Query Concept Learning

With multimedia databases it is difficult to specify queries directly and explicitly. Relevance feedback interactively learns a user’s desired output or query concept by asking the user whether certain proposed multimedia objects (e.g., images, videos, and songs) are relevant or not. For a learning algorithm to be effective, it must learn a user’s query concept accurately and quickly, while also asking the user to label only a small number of data instances. In addition, the concept-learning algorithm should consider the complexity of a concept in determining its learning strategies. This chapter\(^\dagger\) presents the use of support vector machines active learning in a concept-dependent way (\(\hbox{SVM}^{\rm CD}_{\rm Active}\) for conducting relevance feedback. A concept’s complexity is characterized using three measures: hit-rate, isolation and diversity. To reduce concept complexity so as to improve concept learnability, a multimodal learning approach is designed to use the semantic labels of data instances to intelligently adjust the sampling strategy and the sampling pool of \(\hbox{SVM}^{\rm CD}_{\rm Active}.\) Empirical study on several datasets shows that active learning outperforms traditional passive learning, and concept-dependent learning is superior to concept-independent relevance-feedback schemes.