Recommendations and Adaptation of content in New Domains (RIND)

Configurators assist users in selecting attributes and features such as customer requirements and product attributes of a complex product. In a joint project between SICS AB and Tacton Systems AB, a commercial company selling and marketing a configurator, we have investigated new and complementary ways of assisting the user in the task of configuring complex products. The result of the project is a prototype in which we have added i) automated, collaborative recommendations for displaying social trails associated with the configuration, and ii) a help interface. The prototype is built on top of the Tacton configurator in the domain of PC configuration.
By tagging the configuration process with social information we aim to give the user a better idea of which attributes or complete configurations that are common or not. The recommendations guide the user by displaying two types of social trails. First, the user can get recommendations on selected product attributes. Attributes that are recommended are those that other people have selected in similar situations. Second, the user can ask: Given my current selections, what is the most popular final product configuration?
For bootstrapping, we propose an approach that makes use of prior knowledge acquired from asking experts, analyzing the recommended items, and using prejudices and good guesses. The acquired knowledge is often uncertain and thus represented with a probabilistic model.

The help system is designed to adapt to the current user selections in the configuration. For experienced users, detailed help can easily be found with keyword search. For users who are new to the product domain, the structure of the help system functions as a guide to the domain.

Using a rule-based knowledge system, the configurator calculates and displays which attributes are compatible with the user's previous selections. Whenever a choice is incompatible, the configurator will inform the user what needs to be changed in order to keep her most recent selection. The configurator can also select attributes that optimize some product variable, e.g. price. In this way, a user may select the most important product attributes and let the configurator select all the other.

The configurator is good at handling product attributes. Adding recommendations and cluster-based help takes us yet another step closer to the underlying customer needs.

A paper on RIND, "Enhancing Web-Based Configuration with Recommendations and Cluster-Based Help", was presented at the Workshop on Recommendation and Personalization in eCommerce at AH2002. See http://ectrl.itc.it/rpec/schedule.htm.

RIND was funded by Vinnova.