The Recommendation in the Product Line Configuration Process

In Product Line Engineering (PLE), product configuration describes the process of developing a product according to user requirements, by reuse from a Product Line model (PLM). The problem is that there are so many products in a PL that it is impossible to specify all of them explicitly. Then, when a user makes a decision (e.g. require or reject a reusable artifact), this can be contradictory with former decisions. Thus, the user will be confused and she/he will eventually abandon the configuration process. Consequently, it's crucial to guide the user by combining recommendation and configuration in the PL configuration process. Recommendation helps users to identify relevant products according to their requirements which are elicited by observing purchase habits, features of products formerly acquired, etc. Several recommendation techniques already exist [1]. Many of these techniques are made for simple products and are not adapted to complex systems such as product lines or configurable software. Our research goal is to help user to make his choice in a dynamic way by combining recommendation and configuration. This aim to inform the user in real time about possible/unattainable features according to her/his choices, and to suggest decision by reasoning with known configurations. Thus, the main objective of our research goal is to answer the following research question: How to combine the recommendation and the configuration in a product line? In addition, another problem arises: at which configuration process level the recommendation should be applied? In order to achieve our research goal, we are led to study and experiment the different recommendation techniques by exploring the limits and interests of each one. This study allows adapting these techniques to the PL configuration process. The idea of our approach is to recommend only partial configurations that are valid with respect to the PL constraints, and satisfy the requirements that the user has already specified. Our approach consists in intertwining recommendation and configuration activities in an iterative way. At each iteration (i) a series of decisions is offered to the user, (ii) the user makes choices, (iii) testing the user partial configuration, (iv) recommendation (v) configuration and constraint propagation, (vi) final decision.