Revealing hidden treasures of museums based on public preferences: A case study

(1)

Revealing hidden treasures of

museums based on public

preferences: A case study

Casper E.C. Broertjes 10421300

Master thesis Credits: 18 EC

Master of Science in Information Studies, Business Information Systems University of Amsterdam

Faculty of Science Science Park 904 1098 XH Amsterdam

Supervisor Prof. dr. Marcel Worring

July 23rd, 2016 ‘

(2)

1 Abstract

Digital technology allows museums to become more active on the web. By making use of digital collections and social media, museums attempt to share their cultural heritage with the public in new ways. The shift from a physical museum to a digital museum allows data analysts to gather new types of data, which can be used to analyse public preferences. This research attempts to make use of heterogeneous data sources to find out more about public preferences of art objects in museums. By means of a case study at the Rijksmuseum an attempt was made to find out the added value of the analysis of web traffic and social media use of visitors. It was found that the most information about public preferences could be gathered from web traffic analysis and less information from analysis of social media like Instagram and Facebook. Web traffic analysis provided complementary information about the popularity of art objects which are not on display (hidden treasures) in the museum. Over 30% of the mostly visited art objects were hidden treasures. Also, it was found that web traffic preferences do not always correspond with social media preferences.

(3)

2 Introduction

The technological environment of museums has changed over the years, allowing them to become active on the web. Digital technology gave them the oppor-tunity to reach a broader audience in new ways, by digitising their collection and archives or providing extra services (Throsby et al., 2010). The use of the Internet has been argued to be one of the most important trends in the sector of museums (Pulh et al., 2008). This was already identified in the annual Mu-seums and the Web conference in 1997, where the growing interest in utilizing the Internet as a cultural information sharing tool was introduced.

Visitors of museums have changed in the sense that they want more than ’just’ a physical exposition of artwork. To overcome the limited space of a phys-ical museum, the Web can provide a platform for a digital museum offering complete collections of museums. Instead of artwork being hidden in the base-ments of museums (’hidden treasures’), it can now be made publicly available on the Internet.

A digital museum provides a museum with both challenges and opportuni-ties. Challenges are for example the representation and categorization of the large multimedia collections. Data can be collected by tracking activity of vis-itors in the digital museums, which provides museums with new opportunities. These opportunities lie in the analysis of public preferences, which is a major part of what our research focuses on.

Furthermore, the technological environment has changed to be a participa-tory environment of visitors within social media. Facebook, Instagram, Twitter and other media are places where visitors can share their interests and prefer-ences with other users. Analysing social media may also provide us with insight into public preferences, which is also a focus in this research.

This research is aimed at finding out how online channels can help to under-stand public preferences of art in the domain of museums. With this information an attempt will be made to reveal hidden treasures to museums, by providing an overview of popular hidden treasures. By means of a case study at the Ri-jksmuseum, multiple channels will be analysed. Since the Rijksmuseum has a large digital collection, user activity in this collection will be analysed. Also, social media channels like Instagram and Facebook will be analysed for public preferences. As an extra, this paper proposes an information visualization for museums to navigate through their data, to gain a better understanding of pub-lic preferences. The following research questions summarize this paper’s main objectives:

RQ1: How can we use heterogeneous data sources of museums to gain insight in public art preferences?

RQ2: For each data source, what is the added value when analysing public art preferences?

To answer these questions, this paper has the following structure. First of all, a literature review will be done to create an overview of the current state

(5)

of museums on the Internet. This will be done by looking at both how muse-ums have become digital musemuse-ums and by looking at how musemuse-ums currently use social media. After this, a case study will be introduced, followed by the proposed research methods, concluding with the results and conclusions.

3 Related work

3.1 Digital museums

Before the emergence of the Internet, the most important activities of muse-ums were exhibitions and research. Since musemuse-ums understand the relevance of visitor needs, their core mission is shifting, by adding new activities to their collection. Anderson (2004) claims that museums shift from collection-driven institutions to visitor-centered museums, which was also stated by Vergo (1989), who called this the “New Museology”. This resulted in museums becoming fo-cused on a combination of education and entertainment, which was labelled as “edutainment”. As a result of this shift, Ross (2004) claims that museums should try to reach a wider public.

3.1.1 Importance of online presence

Almost a decade ago research already showed that online presence of museums was becoming more and more important (Marty, 2008). Even before the emer-gence of the Internet it was proposed by Hooper-Greenhill (1999) that the main mission of museums should be the dissemination of information rather than pre-serving artefacts. Over the last years the Internet has become the main conduit of information, which indicates the importance of this medium. When looking at artefacts from an information transmission perspective, there is a negligi-ble difference between the museum object and their digital representation (as long as there is no perceivable loss in the transformation) (Besser, 1997). In that sense, it seems logical for museums to become active on the web, since the experience may be more or less the same as in reality.

Considering the typology of websites created by Hoffman et al. (1995), mu-seums can be seen as “Internet presence websites”. For this type of websites the main objective is not sales, but is about advertising and providing information Hoffman et al. (1995). Since the actual sales are at the physical museum, the website is more of a medium to lure visitors to the museum. This statement is supported by Lagrosen (2003), who researched the use of Internet of Swedish museums. He explained that the visit and experience of visitors is the real prod-uct of a museum. Therefore, stimulating visits is an important goal of a museum website. However, it could be argued that when museum websites become too elaborate, this digital museum may become a substitute of the physical museum. However, research by Pallud and Straub (2014) showed that a positive attitude towards a museum website did not influence intentions to visit a museum in real life. It did however positively influence the probability that a visitor would return to the digital museum. Belanger et al. (2006) state that the goals of

(6)

Table 1: Visiting museum website before visiting museums (Marty, 2008) Very unlikely Unlikely Neutral Likely Very likely In general, before

visiting a museum, how likely are you to visit the museum’s website?

