Social media data are increasingly used as the source of research in a variety of domains. A typical example is urban analytics, which aims at solving urban problems by analyzing data from di erent sources including social media. The potential value of social media data in tourism studies, which is one of the key topics in urban research, however has been much less investigated. This paper seeks to understand the relationship between social media dynamics and the visiting patterns of visitors to touristic locations in real-world cases. By conducting a comparative study, we demonstrate how social media characterizes touristic locations di erently from other data sources. Our study further shows that social media data can provide real-time insights of tourists' visiting patterns in big events, thus contributing to the understanding of social media data utility in tourism studies.
Today, our society is increasingly hyper-connected with the unprecedented rise
of social media; currently, social media has 2.206 Billion active users with 30%
global penetration1. Social media activities are therefore an important class of
daily activities performed by people worldwide to ful ll their social needs. These
social media activities have generated a wealth of social data, which can provide
meaningful and even possibly, real-time insights to a variety of studies. Social
media has thus been leveraged as part of marketing strategy for industries, through
\passive marketing" (as sources of market intelligence to gain insights of the
users) [
Among di erent domains, urban science has been shown as an important
domain where social media data can contribute [
On the other hand, the high availability of social media data raises some challenges. In order to retrieve relevant social data, one should take into account e ective parameters for data ltering, such as hashtags, keywords or geographic boundaries/coordinates. When it comes to tourism studies, the selection of parameters is crucial and needs to be carefully designed, to avoid biasing the interpretation of the results. For example, to capture the popularity of Ei el Tower in Paris w.r.t. the number of visitors, we should avoid only using keywords or hashtags (e.g. #Ei elTower, #TourEi el, or #Ei el) to lter the data, as we cannot assume whether people posting tweets with these hashtags are indeed currently visiting that location. In this case, geographic coordinates become necessarily to be used as an additional parameter for ltering social media data.
By carefully selecting the parameters for ltering social media data, we created a dataset to explore the potential of social media data for tourism studies. Speci cally, we focus on Instagram, which has been shown to be highly popular among tourists [22] as it features creating and sharing visual content (i.e. images) by users. We then seek to answer the following research questions:
RQ1. How does social media data characterize touristic location di erently from other data sources?
RQ2. Can social media data provide real-time insights about visiting patterns of tourists to di erent tourist locations in big events?
We tackle these questions by quantitatively and qualitatively analyzing the social media data we collected from Instagram. By comparing the social data results with o cial tourism statistics and online review data, we nd that social media characterizes touristic locations di erently, featuring more landmark locations. Our study further shows that, during big events, social media data can characterize the visiting patterns of tourists at di erent locations with high temporal resolution, thus contributing to the understanding of how social media data can be used for obtaining real-time insights for touristic locations.
Our study provides preliminary however useful results that support social
media as a useful data source for tourism studies. In 2014, World Travel Tourism
Council (WTTC) stated that tourism industry has contributed US$7.6 trillion
and also support 276 million jobs across the globe [21]. Because of the high
economical and societal relevance, governments and the tourism industry are
devoted to attracting more tourists to visit their cities' touristic locations [
Social media data analytics. The unprecedented popularity of social media
has o ered opportunities for a variety of domains. An important application is
user modeling, for which social media data has been shown to be useful for
modeling user attributes, including personal traits and personality [
Social media data in urban analytics. While little work can be found in
tourism studies, social data has enabled a wealth of research works in urban
analytics. These include using social data for event-detection [13], venue
recommendation for city-scale events [
Data sources for tourism studies. United Nations and European Union have
provided guidelines and recommendation regarding the comprehensive
methodological framework for collection and compiling of tourism statistics [
In 2015, France was one of the top touristic destinations over the world [20], with more than 83 million tourist arrivals. In addition, the city of Paris itself also ranked as the most visited city in the world [19]. Therefore, this paper will scope the sample of touristic locations within the city of Paris (France). This section introduces our method in creating the dataset for answering our research questions, including the o cial tourism statistics, TripAdvisor statistics, and Instagram data. 3.1
O
cial Tourism Statistics and TripAdvisor
We are interested in understanding the characteristics of data from social media,
compared with those from other data sources. To this end, we rst determined
the touristic locations considered in our study. We chose the top-5 touristic
locations based on their popularity in Paris' annual visitor statistic [
Comparing the popularity of touristic locations in the o cial report and in TripAdvisor, we can nd that touristic locations have di erent popularity between the one reported by o cial statistics and the one observed through TripAdvisor. In particular, while there are more visitors in Cathedral than Eiffel Tower according to the o cial report, Ei el Tower appears to be the one reviewed more often on TripAdvisor. In addition, di erent museums also show di erent popularity: Musee du Louvre is more popular than Musee d'Orsay in terms of visitors. However, people tend to review Musee d'Orsay more often on TripAdvisor. These observations clearly show a di erence between the reality (o cial statistics) and statistics from the online review platform (TripAdvisor). # Touristic Locations O cial Statistic TripAdvisor
Visitors Reviews 1 Notre Dame Cathedral 2 Musee du Louvre 3 Eiffel Tower 4 Musee d'Orsay 5 Arc de Triomphe 3.2 Instagram Data This work requires data acquisition from Instagram as the main object of study. Currently, Instagram has provided access to their API Endpoints2 through which developers are able to interact with its data. Unfortunately, due to the change of Instagram's platform policy starting on 17 November 2015, every newly created app won't have full access of the API Endpoint (Sandbox Mode)3. Developers are required to submit their applications through a review process to be granted will full API access, which requires a long period of time. Therefore, we developed an alternative method here to track real-time posts available on Instagram, that was, using its search function.
