Deep Learning based Tourism recommendation system

Ismat Fathima1a*
Bonthu Kotaiah2
1a*
Research Scholar,
Department of Computer Science & Information Technology,
Maulana Azad National Urdu University, Hyderabad. India.
2a
Assistant Professor,
Department of Computer Science & Information Technology,
Maulana Azad National Urdu University, Hyderabad. India

1 a)*
ismat@manuu.edu.in
2a)
kotaiah.bonthu@manuu.edu.in

Abstract - Many different fields make use organizing a tour to make the most of their
of intelligent computer-based travel outings. Modern travelers want to
recommendation systems (RS). Social make their judgments about their trip,
media sites produce a lot of posts, likes, select specific options, complete bookings,
and other interactions because of their and make direct payments to avoid using
massive user bases, and these interactions intermediaries. For this, they turn to the
operate as a database for various internet. However, there is an abundance
recommendation algorithms. This study's of information on the web about many
core area of interest is the tourism sector, tourism-related topics, making it time-
and RS is a crucial tool for trip planning consuming to find a suitable trip package,
for travelers. The research in this broad product, or service. Some tourists plan
area has been studied extensively, and this their trip, right from deciding which
paper provides a current overview of that country to visit and the activities/events
research, taking into account the various they will be participating in, while others
interface types, the range of recommender may be more spontaneous in first reaching
algorithms, the features that such models a destination and then planning their
provide, and their application of artificial itinerary as the trip progresses. In both
intelligence techniques. This survey also cases, the user is flooded with a plethora of
provides some insights on the development options and may not be sure what to pick,
of the industry's most promising research which is where an automated
areas for the following years. recommender system (RS) comes into the
picture.
Keywords: Social media, Points of Interest A recommender system can help
(POI), Recommender System (RS), Deep with a variety of tour-related questions,
Learning such as the best vacation spots in a
particular season, the best weather to visit
1. Introduction a place of interest (POI), the quickest
mode of transportation, or advice on where
Traveling is an experience in itself to find good spots for photography
a combination of the journey, the enthusiasts. For example, Umanets et al
destination, time spent traveling, the stay, developed a service that could be run on
the season, and the activities undertaken. mobile phones and the web, targeted at
Tourists typically seek assistance while finding previously unseen points of

Electronic copy available at: https://ssrn.com/abstract=4286575

interest [2]. This would help returning Korea, Jeong et al argue that the
tourists to visit places they might have personality type of the tourist plays a
missed before. To improve the major role in selecting a tourism
effectiveness of the RS, researchers have destination [3].
employed various techniques. The primary Wang processed users’ information
sources of user data are blogs, geo-tags, like demographic variables viz age,
check-in information, GPS records, and so gender, profession or occupation and city,
forth. Every method used to improve the and review data to provide a system to
effectiveness of the travel recommendation recommend personalized travel products.
systems has some benefits and drawbacks. The researcher used a total of 1,283,715
In the next section, related work is reviews for his study [13]. G and H.
discussed. The most popular Verma et al focussed on rural tourism in
implementation strategies in the field of India using opinion mining. They
travel recommendation to date are covered employed a supervised machine learning
in Section 3. The last section concludes method to categorize the sentiments from a
with future scope of travel RS. variety of travel-related companies and
websites, reviews of hotels, and tourism
2. Related Work agencies left by foreign travelers. “Term
Frequency - Inverse Document Frequency
Previous publications that are (TF-IDF)” metrics are calculated to find
briefly covered in this section illustrate the relevance of the words identified for
how recommendation systems are applied sentiment analysis, and a robust model is
in the tourism industry. Many systems proposed [5]. Offering a combined
employ machine learning, while some Lexicon-based and rule-based NB
apply Deep Neural Networks for their sentiment analysis approach to extract
working. tourist characteristics from mobile apps,
Lucas et al in their study, apply Muthukrishnan and others in their paper
classification based on association in a collect different reviews on Twitter
recommender system for tourism, connected to travel and processes them to
implementing a hybrid recommendation identify the polarity and categorize each
technique. It is named Personalized into different sentiments [6].
Sightseeing Planning System [8]. A. In their paper, Zelenka et al.
Umanets and others developed an present a conceptual framework for
application called GuideMe and its offering validated destination assessments
primary features include its ability to and verified ratings of tourism services.
integrate with social networks. This Tripadvisor and Booking.com are the two
prototype works on both Android and iOS case studies used in the main research. A
in helping suggest unexplored tourist trust model is created after conducting a
destinations based on user ratings [2]. SWOT analysis of the review and
Employing the Amazon Reviews dataset,. verification processes of the
Kulkarni et al rank tourist places based on aforementioned websites [7]. To define the
the positive and negative reviews about geographical, temporal, and demographic
them. They use a deep learning algorithm tourist flows throughout a tourist region,
in arranging the sequence of POIs [1]. the research by Paolanti and others
Proposing an RS using Social Network presents and evaluates a deep learning
analysis for the city of Seoul in South geodata framework. The dataset included