3.2% 4.5% 10.3% 29.2% 52.7%

Table 2: Visiting museum website before visit in specific situations Marty (2008)

In each of the following situations, how likely are you to visit a museum’s website before visiting the museum?

Very unlikely Unlikely Neutral Likely Very likely

You are visiting a museum you visit regularly 5.4% 16.0% 19.1% 37.8% 21.7% You are visiting a museum unfamiliar to you 2.1% 2.7% 6.1% 21.7% 67.3% You are visiting a museum by yourself 3.3% 4.7% 14.0% 32.6% 45.3% You are visiting a museum with friends or family 2.6% 4.0% 13.0% 34.9% 45.6% You are visiting a museum in your home town 5.5% 12.7% 20.1% 37.8% 23.9% You are visiting a museum while traveling or on

vacation 2.9% 7.1% 9.1% 21.4% 59.6%

You are visiting a museum where the exhibits

change frequently 2.0% 2.6% 7.2% 27.1% 61.1%

You are visiting a museum where there is too

much to see in one visit 2.3% 3.1% 11.1% 27.5% 56.0%

museums also include: life enrichment, online learning, knowledge enhancement and entertainment (Fantoni et al., 2012).

From the visitors’ perspective an exploratory research by Marty (2008) was conducted, where 1200 visitors at nine different museums were asked to fill in a survey about the role of museum websites in their lives. A majority of the respondents state that they use museum websites often and explain that they have clear expectations of the interactivity on museum websites.

Table 1 shows responses from respondents in this survey when they were asked how likely they were to visit the museum’s website before their actual visit. According to survey, a vast majority of the visitors were very likely to visit the museum before their visit (81.9%).

Table 2 shows responses from when visitors were asked how likely they were to visit the website in specific situations. In almost all circumstances visitors were likely to visit the website before their visit.

The survey also had questions about what kind of resources visitors were likely to use. These responses are summarized in table 3. Prior to visiting a

(7)

Table 3: Use of online resources before visiting museums Marty (2008) When visiting a museum’s website before visiting the

museum, how likely are you to use each of the following resources?

Very unlikely Unlikely Neutral Likely Very likely

Online images of artefacts/collections data 3.8% 9.5% 17.4% 33.2% 36.1% Online tours of galleries/interactive exhibits 4.7% 13.4% 20.9% 34.4% 26.6% Online educational activities/learning resources 6.7% 19.5% 30.7% 26.0% 17.2% Online research materials/archives 5.1% 16.0% 24.9% 25.7% 28.2% Information about hours of operation/location/directions 1.9% 1.5% 3.9% 20.4% 72.3% Information about admission fees/pricing 2.5% 2.8% 7.5% 22.0% 65.2% Information about museum facilities/gift shop/restaurants 3.4% 3.8% 13.1% 31.8% 47.9% Information about current and future exhibits 2.5% 1.1% 4.5% 27.6% 64.2% Information about programs/tours/special events 3.4% 3.8% 13.1% 31.8% 47.9% Information about employment/volunteer opportunities 22.7% 25.2% 27.6% 13.4% 11.0% Information about donation/membership opportunities 15.5% 21.1% 36.2% 19.9% 7.2%

museum, visitors were most likely to go to the website and gather information about opening hours, driving directions and other basic information, rather than online images about art objects, online gallery tours or online educational activities. However, visitors were still relatively likely to visit the website for the latter purposes (69% was either likely or very likely to do so). This percentage becomes even higher after a museum visit (72.4%).

By studying the previously mentioned literature, it can be seen that mu-seum websites are popular among mumu-seum visitors and that online presence of museums is increasingly important. This is why research on this topic can be relevant and important. An interesting observation is the high percentage of museum visitors who are likely to search for digital content as well. A vast majority (69.3%) is either likely or very likely to search digital content. 61.% of the visitors are even willing to take an online tour. These observations are interesting for multiple purposes, like revealing hidden treasures to the public or learning more about public preferences in general.

3.1.2 Representational practices

To answer the need of these visitors, museums need to become active on the Internet. Not only for general information about a visit, but also for giving information about art objects in their museums. This requires the representa-tion of large quantities of images. Representing large collecrepresenta-tions of images for museums requires categorization. What one may expect is the usual grouping: artwork from a specific time period or a categorization by artist. However, so-cial tagging is becoming increasingly popular as well. When focusing on search in a digital collection, Smith (2006) states that social tagging can increase ac-cess to images for audiences with a wide range of expertise. (Klavans et al.,

(8)

2014) proposed a matrix for categorizing art. The basis for this matrix was developed by Armitage and Enser (1997). In essence these categorizations are: Who, What, Where and When questions. However, the authors felt that there still was some information missing, namely visual aspects like shapes, forms and colours. Furthermore, they added an "other" category for items that do not fit any category.

"Adopting digital images, multimedia and broadband network is transforming the nature of collections and how the public views collections" (Anderson, 1999). Tang (2005) explored the semiotic aspects of this transformation. He proposed the following three levels of representation.

