Personalization in Business-to-Customer Interaction

As electronic catalogs are visited by users having different interests and knowledge about products, the role of personalization in e-commerce has been increasingly recognized [6]. Several B2C tools have been extended with the one-to-one recommendation of products (for example, BroadVision—see www.broadvision.com). Moreover, other systems have been developed to adapt the presentation of information to the user’s physical capabilities [5] and device [4].

SeTA (Servizi Telematici Adattativi)¹ is a tool kit for the creation of Web-based stores supporting user-adaptive, multilingual interactions with customers. This system aims to tailor the presentation of products to the customer’s interests and information needs and has been used to develop an electronic catalog describing telecommunication products. The personalized presentation of information is based on three main ideas:

• Product features specify different types of information such as general data (price), aesthetic information (size, color), and technical information (functionalities, requirements). This characterization is exploited to customize the product presentations, according to the user’s interests.
• More or less technically detailed descriptions are generated to suit the user’s knowledge.
• The amount of information presented is tailored to the user’s interests and receptivity, showing only the most relevant data and hiding other details that the user can reach on demand as supplementary information.

The system applies personalization strategies to tailor content, layout, and presentation style to the customer’s interests and knowledge about products, according to information stored in a user model for that customer. Moreover, template-based natural language generation techniques are employed to efficiently produce the presentation content [1]. As such content is generated at the granularity level of the individual product features, significantly different descriptions are produced when describing the same product to different users. For instance, the sample page shown in Figure 1 is tailored to a customer with low receptivity and minimal background in telecommunication products: a few features are described using simple terminology. The sample page shown in Figure 2 presents the same product to an expert, highly receptive user: more detailed and technical descriptions are generated for expert customers.

SeTA uses stereotypical information about customers to initialize the user model. Moreover, it applies unobtrusive user modeling techniques to update the model on the basis of an analysis of the user’s behavior. This is possible because specific information needs are associated with each type of action that can be performed on the catalog pages. For example, clicking on “help” buttons denotes a lack of knowledge about features, whereas inspecting technical details denotes technical interest [3].

A subjective evaluation of SeTA carried out in our laboratories showed that users appreciate the personalization features offered by the system because they contribute to filling the gap between the customer’s view on products and the technical information available about them. In particular, the personalized descriptions help inexperienced customers select goods in an informed way.

The system is based on the multiagent architecture described in [2]. Knowledge representation techniques support the configuration of the tool kit in different domains, and artificial intelligence techniques such as Bayesian networks and rule-based systems enhance its reasoning capabilities. Finally, standard techniques such as XML and XSLT are used for generating the Web-based user interface.

Back to Top

Back to Top

Back to Top

Back to Top

Back to top