摘要: Tagging is an important way for users to succinctly describe the content they upload Internet. However, most tag-suggestion systems recommend words that are highly correlated with existing tag set, and thus add little information a user's contribution. This paper describes means determine ambiguity of set (user-contributed) tags suggests new disambiguate original tags. We introduce probabilistic framework allows us find two appear in different contexts but both likely co-occur set. If such can be found, current description considered "ambiguous" recommended user further clarification. In contrast previous work, we only query when needed good suggestions available. verify efficacy our approach using geographical, temporal semantic metadata, study. built system statistics from large (100M) database images their