Wednesday, May 23, 2007

Social influence on the convergence of tags

I'm a big fan of finding out how the social conditions influences the tags. Well, that is why it is one important aspect of my PhD studies, so we will find out - sooner than later.

In order to design a good tool for social tagging, I did quite a lot of research on our current understanding on this. I also came up with the table below, as I was mulling over what would be the best balance between giving social cues to people while tagging (e.g. showing the tags from other users) and still keeping the tags individual, personal, and specific enough.

See, I think too much of the social condition will make people lazy to come up with their own tags, thus the tag base will become less and less descriptive and consequently it becomes harder to find items tagged with these terms. Chi et al (2007) find that out too, but they do not contribute it to social influence! Just to the size of the community.

I think social cues are important for the uptake of tagging, in the first place, and in the second, they are important in terms of using common vocabularies, e.g. seeing those broad folksonomies to emerge. But too much can be too much! Thus, we are planning an experiment on tagging interfaces, where half of the users see the previous tags (social influence condition) and the other half does not (independent condition). We will study the tags from different aspects, how they converge, their originality and some other...not sure yet.

High convergenceLow convergence
Social influence conditionPrompt tags from other users (1) ; high influence from the community (2), better uptake as users may be more motivated to add tags; (3) maybe all tags become the same.
Use of existing tags becomes a habit, not many new tags are created as the time goes by (4); tags become less and less descriptive and consequently harder to find items.
Independent condition (no guidance)Even without social influence, when many users tag popular items, usually broad folksonomies startemerging (4).Original and intuitive user generated list evolves; however, low convergence of tags (1), lower uptake of tagging (3)

Table 1: Table presents two axes that affect on tagging habits and

(1) if people see other's tags (e.g. they are proposed, are auto filled when typing, ...) while they are tagging, vocabularies are more likely to converge than if users are working on their own.

(2) Also, users, who view tags by other people before adding their first tag, are more likely to have their tags influenced by other taggers in the community. The community of other users affects a user's personal vocabulary; there is a strong influence on user's first tag, if they have been exposed to others' tags.

(3) There were overwhelmingly more non-taggers in the group that had not seen examples of tags than in the one that had seen them in their tagging interface. As stated above, pre-existing tags affect the future tagging behaviour.

(4) When looking at how the vocabularies evolve while tagging, it was found that about half of the tags used were tags that the user had previously applied; thus, it was concluded that early habit and investment influence tagging behaviour and grows stronger as users apply more tags.
However, the research shows that habit and investment aren't the only factors that contribute to vocabulary evolution.

Well, there was another point, somewhat related to this, that I talked about with a studdy-buddy of mine: what is a good tag? Firstly, it is important to note that tags do have two main functions: one being a PKM element and the other is the sharing.

So, for the first one, any tag is good, if it makes sense to the user.
For the second category, we thought of different metrics; it could be for retrieval purposes or sharing with other people.

For retrieval, for example, a tag that repeats the terms in the title is not very good, especially if the search looks at the title anyway (like in our case we do have LOM already). So one could say that a good tag has some additional information that we do not have in the metadata already. After all, if we talk about social bookmarking of websites or research papers, there is some metadata already available. A tag that is a synonym of the title, however, can be useful for retrieval purposes, if any automated way for understanding synonyms are used (like a thesaurus that knows the relations of different terms).

A good tag could also hint something in the use of that content, for example in digital content, it could be something that would hint how that content could be used.

Chi, E. H. and Mytkowicz, T. Understanding Navigability of Social Tagging Systems. In Proceedings of CHI'07, February, 2007.

Sen, S., Shyong K., L., Cosley, D., et al. (2006). tagging, community, vocabulary, evolution. Proceedings of CSCW 2006. Retrieved from

Vander Wal, T. (2005). Explaining and Showing Broad and Narrow Folksonomies :: Personal InfoCloud. Blog posting. Retrieved November 13, 2006, from

No comments: