Abstract: This paper explores the intersection of entity matching, machine learning and online marketplaces. Entity matching, or finding references that point to the same real-world object, has existed as a field for more than half a century. The field of machine learning classification began about twenty-five years ago and has seen breakthroughs recently with processing natural language. In the last ten years, online multi-vendor marketplaces have proliferated. To maximize the benefits of sellers competing for customers, the marketplace firm must be able to consolidate the product offers in its catalog that belong to the same real-world product. This paper addresses the following questions: 1. To what degree can machine learning algorithms solve the entity matching problem for online marketplaces? 2. Which types of feature representations of text produce the best classification results when the offer catalog is rife with missing values? The results demonstrate that machine learning classifiers can detect offer matches and non-matches, but exhibit the tradeoff between precision and recall. Additionally, despite the sparsity in the dataset, an attribute comparison approach to feature representation proves superior to the single-document representation.
Key Words: Entity Matching, Product Matching, Machine Learning, Natural Language Processing