Electronic copy available at: https://ssrn.com/abstract=4286575

Tweets in both Italian and English, and takes into account a variety of tourism
four DNNs were used to determine the resources for a set amount of time [9].
sentiment of the Tweets. By manually Based on a deep learning analysis of social
estimating with the help of human media, the study by Mishra and others
annotators in addition to automatically offers a tactical strategy for surviving the
judging by the surrounding sentences or Covid-19 pandemic. It examines the
hashtags, the authors generated a hospitality and healthcare industries as two
multilingual dataset that was more tourism sub-domains [14].
accurate than the ones already available. A unique bagging-based
The paper employs data visualization multivariate ensemble deep learning
techniques such as statistical graphics, strategy that combines stacked auto
charts, and information graphics [12]. The encoders and kernel-based intense learning
study by Li and another analyzes various machines is the B-SAKE model, which
well-known tourist destinations in China Sun et al. propose. This model increases
using sensor detection technologies, taking data volume and improves feature
into account factors like tourist type, extraction efficiency to address the over-
gender, and location. It also combines the fitting issue and improve forecasting
survey findings of tourists' satisfaction accuracy. [17]. The work by Yuan creates
with comprehension with online comments a Hadoop Distributed File System (HDFS)
on Ctrip [20]. based on the Map-Reduce concept and
presents an enhanced support vector
machine (SVM) method based on travel
consumer sentiment analysis. [22]. Zhang
et al in their paper employ text mining,
sentiment analysis, and social media
analytics and explore original reviews
about 160 Chinese cultural theme parks.
Here, 9 cultural aspects are used and
sentiment classification is done on this
basis [18].
The paper as part of the CRUMBS
project, classifies their CRUMBS
Fig. 1 Word cloud of the titles of the selected papers
recommender system as “a Web-based,
pull-based, user constraint-based TRS”
Santamaria-Granados and others
and “Mobile Dynamic Social
suggest a recommender system based on
Recommender System”. The
physiological information gathered from
recommendation mechanism depends on
wearable devices worn by travelers. The
data about tourist destinations and the
engine also creates a list of the best POIs
categories to which they belong,
based on the user's profile, emotional state,
demographic data, explicit tourist
and portfolio of travel experiences [10].
preferences regarding POI categories, tags
The creation of an RS based on big data
about POIs created by visitors and POI
technologies, AI, and operational research
ratings. This is the information used as
to promote tourism in the Daraa-Tafilalet
input by the system's suggested algorithms
region of Morocco is covered in the work
for user modeling and recommendation
by Farani et al. The RS acts as a travel
[4]. Farokhi et al propose a recommender
planner, creating a thorough itinerary that

