Can Social Tagging Be a Tool to Reduce the Semantic Gap between Curators and Audiences? Making a Semantic Structure of Tags by Implementing the Facetted Tagging System for Online Art Museums

Gunho Chae and Jungwha Kim, KAIST, Republic of Korea

Abstract

Social tagging for art museums has been receiving widespread attention as a tool for reducing the semantic gap between curators and visitors by the publicís active participation and knowledge. However, as the number of tags is increasing, social tagging systems have shown limitations in providing meaningful information and supporting semantic relationships between tags and museum collections. Thus, to improve existing social tagging systems, this study proposes a facetted tagging system which gives a guideline or schema for the users when tagging the individual artworks. In order to demonstrate the effectiveness of a facetted tagging system in art museums we first presented six facets  Background, Identification, Theme, Association, Emotion and Figureî – based on explanation and interpretation of art works so that the tags can be organized in semantic structure. Then we implemented the facetted tagging system with Gyeonggi Museum of Modern Art (GMoMA) and discussed the feasibility of facetted tagging systems and their results with GMoMA curators through workshops.

Keywords: Art Museum, Social tagging, Facetted Tagging, GMoMA

1. Introduction

Since the emergence of Web 2.0, online art museums have been evolving into participatory museums, in an attempt to increase the public's participation through the utilization of social media (Durbin, 2003). Among many types of social media, social tagging has been receiving widespread attention as a tool for reducing the semantic gap between curators and audiences, through the group knowledge obtained from the active participation of the public (Trant and Bearman 2006; Smith, 2006; Chun, 2006; Chan 2007).

In this circumstance, Gyeonggi Museum of Modern Art (GMoMA) embarked on an ongoing project with us to explore the potential of social tagging and applying it in museum management strategy. In late 2009, we built our own tag database based on the collections from GMoMA, and carried out experiments. These experiments were done by building a testbed on a website that was created to collect tags on 128 pieces of artwork.

After collecting these tags, we evaluated the feasibility of social tagging systems through workshops with curators from GMoMA. Through interviews and discussions with curators at the workshop, we found the potential of social tagging systems in museums and identified improvements that could be made in order to apply it to actual museums.

As a result, we decided to apply a social tagging system to GMoMA. In December 2010, the homepage of GMoMA (http://www.GMoMA.org) went under construction to incorporate a social tagging system which has currently been applied. However, a new issue arose after applying a social tagging system to GMoMA. Though a social tagging system seemed to bridge the semantic gap between curators and audiences, vast amounts of information from social tagging are still inadequate to be applied the museum management (Chae and Kim, 2011).

Thus, to improve the existing social tagging system and enhance semantic appreciation in online art museums, we developed new tagging system called a facetted tagging system which provides a guideline or schema to users when tagging the individual artworks.

2. Challenges in Applying Social Tagging System for Art Museum

Through social tagging, the curator has a chance to view and understand artwork from the perspective of the audiences. Thus, curators can extend their understanding of artwork beyond their expertise through tags. Despite the positive aspects of social tagging in museums, the flaws of such systems are rising with the increase in the number of tags.

We were able to receive feedback on these challenges and the curators' needs through workshops with GMoMA curators and staff. According to them, in order for tags to be practical information, the initial intention of the viewer, which becomes hidden within the diversity of tags, has to be clarified. In other words, it is difficult to understand "why" a tag was associated with an artwork just by looking at the tag. Also, since tag clouds arrange tags in alphabetical order with no information on their relationships, it is difficult to support semantic relationships among them. Lastly, they stated the necessity of taking into consideration the linguistic issues of tags; such as polysemy, homonymy, plurals, synonymy, basic level variation, etc.

These issues show that the current social tagging system used in GMoMA cannot be utilized actively in museum management yet. For that reason, we started to research an alternative method of using existing social tagging systems to solve these issues.

3. Related Works

These GMoMA issues are in the same context as existing museum social tagging research, but also offer new implications. It is to extend the usage of tags from a tool to minimize the semantic gap between the public and the museum to a tool for museum management.

These issues are similar to those that are highlighted in existing media research on social tagging. Recent research on social tagging point out the limits of social tagging systems and at the same time provide a new method of tagging.