Firstly, he speaks of the narrative-centered mode of representation. This mode has the intent to tell a story, where the cultural artefacts present evidence to the storyline. The objects are presented so that they invoke imagination associated with the narrative. From the three levels of representation, this structure seems to be most suitable of conveying a message. The narrative-centered mode is also discussed by Segel and Heer (2010). They conducted an analysis of narrative visualizations intended to convey stories. Their analysis identified distinct genres of visualization, like the martini glass, interactive slide show and drill-down story. The basis for this distinction lies in the balance between author-driven elements and reader-driven elements.

The second mode is object-oriented. This mode comes the closest to the traditional sense of a museum, where objects are mostly accompanied with a literal explanation of the object. This allows the viewer to both grasp the aesthetic value as well as the cultural significance of the piece. So in short, both the integrity as well as the intrinsic value is carefully stored.

The last mode is an information-centered mode of representation. Here the representation of images is illustrative rather than symbolic. The images either function as examples of a group of objects or as a statistical representation in the form of a graph or chart. Not much is stored about the cultural significance of the objects. This representation mode is used more often in natural science or for presenting scientific phenomena.

The representational practices discussed in this section will be useful when creating an information visualization.

3.2 Social Media and museums

Over almost a decade, social media has been a way for organizations to strengthen their relationship with their stakeholders. This has resulted in the change of strategies on policy development within organizations like museums. Since au-diences continuously consider how to spend their time and money carefully, museums have to find ways of communication to keep up with the audience by sending personalized messages (Liebenson, 2009). This section will explain more about the current state and emerging use of social media within museums and will also explain more about the challenges that museums are facing when dealing with social media.

(9)

Table 4: Social network survey results at Museum of Natural History in Florence (Lazzeretti et al., 2015)

Social network Year of entrance Objectives Staff ’s potential perception

Commitment (time devoted) Trip Advisor 2009 1. Increasing visibility for potential visitors Limited Limited

Flickr 2010

1. Promotion of collections and additional activities, also through the collection of user-generated

contents

High Average

YouTube 2010

1. Promotion of additional activities, also through the collection of user-generated contents

Average Limited

Facebook 2010

1. Information regarding collections and additional activities

2. Managing and strengthening relations with visitors

3. Connection with scientific educationnet works

High High

Twitter 2010

1. Information regarding collections and additional activities

2. Managing and strengthening relations with visitors

High High

Four Square 2012 1. Increasing visibility for potential visitors Cautious (new) Limited

3.2.1 Current state

For types of organizations like museums, involvement with social media has been argued to be very important for survival of the organization. Without strong ties with the visitors, volunteers and donors, museums may cease to exist. Therefore it may be particularly important for museums to stay active on social media and keep strong relations with the audience. Museums have started to use many kinds of social media, like: social networks, bookmarking, pod casting and blogging (Russo and Peacock, 2009). This emergence has its origins in the beginning of Web 2.0, where users can contribute to, rate and customize content found on the Internet.

Researchers have investigated the quality of social media content and they have shown that social media is more of an instrument for traditional communi-cation rather than user engagement. When Schick (2010) investigated Facebook profiles of Danish art museums, he discovered that the quality of user gener-ated content is generally limited and rarely goes beyond small talk. It lacks any theoretical or cultural importance. They stated that Facebook is rather used as a way for museums to reach the audience, rather than interacting with them. These results are confirmed by a survey of Fletcher and Lee (2012) about American museums. In their survey social media was also seen as a one-way

(10)

communication, promoting events, promotions or announcements to reach large audiences. If user generated content on social media is usually rather limited, this could have implications for the added value of this content when analysing the medium for public preferences. Taking the one-way communication aspect of Facebook into consideration, the quality of posts by organisation might be higher, but does not help with finding out more about public preferences.

In research by Lazzeretti et al. (2015) a social media analysis has been done at the museum of natural history in Florence. Interviews were conducted with museum staff to learn more about their social media strategy. In table 4 an overview has been given on the museum staff’s view of the objectives, potential and commitment for each social media channel. To get a feeling of how museums may use social media, the following paragraphs will discuss how this museum is involved with these social media.

TripAdvisor is a travel portal where users can share information and reviews about accommodations, attractions and restaurants. TripAdvisor is one of the largest information sources for global travellers, which has not gone unnoticed by hotel and attraction managers. These managers often view what is being said on TripAdvisor as an indicator of their quality. Four Square is a social network where users can share their locations via check-ins at accommodations. Four Square is limited compared to TripAdvisor when looking at interaction opportunities for museums. Overall, it was found in the survey that these social media channels were used as a ‘virtual visit card’. However, the museum’s staff stated that they were not yet taking full advantage of these mediums. They also perceived the channels either limited or cautious.

Youtube is the largest social network for video sharing, where users can upload video’s, comment to video’s and view them. The museum has opened its own YouTube channel and uploaded a collection of video’s which were viewed about 50,000 times. Flickr is similar to YouTube since it is a video and photo sharing platform. One of the most important feature offered by Flickr is social tagging (adding keywords which describe photo’s). The museum has uploaded 153 photo’s which have been viewed a limited number of 550 times. Despite the efforts of the staff, the number of followers and views is still limited. In table 4 it can be seen that the potential is perceived as high, which is why they do spend much time on this social media.

