A categorical comparison between user-supplied tags and web search queries for images
Although social tagging has received attention in the LIS field as a promising information organization mechanism, there is little research comparing user-supplied tags and search queries. For using user-supplied tags as an alternative or supplementary tool for existing image representation schemes, there should be evidence which demonstrate to what extent there are similarities and/or differences between user-supplied tags and search terms. Based on these considerations, this study investigates categorical differences between user supplied tags and Web search queries. Using Shatford's categories, the study identifies categorical characteristics of tags and queries. Using the 1st level of categories, a rough comparison was made. User-supplied tags contain Generic terms for the majority, but Web search queries include Specific terms, rather than Generic terms. For a more detailed comparison between tags and queries, the 2nd level of categories from Shatford's was adopted. In most of subcategories, the categorical comparisons between tags and queries demonstrate considerable differences. Although this result is based on a small size of data sets, the result of this study can impact the way of social tagging systems and indexing/ranking systems of Web search engines for images.
Introduction and Background
Recently, social tagging has received attention in the LIS field as a promising information organization mechanism. Based on the idea that users not only organize information for their own use but also share their organized collections with others, researchers in the LIS field expect that user-supplied tags can serve as a user-oriented indexing approach. A social tagging system includes several promising advantages. First, information loss which inevitably occurs during the information representation process can be overcome or at least decreased through social tags, because for most of the social tagging systems, the information loss is from people having different viewpoints and is not from a single indexer's discern (Shirky, n.d.). Second, users engaged with social tagging systems describe content with their own vocabulary without referring to any formally structured set of vocabulary. Therefore, tagging systems can reveal the conceptual structure and current terminologies of the user community (Furnas, Fake, Ahn etc., 2006). Finally, by offering opportunities to find related topics and users, tagging systems can increase the possibility of discovering unexpected resources through browsing or serendipity (Kroski, 2005). The potential of social tagging seems more beneficial for the image indexing and retrieval process. First, information loss has been identified as one of main obstacles in representing image documents. In other words, since an image document conveys multi-levels of meanings including subjective impression, it has been discussed that a single indexer cannot provide all possible index terms with an image document. Second, existing controlled vocabulary systems for image documents have been criticized because concepts included in those thesauri are too specialized terms for general image collections. Third, browsing has been addressed as a significant activity during the image retrieval process, because verbal queries have limitations in expressing information needs for visual materials.
For employing tags as an organization tool, researchers have attempted to understand features of social tags (Golder & Huberman, 2006; Stvilia & Jörgensen, 2007). They performed several tag analysis studies which quantitatively or qualitatively analyze and categorize user-supplied tags. Although these studies demonstrated that social tags have features which can be utilized as user-centered index terms and organization tools, there is little research comparing user-supplied tags and search queries. For using user-supplied tags as an alternative or supplementary tool for existing image representation schemes, there should be evidence which demonstrate to what extent there are similarities and/or differences between user-supplied tags and search terms. This study aims to compare user-supplied tags for image collection with image search queries. More specifically, this study compares tags and queries for images by categorical analysis and demonstrates that to what extent there are differences in category distributions among tags and queries. The result of this study is expected to provide a guideline on how to integrate or utilize user-supplied tags as an information organization mechanism for an image collection.
This study uses two data sets: a set of user-supplied tags and a set of search queries. As user-supplied tags, this study collected a data set from Flickr. Flickr users can upload and tag their photos and then their photos can be viewed and searched by the public. Using the API provided by Flickr website, 33,742 tags assigned to 8,998 photos were collected. Then, 535 unique tags which were provided by more than two users were identified as a data set. As search queries, the Web search log of Excite 2001 was used. The Web search log of Excite 2001 contains 262,025 sessions and 1,025,910 queries (Spink & et al., 2002). For this preliminary study, 100 randomly selected queries for image searches were analyzed. For examining category distributions among tags and queries, this study adopted a categorization scheme proposed by Shatford (1986). Shatford's categorization scheme has been used in examining which categories of meanings are included in an image and which categories are dominantly used during the image search process (Choi & Rasmussen, 2003). This study investigates whether there are differences in category distributions between tagging and searching processes.
The percentile of user provided tags and Web search queries for images were calculated by Shatford's categories containing Abstract, Generic, Specific, and Others with 2nd level categories. As shown in Table 1, there are considerable differences in Generic and Specific categories. While the majority of user supplied tags is in the Generic category (63.18%), 59% of search queries is in the Specific category. This pattern can be explained from the second level of Shatford's categories. In the Specific category, terms in the Specific object subcategory are frequently used by search engine users who find images of celebrities or famous people, such as ‘Shakespeare photos’ ‘photograph of Adolph Hitler’ ‘Pitt Aniston wedding photo’ and so on. Terms categorized in the Specific event/activity, such as ‘Passover images’ ‘Easter images’ ‘Nobel prize photo’ and so on, are also often employed by search engine users. Among the subcategories in the Specific category, the Specific location subcategory is more often found in tags than in search queries. It is inferred that many of Flickrs users take their pictures during travel and they assigned tags for their travel locations. In the case of the subcategories of the Generic category, it is found that Flickr users tag their images with terms which represent generic object, event, activities, and so on. Search engine users also often used terms categorized in the Generic object, but this is less frequent than the terms used in the Specific categories. In the Abstract category, web search queries and user supplied tags demonstrated similar patterns in general, but it is demonstrated that the Flilckr users more often used terms in the Emotion/Abstraction subcategory than the search engine users. In the Others category, user supplied tags included Flickr oriented tags (e.g., flickr, Fujifilm, geotagged, etc.) and color tags while Web search queries contained queries of image research (e.g. image segmentation, image and data, etc.) and vague queries (e.g. photos, free images, etc.)
Table 1. The percentile of user provided tags and Web search queries by Shatford's categories
|Abstract (A)|| ||8.04%||6.00%|
| ||Abstract object (A1)||0.19%||0.00%|
| ||Emotion/Abstraction (A2)||7.10%||3.00%|
| ||Abstract location (A3)||0.75%||2.00%|
| ||Abstract time (A4)||0.00%||1.00%|
|Generic (G)|| ||63.18%||22.00%|
| ||Generic object (G1)||46.35%||20.00%|
| ||Generic event/activity (G2)||5.61%||2.00%|
| ||Generic location (G3)||9.72%||0.00%|
| ||Generic time (G4)||1.50%||0.00%|
|Specific (S)|| ||19.63%||59.00%|
| ||Specific object (S1)||3.17%||47.00%|
| ||Specific event/activity (S2)||0.56%||6.00%|
| ||Specific location (S3)||14.21%||6.00%|
| ||Specific time (S4)||1.68%||0.00%|
|Others||Color, Flickr related, Part of speech||9.15%||13.00%|
Discussion and Future Studies
The purpose of this study is to investigate categorical differences between user-supplied tags and Web search queries. Although this preliminary study compares user supplied tags and Web search queries based on small data sets, it demonstrates that user-supplied tags and Web query terms have some different characteristics. Therefore, for utilizing user-supplied tags in designing user-centered image access mechanisms, it should be based on further understandings of behavioral and cognitive features of image tagging system users as well as features of tags assigned by users.
The authors gratefully acknowledge the assistance of Stacy Davis of University of South Florida in data analysis for this research.