• Xu (2006) stated that existing social tagging systems were free form, thus too chaotic, and that this lack of order and depth could result in a disaster. Xu proposed collaborative filtering as a solution.
• Quintarelli (2007) suggested a lack of precision, a very low findability quotient, and a limited scalability for the intrinsic variability of language as problems of existing social tagging systems. Quintarelli introduced Facetag, an algorithmically generated hierarchies of tags, as a solution.
• Wu (2007) suggested that the low quality and lack of structure in tags were problems of social tagging systems, stating that the ambiguity and noise arise from the linguistic nature of tags. Wu proposed wiki-fashioned facets and categories for tags as a solution.
• Smith (2006) also pointed out issues in museum tagging in 2006, stating that social tags cannot be meaningful bridges between visual elements and deeper meanings in artworks.

This research on the limits of existing social tagging systems and solutions to overcome them, combined with research on the limits of museum tagging systems, provide insight on where the GMoMA tagging system should be headed.

4. A Faceted Approach for Organizing Social Tags for Artwork

This paper proposes a facetted tagging system in order to overcome the limits of existing social tagging systems. Facets are orthogonal descriptors (i.e. categories) within a metadata system. Each facet has a name, and it addresses a different conceptual dimension or feature type relevant to the collection. Also, according to Quintarelli (2007), facets can be flat or hierarchical, and they can be assigned single or multiple values.

Research that finds that utilizing facetted tags creates a more structured social-tag database are emerging. Bar-Ilan (2008) reveals that through facets, structured tagging may be able to produce stronger user guidance, hence possibly resulting in higher quality descriptions. In addition, Hearst (2006), who took into account the user's perspective, stated that information seekers in large domains of objects prefer meaningful groupings of related items, in order to quickly understand relationships and so decide how to proceed. As shown here, facetted tagging is able to process the chaotic wisdom of the public into useful information through proper control.

However, as pointed out from much research, facetted tagging can process useful information, but also has the potential to invade the user's freedom and be a hindrance in the formation of the public's wisdom. In addition, facets that are not detailed may only cause confusion.

This research proposes a method to reduce the semantic gap between the public and the museum by using facetted tagging to process useful information on social tags. In order to do this, facets are created by combining theoretical studies and existing criteria used in classifying tags from other research. Also, we did an evaluation in order to increase the accuracy and usefulness of the facets. By applying the facets to the social tags that were collected on the artworks of GMoMA, we researched the validity of using such facets. We also analyzed the artwork descriptions from four museums through facets to see if it would be useful for specialists in artwork (i.e., curators).

5. Constructing the Facet Structure and Evaluation

Literature Reviews and Manual Indexing for Constructing Facets

Before constructing facets, a review on related works was done. Such research on facet and classification of tags includes facetted tagging research, classification of tags, elements of impressions of artwork, and real-world facets from an image-based facet tagging website. After extracting the facets and classifications from each piece of research and website, we went through the process of manual indexing to apply them on GMoMA's social tag database, and finally extracted facets (or classifications) that were applicable in viewing artwork.

Table 1: Research on facet and classification of tags

After the literature reviews and manual indexing process, we came up with 29 unique facets (or classifications) that were valid to be used in museum tagging. With these 29, we extrapolated six facets based on semantic similarity. Thus, each facet embraces its sub-facets in its definition.

Table 2: Extrapolating six facets

As shown in Table 2, a total of six facets was created after researching related works, manual indexing, and combining these concepts. These six facets are "Background, Identification, Theme, Association, Emotion and Figure".

Verifying Feasibility of Facets through Closed-Card Sorting Test

In order to test whether the six facets were a feasible categorization system for users tagging museum artwork, we began an evaluation to see if the six facets were useful in categorizing social tags from the user's point of view.

We chose the closed-card sorting test method for the evaluation. The card sorting test is commonly used to test the reliability of a categorization system and has the advantage of receiving user feedback in a simple manner. Ten artworks were randomly selected from GMoMA to be used in the experiment. Twenty experiment participants were each given five artworks and asked to classify the associated tags into either one of the six facets, or "other". The participants were not given any time limit, but took an average of five minutes for each artwork.

According to the results, 91% of the tags fell into one of the six facets, with "Identification" having the highest tag density, followed by "Background" and "Association", as shown in the figure 1 below.

Fig 1: Result of the closed-card sorting test

With each artwork, we also analyzed the consistency of categorization among the participants. Each artwork had five participants categorizing its associated tags. We checked how consistently the same tags were categorized among different participants: 61% of the results indicated that three or more participants categorized the same tag to the same facet, and 75% of them showed two or more participants categorized the same tag to the same facet.

