Fuzzy knowledge of consumers’ frequent purchase behaviors can be extracted from transaction databases. To effectively supporting decision makers, it is necessary to use fuzzy knowledge to assess weights or degrees of consumers’ attentiveness to product attributes. From the standpoint of habitual domains, frequent purchase behaviors can be viewed as ideas that are contained in the reachable domain of customers. In addition, this reachable domain is changeable with time, due to the dynamic environment. This paper thus proposes a two-phase learning method with adaptive capability. The first phase builds a fuzzy knowledge base by discovering frequent purchase behaviors from transaction databases; the second phase finds weights of product attributes by a single-layer perceptron neural network. Indeed, customers are asked to evaluate alternatives and attributes through questionnaire. Then, each alternative can be transformed into a piece of input training data for the neural network by the fuzzy knowledge base and part-worths of attributes’ levels. After completing the training task, we can find weights from connection weights. Simulation results demonstrate that the proposed methods can use fuzzy knowledge to effectively find customers’ attentive degrees of attributes.