Personalizing Web Sites with Mixed- Initiative Interaction

T he IT industry has interpreted “personalization” in many ways (D. Riecken,“Personalized Views of Personalization,” Comm.ACM, vol. 43, no. 8,Aug. 2000, pp. 26-28). Personalization refers to the automatic adjustment of information content, structure, and presentation tailored to an individual user. Commercial Web sites increasingly employ personalization to help retain customers and reduce information overload. For instance, Amazon’s ecommerce site is estimated to have at least 23 different types of personalization (J. Riedl, “Personalization and Privacy,”IEEE Internet Computing,Nov.-Dec.2001,pp.29-31). But what does it mean for a Web site to be personalized? Some Web sites, especially e-commerce sites, welcome returning users and remember personal details, such as credit card numbers. Other Web sites track purchase patterns and recommend specific products. Still others provide browsing aids, such as top-10 visited links. A Web site is personalized if a user can interact with the site in an expressive way to achieve his information-seeking goals. Thus, personalizing the user’s interaction is the best way to achieve personalization. This approach is complementary to thinking of personalization in terms of content relevance, delivery speed, and qualitative criteria such as utility and customer satisfaction. To recognize personable interaction with a Web site, consider the human-to-human dialogues between a camera buyer and a dealer in the “Human Dialogues” sidebar. Both conversations involve the specification of camera attributes, but they differ in important ways.The buyer in the first dialogue responds to questions in the order the dealer chooses to pose them.The dealer has the initiative at all times,and we refer to such an interaction as a directed dialogue. In the second dialogue, when the dealer prompts the buyer about camera manufacturer, the buyer instead responds with information about single-lens reflex (SLR), his choice of type, in line 3 of dialogue 2. The buyer in dialogue 2 thus takes the conversational initiative from the dealer. Nevertheless, the conversation does not stall, and the dealer continues with the other aspects of the information-gathering activity. In particular, the dealer registers that the buyer has answered a different question than the one he was asked, and the dealer refocuses the dialogue in line 4 to the issue of manufacturer (this time, narrowing down the available options). Such a conversation—where the dealer and buyer exchange initiative—is called a mixed-initiative interaction (J.F. Allen and colleagues, “Towards Conversational Human-Computer Interaction,” AI Magazine,Winter 2001, pp. 27-37). Our goal is to provide an interaction instrument so the user can take the initiative in Web site interactions.