The relevance feedback approach is a powerful technique in content-based image retrieval (CBIR) tasks. In past years, many intra-query learning techniques have been proposed to solve the relevance feedback problem. Among these techniques, Support Vector Machines (SVM) have shown promising results in the area. More specifically, in relevance feedback applications the SVMs are typically been used as binary classifiers with the balanced input data assumption. In other words, they do not consider the imbalanced dataset problem in relevance feedback, i.e., the non-relevant examples outnumbered the relevant examples. In this paper, we propose to apply our Biased Support Vector Machine (BSVM) to address this problem. Moreover, we apply our Self-Organizing Map-based inter-query technique to reorganize the feature vector space, in order to incorporate the information provided by past queries and improve the retrieval performance for future queries. The proposed combined scheme is evaluated against real world data with promising results demonstrating the effectiveness of our proposed approach.