering approaches. Bayesian networks are helpful in

domains with high stability in user preferences,

meaning user preference changes slowly concerning

the time required to build the model. The articles that

refer to the Bayesian network classification method

are: [A01, A08, A10, A13, A23, A24, A27, A28,

A29, A34, A42, A43].

The trend of using data mining techniques occurs

because it allows exploring the relationship between

elements based on how users have rated them (Ricci

et al., 2011). However, these techniques need to

compare each user with others, which is not practical

with huge data sets as well as requiring a class tag

and, in many applications, such tags are not available,

leading to domain wide expert tagging. They also

require a class label, and in many applications such

labels are not available, leading to domain-wide

expert labelling (Alabdulrahman et al., 2018). So, it

can be beneficial to perform a dimensionality

reduction that, although they require an extensive

offline calculation, the results scale much better

(Ricci et al., 2011).

It is essential to consider that clustering

techniques can identify groups with similar

characteristics. Once these groups have been

discovered, it is possible to make predictions taking

these characteristics into account. Although these

grouping methods have less precision than the

ranking methods, they can be applied as a preliminary

step to reduce the number of candidates or distribute

them to different recommendation engines. A pre-

grouping can be a valuable trade-off between

accuracy and performance (Ricci et al., 2011).

Figure 4: Most used classification methods in the various

approaches.

Data Mining Techniques Applied to Recommender Systems for Outdoor Activities: A Systematic Literature Review

233

5.4 Research Developed in

Recommender Systems for

Outdoor Activities

The criteria EC13, EC14, and EC15 are considered to

answer this question. A large majority of studies have

been developed in the field of academic research

[A01, A02, A03, A04, A05, A06, A07, A08, A09,

A10, A11, A12, A13, A14, A16, A18, A19, A20,

A21, A22, A23, A24, A25, A26, A27, A28, A29,

A30, A31, A33, A34, A36, A37, A38, A39, A40, A41

A43, A44]. The results also indicate that evaluations

are developed through experiments and case studies

[A01, A02, A04, A06, A07, A08, A09, A11, A12,

A13, A15, A16, A17, A18, A19, A21, A24, A26,

A28, A32, A34, A35, A41, A42, A44] [A03, A05,

A29, A31, A33, A37, A40].

In [A44], a group recommendation system has

been made to explore network document resources

using the knowledge graph and Long short-term

memory (LSTM), carrying out experimentation of the

system in the field of Big Data applications in the

packaging industry.

A case study based on a national level project

focused on the pre-processing and analysis of the data

collection of a city was developed in order to provide

the public with a guide by establishing a Big Data

platform for fine dust and provide administrative

guidance to public institutions and local governments

and inform the integrated indoor and outdoor air

management service [A37].

6 THREATS TO VALIDITY

6.1 Lack of Important Primary Studies

The primary studies were obtained from 9 sources

between libraries and indexes. Although the selected

databases are compliant for this type of study, the

possibility of a small group of primary studies being

eliminated is not ruled out. In the same way, when

building the search chain, certain words can be

chosen incorrectly, obtaining wrong results.

A pre-evaluation was considered when

determining the search chain to observe the results

obtained to mitigate this threat. In the beginning, the

chain lacked words like "Sport" or "Health”, omitting

a considerable number of results. Later, these words

were included for a second search. Then all the

connectors used in the chain were of type "AND".

These actions caused a shortage of results, obtaining

one or in specific libraries no results.

Finally, changing the AND connectors that joined

the secondary terms for connectors of the OR type

was considered, solving these problems. Although

many studies were obtained, one more stage was

included when choosing the primary articles.

6.2 Reliability of the Selection

In each of the previously defined selection stages,

primary articles related to the proposed topic may

also be excluded. Each researcher scored Yes, No,

and Review each article to mitigate this threat. The

objective is to reach a consensus among all

researchers on specific issues that are not clear. Then,

there is no doubt about the selected options. This

procedure was carried out at each selection stage to

prevent the elimination of articles relevant to the

proposed investigation.

6.3 Data Extraction

A threat during data extraction can arise due to

misunderstanding or disagreement between the

reviewers. So then, five papers were randomly

selected, and each of the researchers involved issued

their respective interpretations of each one of them.

Subsequently, the Fleiss ’kappa calculation was

performed with a resulting value of 0.65.

According to Landis and Koch (Landis & Koch,

1977), values between 0.61 and 0.80 are interpreted

as substantial agreement. Many studies did not

provide clear answers to the extraction as mentioned

above criteria, so this obtained result provides a good

coefficient of agreement.

7 CONCLUSIONS

The systematic review process has satisfactorily

complied with the proposed validations and

evaluations. The agreement for the selection of

studies and the understanding of the extraction

criteria applied later in the reading of said studies was

verified and improved. In addition, a reasonable

assessment of the quality of the articles was obtained

using a quantified score. Likewise, there is an

acceptable validity in all stages of the systematic

review, so it can be deduced that the planning of said

review was adequate

As the main conclusion, it has been determined

that currently, the recommender systems use

collaborative filtering to make recommendations,

whatever the case. However, there is very little

ICT4AWE 2022 - 8th International Conference on Information and Communication Technologies for Ageing Well and e-Health

234

research and applications that use a knowledge-based

approach. Therefore, it would be optimal for this type

of application; that is, it does not need or depend so

much on the information provided by the system's

users, but rather that the system can provide

recommendations based on the history that has been

saved in the system.