Electronic copy available at: https://ssrn.com/abstract=4286575

using multi-criteria using a collaborative depends on the optimisation techniques
filtering technique, which will output user- used by it to present the best results to the
based and item-based recommendations. users. This section will cover the popular
They used offline as well as online training strategies implemented and the algorithms
of data [11]. Khomash et al use big data used by researchers for implementing
analytics to extract the opinion of tourists tourism recommender systems.
by the reviews they leave on Google The recommendation framework
Maps. Focusing on 8 tourist destinations in by Lucas et al [8] consists of using “CBA-
Indonesia, they concluded that there is a Fuzzy”, an algorithm developed
high correlation between sentiment and specifically for this purpose. This
reality, and that user reviews help a great algorithm “mines the fuzzy association
deal for the tourist to decide about rules that constitute the associative
traveling to a point of interest [16]. classification model used for
recommendation”. Their method
Baharmian and others propose an constitutes two stages - The method's
ontology-based tourism recommendation initial phase entails creating user groups
system that employs the spreading with comparable interests and traits. The
activation model. Their technique uses active user is categorized into one or more
three steps to address the issue of of these groups in later phases so that
information overload - “ontology-based recommendations can be made to him
content analyzer, ontology-based profile based on his profile. A clustering
learner, and ontology-based filtering algorithm is used to create the user
component”. Considering 1573 POIs in groupings using attributes holding
Tehran, Iran, they limited the demographic information about users
recommendations to only those which are (such as age, postal code, and degree of
at a predefined distance from the user [21]. education), as well as attributes of goods
Bhaidkar et al created a prototype in which to be recommended, which users have
they used the triangulation method with rated or purchased. The Personalized
GPS to track the location of the user and Sightseeing Planning System (PSiS) is
recommend “nearest as well as categorized used which was designed to help tourists
tourist visiting places with rating”. This in Oporto, Portugal, obtain a customized
will give the tourist a list of tour itinerary. The approach starts by using
recommendations to choose from on a data from the 241 sites of interest that are
user-friendly mobile app [19]. already in the PSiS, the dataset of which
was provided by the user community to
assess the procedure.
3. Methods Used The main focus of the mobile
application developed by Umanets and
There are many factors which others [2] is “the users’ current location,
tourists take into consideration before which is represented by the geographic
planning a trip - the popularity of the coordinates obtained using the positioning
destination, user’s interest in the place, service available on the user’s device,
traveling time to reach the place, the mode provided by GPS or Wi-Fi connections”.
of transport and the traffic, and finally the The app provides local points of interest
cost involved in the entire trip. The and their respective distances based on the
popularity of a recommender system users' current location. It enables users to

Electronic copy available at: https://ssrn.com/abstract=4286575

examine points of interest without showing implemented. After the emotion is
the user's distance from the nearby tourist identified, the “Tourist Experience
attraction when positioning or location Recommendation System” is set to work.
services are unavailable. To obtain the contextual data, an interface
206,387 users, 7,300 scenic is employed using the “Tourist
locations, and 1,283,715 reviews were Traceability Ontology (OntoTouTra)”.
included in the dataset for the model This recommender system works with TE
created by Wang [13]. Data was acquired and MongoDB datasets, and following the
using the distributed web crawler system analysis, a list of travel suggestions is
made possible by presented [10]
Scrapy+redis+MongoDB, which collects
network data and converts it into JSON 4. Conclusion
format. A few of the user's details include
their gender, age, occupation, city, All the systems reviewed are based
previous review items, and review on enough data collected for the model
information. The user's behaviour features suggested or prototype created. However,
are retrieved in order to analyse user most systems are specific to a particular
interests and build the feature model. The place, generally to a city. The interfaces
information regarding the service includes used are either web-based or mobile
the name, address,, and review details of device based. Some are built both for
the tourism service products. Data Android and iOS. Most people looking for
processing was done using word a flawless itinerary generally use web-
embedding technology. The review text based recommender systems because of
data of users and objects were processed the ease of using them, without having to
using Chinese word segmentation download or install extra apps. The
technology. Abeysinghe et al present in common factor in all these systems is the
this paper that the social media analytics ‘intelligence’ used to return suggestions to
platform for the tourism industry can be the users. Research papers and
utilized to provide economic value and publications from the last 8 years were
support decision-making. Their suggested analyzed to conclude that recommender
system has the capacity to recognise systems in tourism have increased in leaps
conversation threads and topics covered in and bounds for the modern tourist. The
microblogs, as well as to examine paces at which these systems are
sentiments and emotions that go along enhancing with each passing season say
with them. Only text data is gathered and that many people are turning to AI for
used in the insight generating process for their trip planning. Though most authors
the system's current iteration [15]. focus on a particular tourist destination
Granados and others use Emotion with multiple points of interest for the
Recognition (ER) for their prototype. tourist, these studies can be generalized for
Tourists use a wearable smart phone/band wider audiences and other tourist places.
that gathers information about their Tourists are looking for more personalized
emotions and registers it on the band app. recommendations for their travel, and
It extracts the emotion from the dataset, would want a comprehensive service for
and a hybrid Deep Learning (DL) all their travel needs, which would include
approach from “CNN and Long Short- the best season to visit a place, recent
Term Memory (LSTM) networks” are change in visiting timings of any POI,

