Interaction Models for Preference Elicitation


Researchers have expressed the consumer buying behavior as a 6 step process: need, product brokering, merchant brokering, negotiations, purchase, delivery. This process is not essentially sequential. At different stages there are iterative processes as well, primarily to refine the options chosen so far. In general all the present e-commerce applications are based on this model. But often the users are quite ignorant of the taxonomical classifications of the objects they set out to buy, for example, buying a very specialized product such as a digital camera. Hence, it is quite unrealistic to assume the users to have a well formulated need. Hence the system has a glitch in the very beginning. In the absence of a clear formulation of needs, the user depends on the system to present him a deal worthy of his consideration and money. Often it happens that the consumer ends up having deal that sound to them as “Ok” but not “great”. Here consumer satisfaction is at stake which is quite a critical factor. All businesses identify consumer satisfaction as the most important commodity they sell with their products. All economic theories of price are also based on the idea of satisfaction achieved. The basic drawback of the current system is that the user is not involved in the process of product/solution selection. Hence there is always a psychological barrier that makes him doubt the deal even if it’s the best he could have got. Humans need a feeling of being in control, which is primary to their satisfaction.

Human centered decision making principles have been the center of massive research activities in the recent past. While on one hand traditional decision theory claims to predict exact solution for a decision problem, provided a value function is defined. A value function is one which can predict accurately about decision values given precise value of the attributes. Researchers have long found the problem of defining a precise value function tough in real life scenarios. The problem becomes even more complicated when the decision task is a multi attribute function. A simple example of such a task is the selection of a multi attribute product. Our work centers around selection of a multi attribute product on an e-store. Many tools are currently available which help users in making the crucial decision. The major impediment with such tools is the fact that under uncertainty in preferences, they are only as good in helping the users in their decision task, as users themselves are in their preferences. Experiments done with users have suggested the fact that often preferences with users are fluid and uncertain, hence a proper method has to be designed to elicit those unformed and often unspecified preferences. It is these preferences that lead to a definite decision goal. Under the scenario of uncertain and unspecified preferences, the task of defining a precise value function also becomes highly complex and difficult. We believe that such hidden and unstated preferences can be elicited by providing users with example. This process is called example critiquing. The users are shown examples of the products they are looking for, and then they discover their preferences when attributes of the example violated any of their hidden preferences. The idea is that users do not know their preferences till they see them being violated. Thus, we believe that such example critiquing can be crucial in eliciting preferences of users and help them in better decision making.


focus + context, information visualization


2003 – 2004


Pratyush Kumar and Pearl Pu


Swiss National Science Foundation


  • Pearl Pu and Pratyush Kumar. Evaluating Example-based Search Tools. In Proceedings of the ACM Conference on Electronic Commerce (EC’04), pages 208-217, May 17-20, 2004, New York, USA. [pdf]
  • Pearl Pu, Pratyush Kumar and Boi Faltings. User-Involved Tradeoff Analysis in Configuration Tasks. In workshop notes of the Third International Workshop on User-Interaction in Constraint Satisfaction, Ninth International Conference on Principles and Practice of Constraint Programming, pages 85-102, 2003. [pdf]


Demo 1: RankedListUI
Demo 2: TweakingUI
Both demos require enabling of java applets.

More information

You can also read the following report: Pratyush Kumar. Interaction Models for Decision Tradeoff Tasks. Unpublished technical report, EPFL, 2003. [DOC, 7.7MB]