These results show that the proposed facets are feasible in structuring social tags on museums and also indicated that the facets have universality.

Verifying Feasibility of Facets through Analysis of Artwork Descriptions

In addition to the aforementioned card sorting test, we carried out another evaluation to see whether the six facets were useful for museum curators in reflecting the public's viewpoint. This was done by analyzing the actual artwork descriptions used by curators with the proposed six facets. The experiment was done using artwork from four museums – SFMOMA, IMA, MOMA, and Guggenheim – as case studies.

Ten artworks were randomly selected from the highlight section of each museum's website, and the corresponding descriptions were classified into the six facets. For example, Table 3 represents an analysis based on the description for Emergency Room by Robert Colescott (Figure 4).

Fig 2: Emergency Room by Robert Colescott from the MOMA website

Table 3: Example of analyzing a description of "Emergency Room" based on six facets

Using this method, we analyzed the artwork of four museums. The result of the analysis is shown in Table 4, with an average of 4.9 facets being used in the artworks' descriptions. From this result, we concluded that these six facets are appropriate components for describing and appreciating artworks.

Table 4: Result of analyzing 4 museums' artwork descriptions based on six facets

With the positive outcomes from the two feasibility tests, we firmly believe that these six facets can cover both the curators' and audience's perspectives, thus providing a plausible cause to pursue the implementation of a facetted tagging system.

6. Is it Possible to Reduce the Semantic Gap between Curators and Audiences through a Facetted Tagging System?

Implementation of a Facetted-Tagging System for Art Museums

In this research, we implement the six facets that were shown to be valid in both the curators' and audience's perspectives. This experiment is a testbed to apply facetted tagging systems to actual museums. In addition, the two major objectives are understanding the user experience, and finding the benefits of using facetted tag databases compared to existing social tag databases.

Twenty-four artworks were randomly selected from GMoMA, and an online testbed was created for the experiment. It lasted for one week, during which users could participate through a website (http://agarpe.cafe24.com/facet/) and begin the experiment after filling out a simple form for particulars. As shown in Figure 3, for each artwork, six facets and "etc" were provided, to be used by the users to classify the associated tags.

Fig 3: A testbed for implementing facetted tagging system for art museum

During one week, 100 users participated in the experiment, and approximately 9400 tags were collected, with an average of 391 tags per artwork. Among all the tags, 165 tags did not belong to any of the six facets and were classified as "etc", accounting for less than 2% of total tags. Based on these results, we examined the potential of a facetted tagging system through analyzing the tag-collection results from the existing social tagging methods and facetted-tagging systems.

Potential of Facetted-Tagging Systems for Art Museums

Through workshops with six of GMoMA curators, we wanted to verify the usefulness of the actual implementation of facetted tagging system on museum artwork, based on the implementation results. In order to do this, we compared and analyzed the existing social tags with the tags that resulted from the implementation. Tables 5 & 6 show social tags and facet tags from Figure 4 which is one of the artworks used in the workshop with curators. Table 5 shows tag results from existing social tagging methods, and Table 6 shows the results of facetted tagging on the same artwork (Figure 4).

Fig 4: Idols on Narrative Stage, Hong Youngin, 2007

Table 5: Social tags for Figure 4

Table 6: Facet tags for Figure 4

The workshop with the curators from GMoMA progressed with comparisons of social tags and facetted tags, as shown above. The focus of the workshop was to see whether facetted tagging, compared to existing social tagging methods, reduced the semantic gap between curators and audiences.

The curators stated that compared to existing social tags, facetted tags "were extremely interesting in that it was easier to see the intentions of the users". They also stated that, because tags were grouped, "it would be possible to use such information when users search or categorize artwork." In addition, they made a comment that "it would be possible to understand the semantic relation between tags in tag clouds created for facetted tagging."

However, we noticed some limits that facetted tagging systems had when applied to the actual museum field. According to curators of GMoMA, "there is a chance that the requirement for users to classify tags into one of six facets might make them feel like a taking a test and become a drawback from freely appreciating art" and that "the distinction between 'association' and 'emotion' was unclear." Thus there needs to be an improved method despite the fact that facetted-tagging has sufficient value to be used in museums .This improvement could be achieved by having fewer facets, or automatically categorizing tags into the appropriate facets.

