Using Relevance Feedback Techniques for Optimising Media Retrieval Applications

Storage capacity in portable media players has increased to 30 GB and beyond. Increasing bandwidth and connectivity has led to a case where there is an arguably unmanageable amount of content stored in computers and media players. Creating the right interface for managing media, is highly important in consumer electronic devices. Previous works [1], [2], [3], [4] have focused on mechanisms to classify and retrieve audio files. But mere classification is not enough in the case of personal media players, as the musical taste of each user is different, and classification cannot be purely based on what others have classified. For adoption in media players and other consumer electronic devices, the musical tastes of each user have to be taken into account while retrieving, and this aspect is missing in current systems. Also, the speed of retrieval is of prime importance in consumer electronic devices. This paper discusses ways of optimising the retrieval process, using relevance feedback from the user. The feedback is used to tune the classification process using perceptual weights. The feedback can be generated directly, or by using indirectly generated data from the behaviour of the user during the search process. Speed of retrieval can be optimised for the user, by successively optimising the search results. Experimental results show the benefits of this approach in terms of search time to get the most relevant results.