Intelligent Product Brokering for E-Commerce: An Incremental Approach to Unaccounted Attribute Detection

Abstract

This research concentrates on designing generic product-brokering agent to understand user preference towards a product category and recommends a list of products to the user according to the preference captured by the agent. The proposed solution is able to detect both quantifiable and non-quantifiable attributes through a user feedback system. Unlike previous approaches, this research allows the detection of unaccounted attributes that are not within the ontology of the system. No tedious change of the algorithm, database, or ontology is required when a new product attribute is introduced. This approach only requires the attribute to be within the description field of the product. The system analyzes the general product descriptions field and creates a list of candidate attributes affecting the user’s preference. A genetic algorithm verifies these candidate attributes and excess attributes are identified and filtered off. A prototype has been created and our results show positive results in the detection of unaccounted attributes affecting a user.