ACKNOWLEDGEMENTS

The authors wish to thank the Vice-Rector for

Research of the University of Azuay for the financial

and academic support and all the staff of the School

of Computer Science Engineering and the Laboratory

for Research and Development in Informatics (LIDI).

REFERENCES

Alabdulrahman, R., Viktor, H., & Paquet, E. (2018).

Beyond k-NN: Combining cluster analysis and

classification for recommender systems. IC3K 2018 -

Proceedings of the 10th International Joint Conference

on Knowledge Discovery, Knowledge Engineering and

Knowledge Management, 1(Ic3k), 82–91.

https://doi.org/10.5220/0006931200820091

An, R., Shen, J., Ying, B., Tainio, M., Andersen, Z. J., & de

Nazelle, A. (2019). Impact of ambient air pollution on

physical activity and sedentary behavior in China: A

systematic review. Environmental Research, 176,

108545. https://doi.org/10.1016/j.envres.2019.108545

An, R., Zhang, S., Ji, M., & Guan, C. (2018). Impact

of ambient air pollution on physical activity among

adults: a systematic review and meta-analysis.

Perspectives in Public Health, 138(2), 111–121.

https://doi.org/10.1177/1757913917726567

Betancourt, Y., & Ilarri, S. (2020). Use of text mining

techniques for recommender systems. ICEIS 2020 -

Proceedings of the 22nd International Conference on

Enterprise Information Systems, 1(Iceis), 780–787.

https://doi.org/10.5220/0009576507800787

Ismagiloiva, E., Hughes, L., Rana, N., & Dwivedi, Y.

(2019). Role of Smart Cities in Creating Sustainable

Cities and Communities: A Systematic Literature

Review. In Y. Dwivedi, E. Ayaburi, R. Boateng, & J.

Effah (Eds.), ICT Unbounded, Social Impact of Bright

ICT Adoption (pp. 311–324). Springer International

Publishing.

Jahn, H. J., Schneider, A., Breitner, S., Eißner, R.,

Wendisch, M., & Krämer, A. (2011). Particulate matter

pollution in the megacities of the Pearl River Delta,

China – A systematic literature review and health risk

assessment. International Journal of Hygiene and

Environmental Health, 214(4), 281–295.

https://doi.org/10.1016/j.ijheh.2011.05.008

Kitchenham, B. A., & Charters, S. (2007). Guidelines for

performing Systematic Literature Reviews in Software

Engineering. EBSE Technical Report EBSE-2007-01.

School of Computer Science and Mathematics, Keele

University. January, 2007.

Knoch, S., Chapko, A., Emrich, A., Werth, D., & Loos, P.

(2012). A context-aware running route recommender

learning from user histories using artificial neural

networks. Proceedings - International Workshop on

Database and Expert Systems Applications, DEXA,

106–110. https://doi.org/10.1109/DEXA.2012.49

Landis, J. R., & Koch, G. G. (1977). Landis amd

Koch1977_agreement of categorical data. Biometrics,

33(1), 159–174.

Ricci, F., Rokach, L., Shapira, B., & Kantor, P. B. (2011).

Recommender Systems Handbook. In Journal of

Physics A: Mathematical and Theoretical (Vol. 44,

Issue 8). https://doi.org/10.1088/1751-8113/44/8/

085201

Rybarczyk, Y., & Zalakeviciute, R. (2018). Machine

Learning Approaches for Outdoor Air Quality

Modelling: A Systematic Review. Applied Sciences,

8(12). https://doi.org/10.3390/app8122570

Singla, S. (2018). Air ality friendly route recommendation

system.

PhD Forum 2018 - Proceedings of the 2018

Workshop on MobiSys 2018 Ph.D. Forum, Part of

MobiSys 2018, 9–10. https://doi.org/10.1145/

3212711.3212717

Suresh, L. P., Dash, S. S., & Panigrahi, B. K. (2015). A16 -

A Bi-level clustering analysis for studying about the

sources of vehicular pollution in Chennai. In Advances

in Intelligent Systems and Computing (Vol. 325).

https://doi.org/10.1007/978-81-322-2135-7

Thai-Nghe, N., Drumond, L., Krohn-Grimberghe, A., &

Schmidt-Thieme, L. (2010). Recommender system for

predicting student performance. Procedia Computer

Science, 1(2), 2811–2819. https://doi.org/10.1016/

j.procs.2010.08.006

Yu, L. (2018). A35 - Cloud storage–based personalized

sports activity management in Internet plus O2O sports

community. Concurrency Computation, 30(24), 1–10.

https://doi.org/10.1002/cpe.4932

Zaini, N., Ean, L. W., Ahmed, A. N., & Malek, M. A.

(2021). A systematic literature review of deep learning

neural network for time series air quality forecasting.

Environmental Science and Pollution Research, 29(4),

4958–4990. https://doi.org/10.1007/s11356-021-

17442-1

Data Mining Techniques Applied to Recommender Systems for Outdoor Activities: A Systematic Literature Review

235