Based on this research we can see that this museum and probably museums in general are continuously trying to make use of a wide range of social media. 3.2.2 Benefits and challenges

There is no single way to calculate the value of social media within an orga-nization. This is mainly due to the fact that each organization has a different definition of social media success (Peters, 2008). There are however several benefits which are mentioned frequently.

Firstly, the fact that there are no fees or space limitations for online storage for most of the social media websites means that costs (C) are relatively low (Black, 2005). Secondly, social media provide a way to access (A) a large

(11)

audi-ence. This can be seen based on the fact that 67% of the online population visit online communities. This increased access provides the opportunity to increase their communication reach (R). Not only reach is increased, but also speed (S). Museums and organizations in general can send messages instantly. Solis (2007) argues that possibly the most important opportunity of social media is engagement (E) of the public.

Beside these benefits, also some challenges arise concerning issues like trans-parency (T), liability (L), credibility (C), time (T) and privacy (P) (Wright and Hinson, 2008). Peters (2008) states that one of the largest challenges is time, since a correct implementation of social media requires time and dedication. Content now has to be updated regularly to maintain an active online commu-nity. This requires time for the staff to become acquainted with effective social media usage.

Since presence on social media seems to be a logical step for every museum seeing the mentioned benefits, it is now possible to use these media for analysis for museums in general. The analysis may prove that social media have another benefit, namely discovering more about public preferences about art.

4 Case study

For the purpose of finding out more about public preferences via the digital channels (website and social media) which are discussed in the related work, we will consider a case study at the Rijksmuseum in Amsterdam.

The Rijksmuseum in Amsterdam is one of the largest museums in the Nether-lands. Their collection is dedicated to art and history in Amsterdam and con-tains over 500.000 pieces, among which there are some very famous masterpieces by Rembrandt, Vermeer and Frans Hals. The art collection consists of various types of items like paintings and sculptures. The museum has about 6.000 pieces on display, which means there is a large amount of art not available for visitors to see physically. To allow those who are interested to view the collection any-way, the Rijksmuseum had decided to develop a website. This reward-winning website enables visitors to browse through their collection, via different facets. One can browse through some of the most popular pieces, or decide to explore their collection through a time line. Daily there are over 20.000 visitors on the website, either browsing for information for an actual visit to the museum, or for browsing through their art collection. Furthermore, the Rijksmuseum is active on social media like Facebook and Twitter. They post art almost daily. These posts are shared and liked by thousands of users in the active community of the Rijksmuseum. At the Rijksmuseum there is a gap between what is known about the online preferences of their community and what is actually displayed in the museum. Since there is a large amount of art pieces available but not visible for visitors, there is a need at to gain insight in the community to better utilize the potential of the collection. This research therefore attempts to create an overview of these preferences, by making use of the media available (website and social media).

(12)

Table 5: Rijksmuseum’s digital collection - Extracted meta data Meta data Description

ID A unique identifier for each object Title The title of the object

Web image URL A link to the image of the object

Artist The name of the artist who created the object Type The type of the object, like: painting, sculpture etc. Period The age in which an object was created

Place Where the object was created

On display Shows whether or not an object is hidden.

5 Methods

By gathering data from multiple sources, an attempt will be made to create an overview of public preferences at the Rijksmuseum. For each data source the added value to the knowledge about public preferences will be analysed. Public preferences will be measured by analysing web traffic on the digital collection of the Rijksmuseum and by analysing activity on social media. Preferences between the different sources will be compared, so that patterns or similarities between them can be identified.

To gain more insight into what kinds of art is popular among visitors an information visualization will be created, where there is the possibility to filter art objects by various categories, for example: type of object, artist, period of creation, place of creation and whether or not an object is on display. These preferences can be viewed by country. The data collected for all data sources is over an interval of one year (2015-2016).

The following sections will discuss what kind of data will be gathered and how the data will be gathered for each data source.

5.1 Digital collection

The Rijksmuseum has created an online platform where visitors can explore their digital collection freely. Their online collection consists of over 500.000 art objects, each having a web page with (if available) an image and a description. To help visitors browse through the collection, filters can be applied to search by: makers, object type, period, place, material, technique and more. The Rijksmuseum supports a live website API, which makes the website directly accessible to developers. It allows developers to search the collection and obtain lots of meta data about art objects. The information in table 5 shows which relevant information is gathered from the API.

Not for all objects all information is available or known to the Rijksmuseum. Therefore, sometimes null values are given by the API. Since this information might still be of value (for example: do visitors prefer objects which are made by unknown artists?), these objects will also be included.

(13)

In the dataset a distinction is made between objects which are on display, which are not on display and top pieces. This research will refer to these types with the following definitions:

• Hidden treasures - Those artworks which are not on display in the physical museum, but are offered in the digital collection.

• Revealed treasures - Those artworks which are on display in the physical museum.

• Top pieces - Those artworks which are labeled as a top piece by the Ri-jksmuseum.

• All treasures - All artworks, including hidden treasures, revealed treasures and top pieces.

A vast majority of the art objects in this dataset are hidden treasures. Only about 6000 of them are on display, thus revealed treasures. The Rijksmuseum has selected 91 art objects as a top piece. Top pieces are displayed in the museum and also displayed more prominently on the Rijksmuseum’s website.