7. Conclusion and Future Research

This study, starting from the limitations of social tagging systems used in museums, proposed and evaluated a facetted-tagging system as a new attempt to reduce the semantic gap between curators and audience. Using an empirical study we verified that our proposed faceted structure can be effectively used in categorizing artwork tags and is capable of covering the perspective of curators. Furthermore, this study extends further than existing research on museum social tagging and examines the applications of facetted tagging system in museum management. These can be the main contribution points of this paper.

Although facetted tagging, which gives a semantic structure to social tags, has great potential, it also left another research issue. As the curators of GMoMA stated, the existing facetted tagging system can be inconvenient for users, and the facets themselves can trigger ambiguities.

However, we still believe that a facetted structure for artwork tagging can be an effective solution to bridging the semantic gap between curators and audiences. Our ongoing research will be on eliminating the ambiguities in facets and increasing the degree of automation of the facet construction process for artwork tagging.

8. References

Bar-Ilan, J., et al. (2008). "Structured versus unstructured tagging: a case study". Online Information Review, Vol. 32, Issuse 5, pp.635 – 647.

Chan, S. (2007). "Tagging and Searching-Serendipity and museum collection databases". In D. Bearman and J. Trant (eds.). Museums and the Web 2007: Proceedings. Toronto: Archives & Museum Informatics, 2007. http://www.archimuse.com/mw2007/papers/chan/chan.html

Chun, S., et al. (2006). "Steve. museum: an ongoing experiment in social tagging, folksonomy, and museums". Museums and the Web 2006: Proceedings. Toronto: Archives & Museum Informatics, 2006. http://www.archimuse.com/mw2006/papers/wyman/wyman.html

Durbin, G. (2003). "Using the Web for Participation and Interactivity". Museums and the Web 2003: Proceedings. Toronto: Archives & Museum Informatics, 2003. http://www.archimuse.com/mw2003/papers/durbin/durbin.html

Golder, S. and B, Huberman. (2005). The structure of collaborative tagging systems. Arxiv preprint cs/0508082.

Chae, G. and J. Kim (2011). "Rethinking Museum Management by Exploring the Potential of Social Tagging Systems in Online Art Museums". The International Journal of the Inclusive Museum, Vol. 3, Number 3, pp.131-140.

Hearst, M. (2006). "Clustering versus faceted categories for information exploration". Communications of the ACM, Vol. 49, Issue 4, pp.59 – 61.

Housen, A. (1983). The eye of the beholder: Measuring aesthetic development. Ed.D. Thesis, Harvard University Graduate School of Education, UMI number 8320170.

Quintarelli, E., et al. (2007). "Information architecture: Facetag: Integrating bottom-up and top-down classification in a social tagging system". Bulletin of the American Society for Information Science and Technology, Vol. 33, Issue 5, pp.10-15.

Schmitz, P. (2006). "Inducing ontology from flickr tags". In Proceedings of the Collaborative Web Tagging Workshop in conjunction with the 15th International Conference on the World Wide Web.

Smith, G. (2007). Tagging: people-powered metadata for the social web. New Riders.

Smith, M. (2006). "Viewer tagging in art museums: Comparisons to concepts and vocabularies of art museum visitors". In Advances in classification research, vol. 17: Proceedings of the 17th ASIS&T SIG/CR Classification Research Workshop.

Trant, J. and B. Wyman. (2006). "Investigating social tagging and folksonomy in art museums with steve. Museum". The Collaborative Web Tagging Workshop (WWW'06). Edinburgh, Scotland, May 22, 2006. PDF preprint at ttp://www.archimuse.com/research/www2006-tagging-steve.pdf

Wu, H., et al. (2007). "Collaborative classification of growing collections with evolving facets". Proceedings of the 18th conference on Hypertext and hypermedia HT 07.

Xu, Z., et al. (2006). "Towards the semantic web: Collaborative tag suggestions". Collaborative Web Tagging Workshop (WWW'06).

Cite as:

Chae, G., and J. Kim, Can Social Tagging Be a Tool to Reduce the Semantic Gap between Curators and Audiences? Making a Semantic Structure of Tags byIMplementing the Facetted Tagging System for Online Art Museums . In J. Trant and D. Bearman (eds). Museums and the Web 2011: Proceedings. Toronto: Archives & Museum Informatics. Published March 31, 2011. Consulted

http://conference.archimuse.com/mw2011/papers/can_social_tagging_be_a_tool

Copyright © 2011 – Archives & Museum Informatics.