More Efficient Tagging Systems with Tag Seeding

Tags are a useful mechanism for users to find, organize, and understand items on a web site. A tagging system evolves to reflect a user community’s understanding of an information space. However, the evolution can be problematic: specifically, the ratio of tags to items can decrease over time, to the extent that tags no longer effectively discriminate among items. The primary reason is over application of existing popular tags to most of the items. This makes it difficult for a user to search and navigate using specific tags. We address this problem of disproportionate application of existing popular tags by eliciting applications of those tags that are under applied but highly relevant. In this paper, we introduce two metrics to identify such under applied relevant tags from the system. We then design a controlled study to elicit more applications of these tags, to show how the declining discriminating power of tags is counteracted by interventions in tag applications using these carefully chosen under applied tags, namely seeded tags. We found that users were able to apply these underutilized tags, with the result that about eleven months of lost navigational and search efficiency of the tagging system was attained back. We also monitored subsequent usage of these tags in the system and found that the seeded tags does attract more users in navigation (such as clicks and searches) than before our intervention, suggesting that tag seeding has a persistent effect over the system.