5.2 Web traffic analysis

We want to get an insight in public preferences by analysing web traffic on the Rijksmuseum website. The following method will explain how page views will be used in order to learn more about these preferences. The Rijksmuseum makes use of Google Analytics (GA), which keeps track of visitor activity on their website. By means of the Google Core Reporting API, this data can be extracted and prepared for research. GA allows us to categorize user activity by numerous dimensions and metrics. For this research the metric used will be page views.

The number page views are seen as indicators of preferences in this research. If users are browsing through the website of the Rijksmuseum and they see an image they like, it can often lead to a click to that page. Also, users might have searched for this art object specifically via the search bar, indicating they had a particular interest in the object already. A high number of page views does not always mean an object is of a high artistic quality, but it does mean that users are interested in seeing the object. This is why this metric is taken as the main indicator for public preferences.

To filter out page views which are irrelevant for this research, GA allows us to filter by certain URL structures. Therefore, only URL’s containing the following string will be used:

• "/en/collection/" • "/nl/collectie/"

(14)

Table 6: Instagram posts with "Rijksmuseum" - Extracted meta data Meta data Description

ID A unique identifier of a post

Likes The number of likes given to the post Web image URL A link to the image of the object Post URL A link to the post itself

Caption The entered text within a post

In order to learn more about public preferences of art objects, it is necessary to filter out page views that are irrelevant for this purpose. If we only take into consideration URL’s with the previously mentioned structures, we can make sure we only measure what we want. Since the Rijksmuseum provides an English and a Dutch version of the website, both structures will be used. One art object may have a Dutch URL prefix and an English URL prefix, which is why results for both URL’s will be combined later on.

Another important characteristic of the URL of an art object is the fact that the unique identifier of the object is present. This allows us to extract the identifier and link it to the Rijksmuseum API, so that meta data can be acquired.

5.3 Instagram analysis

Instagram is a social network for photo-sharing, video-sharing and social net-working. Users can upload photos and videos and attach a description to it. These descriptions usually contain "hash tags", which are important keywords with a hash tag prefix. To find out more about how Instagram can help to find out more about public art preferences at the Rijksmuseum, these posts and captions are scraped. The Instagram API is used to gather this data. The API allows us to search for posts containing a particular hash tag. In this research one hash tag will be used, namely: "Rijksmuseum". Over 20.000 Instagram posts will be scraped using this tag. These posts go as far back as the beginning of 2015. The information extracted from Instagram is shown in table 6.

To identify which art object is the subject in an Instagram post, an exact match of the art object title has to be present in the caption. Using this (simple) method, an attempt is made to find Instagram posts and Instagram images that have the searched art object as a subject. Different titles will be used per art object, since the Rijksmuseum provides multiple titles per object. Some titles contain a string saying: "Known as", followed by the most popular name of the object. An example of this is: "Militia Company of District II under the Command of Captain Frans Banninck Cocq, Known as the ‘Night Watch’, Rembrandt Harmensz. van Rijn, 1642". This sub string (the ’Night Watch’) will also be used for matching posts to art objects. All titles are converted to lowercase characters and all non-alphanumeric characters are removed so that the highest chance of a hit is ensured.

(15)

5.4 Facebook analysis

The Rijksmuseum is active on Facebook and has recently reached 300.000 fol-lowers. They post art objects, events and other posts almost daily on their profile with many followers responding to them with likes or comments. In contrast to the Instagram API where posts from all users can be gathered, the Facebook API only allows to search through one profile at a time. This is why this research focuses on the Rijksmuseum’s profile only. When the Rijksmuseum posts an art object on it’s profile, the first comment always contains a link to the given object. For this research only posts with this kind of structure will be used, so that it can be made sure that a post is about a given object. To compare objects to each other, the number of likes and the number of shares will be analysed. Comparison is done by comparing the rank of posts, rather than comparing the absolute numbers. This way we can normalize differences between shares and likes, since likes are usually a lot higher than the number of shares.

6 Results

The results section is divided into three parts. The first section will discuss the web traffic analysis, the second the Instagram analysis and the third the Facebook analysis.

(16)

6.1 Web traffic results

The results of the traffic analysis are divided into the three components: hidden treasures, revealed treasures and top pieces. In the appendix, separate lists are given which provide an overview of the top hidden treasures (Appendix A), top artists (Appendix B), top art object types (Appendix C) and the top of all treasures (Appendix D). For more details on these results, the live information visualization can be consulted (http://thesis.cbwebdevelopment.nl/). More on how to use the information visualization can be read later in this paper in the Information Visualization section.

6.1.1 Hidden treasures results

The analysis of hidden treasures showed that 42.715 art objects were visited on the website, which were not displayed in the museum. The most popular hidden treasure has 1240 page views. Compared to page views for all treasures, this object is at rank 15 in the top mostly viewed objects. Of the most popular art objects with page views higher than 100 (304 objects) 107 objects are hidden treasures (35.2%). The page views distribution of the hidden art objects are displayed in 1.

Figure 1 is a snippet of the entire viewed hidden treasure collection. When looking at the results of the page views for the entire hidden treasure collection, we see that the 2% top viewed paintings contain 25% of all page views. The other 98% of the paintings have 20 views or less. 94% of all hidden treasures have 10 or less views. Figure 2 shows a collage of the top 7 hidden treasures that were found.

6.1.2 Revealed treasures results

A total of 1915 revealed treasures were visited on the website. According to the API there are 6768 art objects on display, which means that about 28% of all revealed objects are visited online.