Electronic copy available at: https://ssrn.com/abstract=4286575

current traffic updates, and the local no. 2, p. 34, Mar. 2021, doi:
conditions 10.3390/admsci11020034.
8. J.P. Lucas., B.E.S. Coelho, M.N.M
References García, A.M.A Figueiredo, C.L
Martins, “Applying Recommender
Methodologies in Tourism Sector”,
In Highlights in Practical
1. A. Kulkarni, M.R., P. Barve, and
Applications of Agents and
A. Phade, “A Machine Learning
Multiagent Systems. Advances in
Approach to Building a Tourism
Intelligent and Soft Computing, vol
Recommendation System using
89. Springer, Berlin, Heidelberg,
Sentiment Analysis”. International
2011. https://doi.org/10.1007/978-
Journal of Computer Applications,
3-642-19917-2_13
vol 178. pp 48-51, 2019.
9. K.A. Fararni, F. Nafis, B.
10.5120/ijca2019919031.
Aghoutane, A. Yahyaouy, J. Riffi,
2. A. Umanets, A. Ferreira, and N.
and A. Sabri, “Hybrid
Leite, "GuideMe – A Tourist Guide
Recommender System for Tourism
with a Recommender System and
Based on Big Data and AI: A
Social Interaction", Procedia
Conceptual Framework”, IEEE,
Technology, vol. 17, pp 407-414,
Big Data Mining and Analytics, vol
2014.
4, no. 1, Mar 2021.
doi:10.1016/j.protcy.2014.10.248
10. L. Santamaria-Granados, J.F.
3. C.-S. Jeong, K.-H. Ryu, J.-Y. Lee,
Mendoza-Moreno, A.Chantre-
and K.-D. Jung, “Deep Learning-
Astaiza, M. Munoz-Organero and
based Tourism Recommendation
G. Ramirez-Gonzalez, “Tourist
System using Social Network
Experiences Recommender System
Analysis,” International Journal of
Based on Emotion Recognition
Internet, Broadcasting and
with Wearable Data”, Sensors, vol.
Communication , vol. 12, no. 2, pp.
2, Nov. 2021,
113–119, May 2020.
doi.org/10.3390/s21237854
4. CRUMBS Project, Ministry of
11. M. Farokhi, M. Vahid, M. Nilashi,
Industry, Energy and Tourism,
and O. b. Ibrahim, “A Multi-
Spain
Criteria Recommender System for
5. G. Verma, and H. Verma, “Opinion
Tourism Using Fuzzy Approach”,
Mining on Rural Tourism In India-
Journal of Soft Computing and
Qualitative Perspective,” Turkish
Decision Support Systems, May
Journal of Computer and
2016, pp 19-29.
Mathematics Education, vol. 12,
12. M. Paolanti, A. Mancini, E.
no. 7, Apr. 2021, pp. 717–722.
Frontoni, A. Felicetti, L. Marinelli,
6. H. Muthukrishnan, C. P. T. Selvi,
E. Marcheggiani, and R. Pierdicca,
M. Deivakani, V. Subashini, N. J.
“Tourism destination management
Savitha, and S. G. Kumar,
using sentiment analysis and geo-
“Aspect-Based Sentiment Analysis
location information: a deep
for Tourist Reviews,” Annals of
learning approach,” Information
R.S.C.B, vol. 25, no. 3, Mar. 2021,
Technology & Tourism, vol. 23,
pp. 5183–5194.
no. 2, pp. 241–264. Feb. 2021,
7. J. Zelenka, T. Azubuike, and M.
doi:10.1007/s40558-021-00196-
Pásková, “Trust Model for Online
4ed
Reviews of Tourism Services and
13. M. Wang, “Applying Internet
Evaluation of Destinations,”
information technology combined
Administrative Sciences, vol. 11,

