This paper presents a contribution to design an online preference based system. The objective of the system is to assist a customer in the products selection process. Current e-commerce recommendation systems assist customers in this process. Nevertheless, quality of the recommendations produced remains a real challenge. There are products that are by mistake recommended to customers and inversely. This paper focuses on these quality and relevance of recommendations. More than product objective characteristics, the customer's choice is also based on his/her perceptive expectations. Therefore, to be considered as relevant, products recommendations must reach customer's expectations and particularly perceptive ones, sometimes spontaneously, without specific request. Collaborative filtering and neighbourhood formation are the main tools used. The cluster of "perceptive" neighbours containing the active customer share common perceptive preferences and can guide the propositions. The application case is the comic. The aim is to propose to a customer a "good" product. A test procedure enabling the validation of this algorithm is to be set.