Based on the premises of relevance feedback, we investigate awell-known learning procedure called the Perceptron ConvergenceAlgorithm (PCA) (Wong et al., 1998) for information retrieval (IR)systems. Although PCA is one of the simplest methods with the leastoverhead, it has its disadvantages. One of the main disadvantages isthat the rate of convergence of the algorithm can be slow. In thispaper, we propose and evaluate an algorithm that can be used in IRsystems, called the Fast Perceptron, which improves the speed ofconvergence of the PCA. By considering a prototype image retrievalsystem called Web-IDBS(Image Database System) (Guivada et al.,1994)that supports content based adaptive retrieval using semanticattributes (Guivada and Raghavan,1995), we compared the performanceof the fast perceptron algorithm to that of the PCA. The objective ofthis experiment was to compare the impact of user relevance feedbackon improving retrieval effectiveness with these query reformulatingalgorithms. The metrics used for comparison were the number ofiterations required in each algorithm to get the optimal query, thefinal ranking of the target image, and the number of images for whichthe user provides feedback. The results demonstrated that significantimprovement is achieved in the metrics considered.
展开▼
