Chapter XI: Agent-Mediated Knowledge Acquisition for User Profiling

A. Andreevskaia, Concordia University, Canada
R. Abi-Aad, Concordia University, Canada
T. Radhakrishnan, Concordia University, Canada

This chapter presents a tool for knowledge acquisition for user profiling in electronic commerce. The knowledge acquisition in e-commerce is a challenging task that requires specific tools in order to facilitate the knowledge transfer from the user to the system. The proposed tool is based on a hierarchical user model and is agent-based. The architecture of the tool incorporates four software agents : processing agent maintaining the user profile, validating agent interacting with the user when information validation is needed, monitoring agent monitoring the effects of the changes made to the user profile, and a filtering agent ensuring the safe information exchange with other software.

INTRODUCTION

In the past few years , Internet shopping has been growing rapidly . Most companies now try to offer a web service for online purchase and delivery in addition to their traditional sales and services. For consumers, this means a broader range of online stores from which to buy products. At the same time, this also means that users face more complexity in using these online services. This complexity, which arises due to factors such as information overloading or lack of relevant information, reduces the usability of e-commerce sites.

This fact is supported in a study presented by Schaffer and Sorflaten (1999) that revealed serious usability problems with e-commerce sites. In this study, respondents gave the following top three reasons for abandoning a web site during personal shopping: inability to find the sought item (56%); the site is disorganized or confusing (54%); and low speed in downloading the pages (53%).

Usability is a prerequisite for the success of e-commerce. If people cannot easily find a product, then they cannot buy it. It does not matter how cheap the products are (Nielsen Norman Group , 2001a). Besides that, customer loyalty depends on positive branding, which is associating a logo or a product with a positive emotional experience. When someone has a negative experience with a web site, being unable to find a product or navigate the site, they associate that negative experience with the brand. Firsthand experience is much more powerful in determining whether a customer will remain loyal to a brand, and no amount of marketing can overcome a negative experience such as being unable to use or find information on a web site (Rohn, 1998).

Since its very beginning, the Internet has been growing in popularity and complexity; the largeness makes it difficult for the user to find the information he needs. Often, it is more difficult for users to shop on the Internet than by conventional means. On the Internet, the user finds himself either "flooded" with irrelevant information mixed with some relevant information, or lacking relevant information altogether. In the context of B-to-C e-commerce type, we note the following:

• User interfaces play an important role in achieving user acceptance.

• Queries usually return more matches than the user can consult or fewer matches than expected.

• The user is "flooded" by unwanted and sometimes unsolicited information (e.g., advertisement banners that pop-up or appear as part of the main window of the browser).

• The information is sometimes very badly organized, which makes it difficult to read and scan through.

• Some of the cultural and ethical values of shopping in stores are missing when shopping through the Internet (trust, honesty, negotiation, policy, etc.)

• Some of the sites target a global clientele without adapting the site to the local needs ” e.g., supporting multiple currencies (Nielsen Norman Group, 2001b), offering the information in multiple languages, and using universal metaphors or ones that are not specific to a given region or group of people (Hershlag, 1998). Examples of such metaphors can be found in Haque et al. (2001).

One possible way to help solve these problems is to personalize interactions and content between the user and the e-commerce system based on an appropriate user model. User modelling implies incorporating certain knowledge about the user. This knowledge describes what the user "likes" or what the user " knows " (Chin, 1986). It can help us to decide what kind of information he/she is interested in as well as how to present this information. For e-commerce, user modelling can be useful in four different ways (Lu, 1999):

• Providing personalized services to a particular user. For example, filtering out the information that does not correspond to the user's center of interest.

• Disambiguating the user's search input based on his user model. For example, filling in missing fields by anticipation in a query form.

• Providing proactive feedback to assist the user. For example, a hint message that pops up when the user is taking too long to perform a task.

• Presenting the information in a way suitable to the user's needs. For example, presenting the information in an appropriate language.

Many e-commerce sites (e.g., amazon.com and garden.com) already incorporate user-modelling capabilities for the purpose of personalizing the interactions. Major commercial software packages for e-commerce sites and portals developers often provide personalization capabilities as a standard feature [e.g., IBM Net Commerce (http://www-3.ibm.com/software/webservers/commerce/wcs_pro/), ATG Dynamo (http://www.atg.com), BroadVision (http://www.broadvision.com), and others]. The previous user models have been used to predict the user's preference in narrow and specific domains. The results of such work have been limited to suggesting novels or movies to the user, personalizing the navigation of catalogues, or adapting the information presentation (Ardissono & Goy, 2000). These models have been applied in bookstores, music stores and video stores. The main purpose of this personalization was to keep customers in the store longer or to attract more visitors to the site.

As shown by IBM High-Volume Web Site Team (2000) and Colkin (2001), we are currently witnessing a new shift in personalization toward catering to the needs of repeat customers, as well. Encouraging shoppers to return to an e-commerce web site is beneficial and challenging. One may use special promotions and discounts for this purpose or show tailored contents (i.e., information about specific products deemed to be interesting for this particular customer, different level of details for different users, etc.).

Our ultimate goal is to be able to deal with multiple domains while fulfilling the four tasks mentioned ^[1] . We explore the concepts of dynamic personalization and agent support. We describe a user-model named PIE ^[2] that is ontologybased and parameter driven and that can be helpful in broad domains, such as a shopping mall or a department store, as well as in narrow domains, such as bookstores. The second part of the chapter discusses methods for acquiring knowledge about the user's preferences.

^[1] Preliminary work and first version of the system was described in Abi-Aad et al. (2001).

^[2] The term Preference Indication by Example was introduced in Abi-Aad (2001).