ExpertClerk: A Conversational Case-Based Reasoning Tool forDeveloping Salesclerk Agents in E-Commerce Webshops

  • Authors:

  • Hideo Shimazu

  • Affiliations:

  • Internet Systems Research Laboratories, NEC Corporation, 8916-47 Takayama, Ikoma, Nara, Japan (E-mail: shimazu@da.jp.nec.com)

  • Venue:
  • Artificial Intelligence Review
  • Year:
  • 2002

Quantified Score

Hi-index 0.01

Visualization

Abstract

Conversational Case-based Reasoning (CCBR) has been used successfully toimprove knowledge management in corporate activities as a problemsolver. In our past research, we developed CCBR systems in customersupport domains where CCBR systems played the role of customer supportagents. Based on these experiences, we have applied the same CCBRtechnologies to design the user-interface of e-commerce websites.ExpertClerk was designed as a tool for developing dialogue-basedfront-end systems for product databases. We first analyzed conversationmodels of human salesclerks interacting with customers. The goal of asalesclerk is to effectively match a customer's buying points and aproduct's selling points. To achieve this, the salesclerk alternatesbetween asking questions, proposing sample products, and observing thecustomer's responses. ExpertClerk imitates a human salesclerk. Itconsolidates the human shopper's requests by narrowing down a list ofmany products through a process of asking effective questions usingentropy (navigation-by-asking) and showing contrasting samples with anexplanation of their selling points (navigation-by-proposing). Thisrequest elaboration cycle is repeated until the shopper finds anappropriate product. In this article, we present the systemarchitecture, algorithms as well as empirical evaluations.