Figure 3 shows a similar graph as the one for the hidden treasures. About 27% of all visited revealed treasures contain 95% of all page views. Most inter-estingly is that the top viewed object ("The Night Watch" by Rembrandt van Rijn) contains 50.000 of the total 188.000 page views and the number two ("Het Melkmeisje" by Johannes Vermeer) contains almost 26.000 views. 198 objects have 100 or more views and 75% has less than 20 views.

6.1.3 Top pieces results

The results of top pieces and revealed treasures are quite similar, which seems logical since top pieces and revealed treasures overlap. Of the 91 top pieces selected by the Rijksmuseum, only 71 are viewed more than 100 times.

The same two objects as in the revealed treasures section are the most pop-ular. To compare the top pieces to what is actually most popular on the web, figure 4 shows an overview of the mostly viewed paintings (both hidden and

(17)

Figure 2: Top hidden treasures collage - Gives a quick grasp of the 9 mostly viewed hidden treasures

Figure 3: Revealed Treasures - Page views on the Rijksmuseum website.

Table 7: Page views comparison between top pieces & all pieces Group Average Mode Median Min Total All pieces 1459 50167 271 363 154781 Top pieces 1719 50167 643 8 131.355

(18)

revealed treasures). Here we can see that there are many pieces not labeled a top piece, despite being very popular on the Internet. Appendix D shows an overview of these paintings. A comparison of the statistics of these two groups can be seen in table 7.

Figure 4: Top pieces - Page views per art object

6.2 Instagram results

From the 20.000 Instagram posts that were crawled via the Instagram API, only 96 matches to an art object were found. In figure 5 an overview is given of these matches. In total there were matches for 28 different art objects. There were two outliers where there were many different matches. These two posts are the same two posts that were outliers in the web traffic analysis ("The Night Watch" and "The Milkmaid"). Figure 5 shows a combination of web traffic and Instagram matches. It is visible here that most art objects have very small amount of matches and page views. On average there were 3.4 posts about all found art objects that had a post on Instagram. The median is 1 match and the mode is 44 matches. Compared to the analysis of the traffic on the website, there is a similar trend where a few objects have a high percentage of all hits and many only have very few hits.

6.3 Facebook results

A total of 293 Facebook posts were extracted from the Rijksmuseum’s profile. For all of these posts a single art object could be identified as the topic of the post. On average a post had 957 likes and 160 shares. In figure 6 we can see

(19)

Figure 5: Instagram posts compared to page views per art object on the website

that when the number of likes decreases the number of shares most of the time also decreases (with some outliers). The top 2 Facebook posts with most shares is different from the other data sources analysed before, namely "Naval battle near Gibraltar" by Adam Willaerts (5376 likes, 2584 shares) and "Portret van vier Ajaxspelers" by Paul Huf (5924 likes, 1274 shares) are most popular on Facebook. The minimum number of likes given to an art object was 54 and the minimum number of shares was 7.

If we compare Facebook shares with page views (figure 7) for a given object, there does not appear to be any consistency between objects with many shares and objects with many page views. For some objects there are many page views and below average shares ("The Night Watch" 50.167 page views, 73 shares). For others there are many views and also above average shares ("The Milkmaid" 25.981 page views, 604 shares). When looking at the most popular art object on Facebook by Adam Willaerts, this object is not so popular on the website, having only 22 page views. What is consistent is that there are many objects which are not so popular on Facebook and also not so popular on the website.

7 The Information Visualization

An information visualization was developed using the collected data, which has helped to get an overview of the dataset. This section will explain how the vi-sualization works and how it can help to find out more about public preferences.

(20)

Figure 6: Facebook - Likes and shares on Facebook per art object

Figure 7: Facebook compared to web traffic - "The Night Watch" (50.167 page views, 73 shares) and "The Milkmaid" (25.981 page views, 604 shares) are ex-cluded here for visibility.

(21)

Figure 8: A snapshot of the information visualization showing statistics for hidden treasures in the Netherlands.

7.1 Filter by category

In figure 8 a snapshot of the information visualization is given. Before any statistics are provided the user has to select a country on the map. All coloured countries on the map have page views within the selected time range. It then provides three bar charts, showing page views for a given category (art type, century and artist). Every bar in every bar chart can be clicked, so that the user can see which art objects have the most page views for that bar. Figure 9 is an example of such a screen. Providing the actual image may help in finding patterns, since it gives some extra visual information about the selected objects. It is possible to view the top art objects without filtering by category. This can be achieved by clicking the "View preferences" button above the map. The same window from figure 9 will be given.

7.2 Filter by revealed/hidden treasures

In the top left corner a user is given the opportunity to filter by revealed trea-sures, hidden treasures and top pieces. For a museum this filter is interesting since it helps to quickly find out which art objects are popular on the web, but are not on display. Also, it allows a museum to switch between all treasures and top pieces. Comparing these two groups may reveal to a museum that there is a difference between what the museum considers a top piece and what the public does.

(22)

Figure 9: A pop-up windows when selecting a group of art objects. This selection shows the top pieces in the category "Beeldhouwwerk" in the Netherlands.

7.3 Filter over time

The visualization allows the user to select a time interval in the year 2015. This interval has increments per month. Changes made in the interval updates the statistics immediately. Using this time range slider can be interesting if a museum wishes to know if certain art objects are popular in specific seasons or during particular national celebrations like Christmas.