Instagram's search function enables us to nd photos based on keyword. Moreover, Instagram website provides \explore" feature that enables a user to explore trending and recent posts on a speci ed location based on the Location2 https://www.instagram.com/developer/ 3 http://developers.instagram.com/post/133424514006/ instagram-platform-update Id4. An example of the Location page is shown in Figure 1. By using webscrapping techniques we were able to retrieve photos and auxiliary information required for this study, including Timestamp, User-Id and Location-Id. Determine Relevant Location-Id. Instagram allows users to tag location information to photos. In doing this, users get automatically location tag recommendations based on their locations or photos' coordinates. We leveraged such location tags to lter relevant posts for this study. Particularly, we noticed that a single location can have multiple Location-Ids. For example, if we want to nd Instagram's posts located at The Ei el Tower we could nd it on Location-Id 216052603, as well as Location-Id 2593354, which is the French version of the name Tour Ei el. To cope with this problem, we proposed to merge multiple Location-Ids for individual touristic locations, in order to gain more data and to avoid biasing the interpretation of the results: including either only the English and the French version of the Location-Id of The Ei el Tower, the results will be biased to speci c type of people speaking English or French.
The discovery of available Location-Ids was performed as follows. First, we queried Instagram's posts based on keyword and/or hash-tags. Second, based on the retrieved posts we were able to identify a list of frequently used Location-Ids 4 E.g., The 218177821/.
Ei el Tower: https://www.instagram.com/explore/locations/ (> 100 posts) for speci c touristic locations. As a result, we were able to nd relevant Instagram's Location-Ids for each tourist destination considered in our study.
Tapping on Instagram's Recent Media Stream. After identifying the Location-Ids of every touristic location in our study, we built a crawler to scrap Instagram's explore page5 using Location-Ids as the parameters. As a result, we were able to retrieve 898,339 Instagram posts. However, when analyzing the retrieved data based on post's timestamp, we found that the number of photos older than one month decreases dramatically. This pattern could happen either because of Instagram's own policy that not revealing all old photos or because of the feature itself only just available since June 20156.
140000 120000 100000 s t so 80000 P # 60000 40000 20000Jan Feb Mar Apr May Jun Jul Aug Sep Okt Nov Dec
Month
As a result of this nding, we used only recent retrieved posts that were not older than 14 days (from 25 December 2015 to 7 January 2016) to ensure reliability. We leave it to future work, to scrap the website for a longer period of time. The resulting data set contains 82,381 posts7. 5 https://www.instagram.com/explore/locations/<location-id> 6 http://blog.instagram.com/post/122260662827/150623-search-and-explore 7 The dataset has been made available to the community at: https://github.com/ arkka/tourism-analytics
Arc de Triomphe
Musee d'Orsay Musee du Louvre
22.8% Notre Dame de Paris In order to answer RQ1, we rst construct a ranking list of the considered touristic locations according to their popularity in Instagram. Speci cally for each touristic location, we count the number of related posts, to be used as a measure of its popularity in social media. Next, we compare this ranking list with the one in o cial tourism report and that in TripAdvisor.
Comparing Metrics. To quantitatively compare the ranking list of touristic locations from di erent data sources, we used a well-accepted measure of nonparametric rank correlations, namely Kendall's rank correlation, which is based on pair-wise agreements between touristic locations. The result shows that there is no correlation (< 0:1, with p-value > 0:1) between the ranking list of o cial report and that of social media. Comparing with the ranking from TripAdvisor, we nd a moderate correlation between social media and TripAdvisor (= 0:59, p-value = 0:14). We next qualitatively analyze the similarity and dissimilarity among data from di erent sources, as we will see later.