Electronic copy available at: https://ssrn.com/abstract=4286575

 with deep learning to tourism Advance Research, Ideas and
collaborative recommendation Innovations in Technology, vol 1,
system”, PLoS ONE, vol 15, no. no. 5, 2015.
12:e0240656, Dec 2020. 20. Y. Li, and H. Gan, “Tourism
https://doi. Information Data Processing
org/10.1371/journal.pone.0240656 Method Based on Multi-Source
14. R.K. Mishra, S.Urolagin, J.A. Data Fusion,” Journal of Sensors,
Jothi, A.S. Neogi, and N. Nawaz, vol. 2021, Article ID 7047119 pp.
“Deep Learning-based Sentiment 1–12, Jul. 2021, doi:
Analysis andTopic Modeling on 10.1155/2021/7047119.
Tourism During Covid-19 21. Z. Bahramian, and R.A.
Pandemic”, Front. Comput. Sci. Abbaspour, “An Ontology-Based
3:775368. Tourism Recommender System
doi:10.3389/fcomp.2021.775368 Based On Spreading Activation
15. S. Abeysinghe, I. Manchanayake, Model”, in SMPR, Iran, 2015.
C. Samarajeewa, P. Rathnayaka, 22. Z. Yuan, “Big Data
M. Walpola, R. Nawaratne, T. Recommendation Research Based
Bandaragoda, D. Alahakoon, on Travel Consumer Sentiment
“Enhancing Decision Making Analysis”, Front Psychol, Feb
Capacity in Tourism Domain 2022. doi:
Using Social Media Analytics” In 10.3389/fpsyg.2022.857292.
Social Media Sentiment Analysis -
UOM CSE, 2018.
16. S. Khomsah, R. D. Ramadhani and
S. Wijayanto, “Big Data Analytics
to Analyze Sentiment, Emotions,
and Perceptions of Travelers (Case
Study: Tourism Destination in
Purwokerto Indonesia)”, Jurnal E-
KOMTEK (Elektro-Komputer-
Teknik), Vol. 5, No. 2 (2021) pp.
284-297 doi.org/10.37339/e-
komtek.v5i2.791
17. S. Sun, Yanzhao Li, J. Guo, and S.
Wang. “Tourism Demand
Forecasting: An Ensemble Deep
Learning Approach.” Tourism
Economics, Jul 2021.
https://doi.org/10.1177/135481662
11025160.
18. T. Zhang, B. Li, and N. Hua,
“Chinese cultural theme parks: text
mining and sentiment analysis”,
Journal of Tourism and Cultural
Change, pp. 37-57, Jan. 2021 doi:
10.1080/14766825.2021.1876077
19. Y. Bhaidkar, P. Bhagwat, P.
Bhalere, R. Gujar, S. Walunj.
"Tourist Place Recommendation
System."International Journal of

Electronic copy available at: https://ssrn.com/abstract=4286575