7.4 View Social Media activity

Once the user reaches the pop-up windows showing the images of art objects, the user is given the opportunity to click on the "Try social media" button. This changes the pop-up window into a collage of Instagram images and a word-cloud. This word-cloud is based on frequently used words on both Instagram and Facebook for the selected art object. As was stated in the Social Media result section, not that many Instagram posts have been collected in this research. Figure 10 shows what the collage and word-cloud screen looks like of there are enough posts to be displayed. This collage helps to learn more about activity of visitors on Social Media.

8 Discussion

The implications of the results will be discussed in this section. This section will also discuss how the results help to answer our research questions.

(23)

Figure 10: An instagram collage when selecting an art object. This case shows "The Night Watch".

8.1 Web traffic

To be able to acquire the results for the web traffic analysis, there are some prerequisites for museums. Most importantly, there must be a digital museum where users can browse the museum’s collection. Secondly, there has to be some kind of tracking of visitors, so that page views can be analysed. Lastly, there has to be some record of which art objects are on display and which are not.

The web traffic analysis of the Rijksmuseum’s digital collection has given insight into the added value when trying to discover public preferences of this data source (RQ2). What is interesting is that a large percentage of the top viewed art objects (revealed or hidden) is actually a hidden treasure. This could have implications for the setting in the physical museum, since this insight could encourage museums to replace art objects.

Another interesting finding is that most revealed treasures are never visited online. This could mean multiple things, namely: the object is not interesting, the object is hard to find on the website or some other unknown reason. Fur-ther research could be finding out why these revealed treasures are not popular online.

Concerning traffic of top pieces, the results may indicate that there is a more preferable selection of top pieces at the Rijksmuseum, namely the objects with the highest amount of page views. There is a difference between the selected 92 top pieces and the 92 most popular pieces on the web. This is interesting given the fact that top pieces are highlighted on the website, but still they are not the objects with the largest amount of page views.

(24)

8.2 Social media

The added value of social media when discovering public preferences differs for the two researched channels. As for Instagram, a limited amount of matches could be made between Instagram posts and art objects from the Rijksmuseum. This may be the result of the simple matching method that was used. Only for some top pieces this method could find matches, which may indicate that Instagram is useful into giving insight into how the public perceives the most popular top pieces. For a vast majority of the art objects, no match could be found, which means that Instagram adds no value into finding out more about public preferences for those objects. The fact that no matches could be found could have multiple reasons. Since there was a search for title matches, it could be the case that Instagram users are not able to or willing to provide the title in an Instagram post. They would simply take a picture without providing the object title.

Facebook results may have been more promising than the Instagram results, since more data could be gathered. Concerning the added value of this channel, it adds information in the sense that the preferences on Facebook are different from the preferences found in other channels. This helps in identifying more ob-jects as being popular among the public. However, preferences were sometimes conflicting with the preferences found in web traffic. Several explanations can be given for this. Firstly, the Facebook audience could differ from the audience on the website, resulting in different preferences. Secondly, Facebook posts which had a certain art object mentioned in it, could have additional content attached which makes it less appealing for a user to like or share. Lastly, it could be the case that we simply do not have enough data to get a clear overview of public preferences on Facebook. Unfortunately, this research could only focus on posts placed on the Rijksmuseum’s profile. If the Facebook API would have allowed to search by hash tags or simply by content, there would have been much more data to analyse.

9 Future work

In this research data for one year has been used. In future research more data could be collected for both the web traffic analysis and social media analysis. For social media in particular it would help in getting a better view of activity and preferences in online communities. However, this research does not answer the question to why certain treasures are preferred by the public. There could be particular reasons to why certain art objects are popular, which shows when we look at one of the most popular objects on Facebook. A photo of Johan Cruijff is suddenly very popular on Facebook, which seems logical since Johan Cruijff had recently past away. This may be an outlier, but this phenomenon could be something that occurs more often than just once. For museums this given is interesting since these objects may become less interesting in the future, so displaying them in the physical museum may not be something that the museum

(25)

wants. In this case linking art objects to news items may have helped to discover this pattern.

The Instagram analysis in this research was limited in the sense that a simple matching method was applied. In future research more complex ways could im-plemented, like: image recognition (Dollár et al., 2014; Manjunath et al., 1996), word2vec (Goldberg and Levy, 2014) or more sophisticated search algorithms. In this research Instagram did not prove to be valuable, but with different meth-ods it is possible that more value can be extracted from Instagram.

Other future work would be finding out why there are conflicting and dif-ferent preferences on a museum’s website and on Facebook. Finding out the differences between these communities could be an interesting research topic.

10 Conclusion

In this research an attempt has been made to find out how heterogeneous data sources can be used to find out more about public preferences of the public of museums. If a museum possesses a digital collection which is accessible to the public and is active on social media, these sources can be used to find out more about public preferences. By means of a case study at the Rijksmuseum it was found that web traffic analysis provided us with the most information. A distinction was made between revealed treasures, hidden treasures and top pieces to find out how traffic is related between these types. An interesting finding is that about 30% of the 300 art objects with the most views (at least 100 views) is a hidden treasure. This indicates that the added value of web traffic analysis is quite high. Two other data sources have been investigated in this research, namely Instagram and Facebook. The added value of Instagram has proven to be quite low in this research, which may have been the result of a rather simple method that was used. Future work may prove Instagram more valuable. The Facebook analysis has shown that preferences on Facebook are not necessarily the same as preferences in web traffic. There is no correlation between Facebook preferences and web traffic preferences.