Social Media vs. O cial Statistic. It is interesting to see how the relative ranking positions of di erent tourist locations di er in social media list and the o cial list of tourism statistics. In particular, o cial tourism statistics reveals that Notre Dame Cathedral and Muse du Louvre have more visitors than The Ei el Tower, however in social media the most popular one is The Ei el Tower for both Instagram and Tripadvisor. A possible reason for this would be that tourists are more opt to take a photo on landmark touristic locations without entering the attraction such as museums or monuments. Following this kind of scenario, we are able to leverage social media data to complement existing o cial statistics. Social media is able to identify trends independently without the needs of conventional visitor counting methods, such as using ticket sales or gate counters. Moreover, social media also able to provide high temporal resolution data that we can leverage to enhance our analysis which we explore in the later section.
Instagram vs. TripAdvisor. Comparing the popularity of touristic locations between TripAdvisor's user reviews and Instagram's posts, we nd that there are certain amount of similarities between these statistic . Both of these statistics rank The Ei el Tower as the most popular touristic locations, followed by Musee du Louvre and Notre Dame Cathedral respectively. The di erence can be found for Musee d'Orsay and Arc de Triomphe, which respectively rank at the fourth and fth in TripAdvisor, but it is the other way around on Instagram. Overall, it is interesting knowing the fact that these two di erent data sources can relate with each other. The similarity between TripAdvisor and Instagram data could be because that both are user-generated. In the following, we will analyze in detail the utility of social media data in providing ne-grained temporal insights for tourism studies. 4.2
Evolution of Touristic Location Popularity During New Year's
Eve
Social media has been shown to be e ective as sensors for detecting collective
trends of societal events [
Figure 4 shows the dynamics of the popularity of the top touristic locations in Paris within two weeks from Dec. 25th, 2015 to Jan. 7th, 2016 on a daily base. We can observe di erent visiting patterns of these considered touristic locations, according to the evolution of popularity before, during, and after New Year's Eve. In particular, the popularity of The Ei el Tower and Arc de Triomphe drastically increases during New Year's Eve in terms of the number of Instagram posts. In contrast, the popularity of Musee du Louvre deceases during New Year's Eve. Therefore with social data we could draw the conclusion that The Ei el Tower and Arc de Triomphe are the ones attract more social activities from visitors during the big event, while Musee du Louvre loses social attention during the same period of time. A possible reason could be that people are more likely
Arc de Triomphe Musee d'Orsay Musee du Louvre Notre Dame de Paris
Tour Eiffel
Fig. 4: Daily number of visitor trends on 2016 New Years Eve. 445000 AMMruucssdeeeee Tddr'uOioLrmsoaupyvhree 350 TNooutrreEDiffaemle de Paris 300 New Year's Eve ts250 s o P#200 150 100 50 208/12/2015 29/12/2015 30/12/2015 31/12/2015Hour 1/1/16 2/1/16 3/1/16 4/1/16 (a) Dec. 25th, 2015 { Jan. 7th, 2016 450 400 350 300 ts250 s o P#200 150 100 50 000:00 04:00 08:00 12:00 16:00 20:00 0H0:o00ur 04:00 08:00 12:00 16:00 20:00
Arc de Triomphe Musee d'Orsay Musee du Louvre Notre Dame de Paris Tour Eiffel
New Year's Eve (b) 2016 New Year's Eve. to celebrate big events in landmark attractions (e.g. The Ei el Tower and Arc de Triomphe) than museums. We leave it to the future work to nd out more characteristics of touristic locations according to their corresponding visiting patterns during big events.
Daily Visiting Patterns of Tourist Locations during New Year's Day Social media data allow us to analyze ne-grained patterns of user activities with high temporal resolution. We now analyze the visiting patterns of people to touristic locations on hourly base.
Daily Visiting Pattern. Figure 5a shows the visiting frequency of people to touristic locations during the week centred at New Year's Day, observed through Instagram. we could observe that in normal days (excluding New Year's Eve) number of Instagram's posts peaked at 9-10PM on each day. In contrast, we have found a drastic change of daily visiting pattern on New Year's Eve, which is moved earlier to 6PM on 31 December 2015 and followed by another peak on the next day at 1AM with a huge number of posts compared to the rest of other peaks. After the New Year's Eve (1 January 2016) we could observe that the daily visiting pattern comes back normal as usual in the evening, which is peaked at 9PM. These observations indicate a distinct visiting pattern of people to touristic locations on New Year's Day, di ering from other days before and after. Moreover, this gure also shows the distribution of each post by touristic locations. Overall, The Ei el Tower consistently has the highest number of posts compared others, which followed by Musee du Louvre and Notre Dame Cathedral respectively.
