An empirical study on the sociality of tag selection on social bookmarking services
This study attempts to characterize users' sociality in tagging by clustering factors that users consider when selecting tags for their bookmarks. Twenty-three frequent users of social bookmarking services were invited, and Q method and factor analysis have been applied in the study. Each cluster of users has varied focuses on personal, content, situational, and social factors when tagging. The study explores the four distinct types of sociality in tagging by the orientation of tag selection and the complexity of affecting factors. The results show different tendencies of sociality in tagging emerge, i.e., tag for me, us, all, and mixed. As there is the shift from functionality towards sociality for the social tagging applications, the initial findings may benefit the future related research.
Social tagging applications like Del.icio.us have been seen as key components of Web 2.0 services. In recent years, there are considerable amounts of studies investigating user behavior and tag usage patterns in social tagging applications (e.g., Golder & Huberman, 2006; Marlow, et al., 2007). Generally, social tagging describes the process by which many users freely assign unstructured keywords to shared content. Either intentionally or unintentionally, users participate the community through the social tagging process. For example, social bookmarking services inject sociality into information seeking by allowing users to search for resources via community browsing (Millen, et al., 2007). Therefore, the nature of the tags and the act of tagging itself becomes a social or even collaborative activity (Zollers, 2007). Research on the social aspect of tagging behavior begins to receive attention (e.g., Ames & Naaman, 2007; Kipp, 2006). It is well assumed that sociality rather than functionality is the key concept in designing social tagging applications.
The ‘social’ in social tagging comes from being able to view other's tags and share resources with others as well as create tags and organize resources with others. While there are benefits of using user-generated tags for indexing and retrieval, the sociality behind these activities is important for enhancing social tagging applications. This study attempts to explore the priorities of various factors that users consider when selecting tags for their bookmarks, and characterize users' sociality types based on the factored analysis of their tag selection behavior. Social bookmarking service, which is the most typical social tagging application, is used as the study target. 23 frequent users of social bookmarking services were invited, and Q method has been applied to analyze their priorities of factors in tagging, including personal, content, situational, and social factors. Factor analysis has been applied to cluster users, and their social characteristics of tagging were then investigated based on the types of sociality proposed by Bouman, et al. (2007). This study aims to understand the sociality behind the tag selection process, and the results may further our understanding on the social aspects of tagging and benefit the design of social tagging services.
The study invites 23 regular users from 4 major social bookmarking services in Taiwan, namely HEMiDEMi, MyShare, Search 2.0, and Yahoo! Kimo bookmarking services. Their services allow users to bookmark web pages and associate user-generated tags with them. All participants have experiences in using bookmarking services for more than 3 years, and their usage is on a daily basis. Q method (Stephen, 1953) has been applied to explore the priorities of various factors that participants consider when tagging their bookmarks. The study begins by preparing statements about how participants select tags for their bookmarks. There are 4 types of factors consisting of 36 statements collected from literature review, i.e. situational, personal, content, and social factors. Each participant was asked to express her/his opinions about the statements by rank ordering them. This creates a Q sort for each person. The Q sorts are then factor analyzed to find patterns in how statements are related. Follow-up interviews include a few questions concerning with the participant's tagging behavior on the bookmarking websites. To further understand the sociality of tag selection behavior, the study categorized the clusters obtained from the above Q-sorts analysis into 4 distinct types of sociality adapted from Bouman, et al. (2007), which will be described in the next section.
The cluster analysis of Q-sorts obtained results in 5 types of users as listed in Table 1. As suggested by McKeown & Thomas (1988), the clustered type can be characterized by presenting the most agreed and disagreed statements, as well as the factors behind these statements. Table 1 lists top 4 of the most and least considered statements and their corresponding factors in tagging. The original statement has been abridged for ease of reading, such as the statement of “I will consider the topic of the web page collected when selecting tags to describe it” is abridged with ‘topic of bookmark’. Type A users are mostly affected by personal factors, and less by situational factors. The participants put emphasis on topics of the tags as well as the tag usage pattern in the community, and less consider the functional purpose of tags like ease of re-finding. Type B users are mostly affected by situational factors, and less by social factors. The participants have their own way of organizing information, and less consider the communicative purpose of tags. Like type B users, type C users are also affected by situational factors greatly, yet they prefer selecting tags used by the community; while type B users tend to favor personal choice of tags and give more attention to the content of bookmarks. Type D and E users are affected by multiple factors and their tag selection are more complex, such as type D rank high on the topic of bookmark and low on the creation date of bookmark, which both belong to the same content factor. It is assumed that type D put emphasis more on certain aspects of information organization instead of accepting all aspects. Type E is the most complex type of users, and the participants show their tendency of controlling information rather than referring to others.
Adapted from the typology proposed by Bouman, et al. (2007), the study categorized the clustered user groups into 4 distinct types of sociality. As listed in Table 2, the sociality is characterized by the orientation of tag selection behavior and the complexity of affecting factors. The orientation is divided to people-based and artifact-based, which with the former users would refer to community's tags by relating directly to each other and with the latter a perceptible bookmark situated between users acts as a connector. The complexity is divided to one-dimensional and multi-dimensional, which with the former users would strongly focus on a particular factor and with the latter users would refer to more complex relationships. The network-centered sociality presents the social relations are not narrational but informational, where type B in our study shows the strong interests in managing personal information via information exchange. The community-centered sociality evolves from interaction between users and is more complex than network-centered sociality. In our study, type A and C incline or refer to tags used in the community, and often consider the community practice. The artifact-centered sociality relates artifacts or objects to the formation and adaptation of social structures and thus to sociality (Gal, et al., 2004). In our study, type D put emphasis only on the content of bookmarks instead of paying attention to community practice or considering other factors, i.e., the sociality is achieved by the mediating role of objects. The system-centered sociality indicates the complexity of factors involved, which type E in our study show the tendency of treating social bookmarking service as a system triggering sociality instead of a community itself. Based on the above analysis, it is interesting to observe there exists different levels of sociality, the Network-entered users tend to tag for self, the Community-centered users for ourselves, the Object-centered users for all, and the System-centered users for mixed audience depending on various situation.
This study provides some preliminary results on users' sociality in social bookmarking services by clustering factors that users consider when selecting tags for their bookmarks. Investigations and discussions on the four types of taggers with different levels of sociality are also provided. There are taggers tend to tag for me, for us, for all, and for mixed audience. As there is the shift from functionality towards sociality for the social tagging applications, the initial findings may benefit the future related research. While the study did not have enough participants to indicate significant trends, we intend to conduct studies with consideration of a wider range of user groups and longitudinal effects. There are additional research questions worth exploring in more detail in future research. An issue is to have a clear understanding of the relationship among tagging motivation, behavior, and effect, such as analyze gaps or changes along these magnitude. Another issue is to apply the sociality into practical design of social software services, such as develop personalized or community tag suggestion for different types of users based on their sociality tendency and preferred factors in tag selection.
This work was supported in part by the National Science Council, Taiwan, under the grant NSC96-2413-H-003-025.