(26)

11 Bibliography

Anderson, G. (2004). Reinventing the museum: Historical and contemporary perspectives on the paradigm shift. Rowman Altamira.

Anderson, M. L. (1999). Museums of the future: The impact of technology on museum practices. Daedalus, 128(3):129–162.

Armitage, L. H. and Enser, P. G. (1997). Analysis of user need in image archives. Journal of information science, 23(4):287–299.

Belanger, F., Fan, W., Schaupp, L. C., Krishen, A., Everhart, J., Poteet, D., and Nakamoto, K. (2006). Web site success metrics: addressing the duality of goals. Communications of the ACM, 49(12):114–116.

Besser, H. (1997). The transformation of the museum and the way it’s perceived. The wired museum: Emerging technology and changing paradigms, pages 153– 170.

Black, G. (2005). The engaging museum: Developing museums for visitor in-volvement. Psychology Press.

Dollár, P., Appel, R., Belongie, S., and Perona, P. (2014). Fast feature pyramids for object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 36(8):1532–1545.

Fantoni, S. F., Stein, R., and Bowman, G. (2012). Exploring the relationship between visitor motivation and engagement in online museum audiences. In Museums and the Web.

Fletcher, A. and Lee, M. J. (2012). Current social media uses and evaluations in american museums. Museum Management and Curatorship, 27(5):505–521. Goldberg, Y. and Levy, O. (2014). word2vec explained: deriving mikolov et al.’s

negative-sampling word-embedding method. arXiv preprint arXiv:1402.3722. Hoffman, D. L., Novak, T. P., and Chatterjee, P. (1995). Commercial scenarios for the web: opportunities and challenges. Journal of Computer-Mediated Communication, 1(3):0–0.

Hooper-Greenhill, E. (1999). Museum, media, message. Psychology Press. Klavans, J. L., LaPlante, R., and Golbeck, J. (2014). Subject matter

catego-rization of tags applied to digital images from art museums. Journal of the Association for Information Science and Technology, 65(1):3–12.

Lagrosen, S. (2003). Online service marketing and delivery: the case of swedish museums. Information Technology & People, 16(2):132–156.

Lazzeretti, L., Sartori, A., and Innocenti, N. (2015). Museums and social media: the case of the museum of natural history of florence. International Review on Public and Nonprofit Marketing, 12(3):267–283.

(27)

Liebenson, S. (2009). Courting a wary customer: Three ways to build and maintain loyal relationships when customers are running scared. Deliver, 5(3).

Manjunath, B., Shekhar, C., and Chellappa, R. (1996). A new approach to image feature detection with applications. Pattern Recognition, 29(4):627– 640.

Marty, P. F. (2008). Museum websites and museum visitors: digital museum resources and their use. Museum Management and Curatorship, 23(1):81–99. Pallud, J. and Straub, D. W. (2014). Effective website design for experience-influenced environments: The case of high culture museums. Information & Management, 51(3):359–373.

Peters, J. (2008). Social media roi: Mythical monster or concept evolution. Pulh, M., Marteaux, S., and Mencarelli, R. (2008). Positioning strategies of

cultural institutions: A renewal of the offer in the face of shifting consumer trends. International Journal of Arts Management, pages 4–20.

Ross, M. (2004). Interpreting the new museology. Museum and society, 2(2):84– 103.

Russo, A. and Peacock, D. (2009). Great expectations: Sustaining participation in social media spaces. In Museums and the Web 2009, pages 23–36. Archives & Museum Informatics.

Schick, L. (2010). Can you be friends with an art museum? rethinking the art museum through facebook. In Papers presented at the conference in Tartu, 14-16 April 2010, page 36.

Segel, E. and Heer, J. (2010). Narrative visualization: Telling stories with data. IEEE transactions on visualization and computer graphics, 16(6):1139–1148. Tang, M.-C. (2005). Representational practices in digital museums: A case study of the national digital museum project of taiwan. The International information & library review, 37(1):51–60.

Throsby, D., Bakhshi, H., et al. (2010). Culture of innovation: An economic analysis of innovation in arts and cultural organisations.

Vergo, P. (1989). The reticent object. The new museology, pages 41–59. Wright, D. K. and Hinson, M. D. (2008). How blogs and social media are

changing public relations and the way it is practiced. Public relations journal, 2(2):1–21.

(28)

12 Appendix

A

Top hidden treasures

Appendix A: Shows the top hidden treasures on the Rijksmuseum web-site.

(29)

B

Top artists - All treasures and hidden

trea-sures

Appendix B: Shows the most popular artists ordered by page views for both all treasures and hidden treasures.

(30)

C

Top art object types - All treasures and hidden

treasures

Appendix C: Shows the most popular art object types ordered by page views for both all treasures and hidden treasures.

(31)

D

Top all treasures

Appendix D: Shows the top 30 of all treasures based on page views on the Rijksmuseum website. Last column compares the rank of an object to the rank of the top "top pieces". Green boxes indicate that a an object is not selected as a top piece by the Rijksmuseum. Blue indicates that ranks are the same. Red indicates that a rank based on page views is higher than the rank based on the top piece selection.

Revealing hidden treasures of museums based on public preferences: A case study