Using high temporal data that we have retrieved from social media, we are able to relate the number of posts on speci c touristic locations with their opening hours. For example, the opening time of Musee du Louvre is from 9am to 6pm every day except on January 1, May 1 and December 258. Based on the hourly visitor trends depicted on gure 5b, we could see that the number of posts increases dramatically on 9am until reaching it's peak on 6pm and then decreases dramatically when reaching the closing hour on 6pm. We could interpret that as during New Year's eve, people tend to visit a touristic location that has no restrictive opening hours, which usually is an outdoor touristic location, such as The Ei el Tower and Arc de Triomphe, instead of museums in Paris such as Musee du Louvre that are closed on the 1st January 2016. Therefore, it is clear that social data is able to provide additional data and insights that complement the existing o cial statistics.
Visiting Patterns of Di erent Touristic Locations on New Year's Eve. Based on the previous analysis, we have noticed that there is a di erence visiting pattern on New Year's Eve and normal days. On gure 5b, we focused our analysis from 31 December 2015 to 1 January 2016 to identify the trend. Consistent with previous results, this gure shows that on New Year's Eve people are more likely to celebrate at The Ei el Tower and Arc de Triomphe. Moreover, Arc de Triomphe become the second-most popular touristic location for celebrating New Year's Eve. In contrast, Musee du Louvre which is usually the second-most popular touristic location on a normal day becomes less popular on New Year's Eve. These observations indicate that people tend to be selective in celebrating the big event, favouring more at landmark locations than museums. 5
This study aims to explore the potential of social media data in tourism studies. The main limitation (and a threat to validity) is the size of the data we collect. 8 http://www.louvre.fr/en/hours-admission/admission This relates to several important factors, including the length of the time period of the data, the number of city and touristic locations we consider in the data. It would be highly interesting to compare the touristic locations in di erent cities, which we leave for the future work. Second, due to Instagram's Public API limitation, we couldn't explore the full potential of Instagram data using the API. In this paper, we stress our e ort in creating the trustable data crawler, and testing the reliability of the data. 6
While social media has open many research opportunities in urban analytic, the value in the tourism studies has been much less explored and remains an open question. This paper takes a rst step towards understanding the utility of social media data in urban tourism studies. We show that social media can provide di erent re ections of the tourism from other data sources, including o cial tourism report and online review platforms (e.g. TripAdvisor); moreover, we nd that social media data can provide real-time insights of tourists' visiting patterns during big events. In conclusion, our study provides a set of preliminary results that support social media as a useful data source for touristic marketing and decision making. 12. G. C. J. Park and E. Ferrara. Sytle in the age of instagram: Predicting success within the fashion industry using social media. 2015. 13. R. Lee and K. Sumiya. Measuring geographical regularities of crowd behaviors for twitter-based geo-social event detection. In Proceedings of the 2nd ACM SIGSPATIAL international workshop on location based social networks, pages 1{10. ACM, 2010. 14. W. Mangold and D. Faulds. Social media: The new hybrid element of the promotion mix. 2009. 15. M. Mathioudakis and N. Koudas. Twittermonitor: trend detection over the twitter stream. In Proceedings of the 2010 ACM SIGMOD International Conference on Management of data, pages 1155{1158. ACM, 2010. 16. J. Miguens, R. Baggio, and C. Costa. Social media and tourism destinations:
Tripadvisor case study. Advances in Tourism Research, 26(28):26{28, 2008. 17. P. O'Connor. User-generated content and travel: A case study on tripadvisor. com.
Information and communication technologies in tourism 2008, pages 47{58, 2008. 18. United Nations. International Recommendations for Tourism Statistics, 2008. 19. UNWTO. UNWTO Tourism Highlights: 2015 Edition. 2015. 20. UNWTO. Unwto world tourism barometer. 14, May 2016. 21. WTTC. Travel & Tourism: Economic Impact 2015. 2015. 22. J. Yang, C. Hau , G.-J. Houben, and C. T. Bolivar. Diversity in urban social media analytics. In Web Engineering, pages 335{353. Springer, 2016.