Amazon Marketplace provides a platform for individuals and businesses to sell products to hundreds of millions of online customers. Currently, more than 40% of Amazon sales result from third-party sellers.

In order to improve and optimize the experiences of their third-party sellers, Amazon provides Seller Forums boards on which sellers can post questions and answers to questions.

Worldwide, Amazon sellers post about 65,000 questions and 2,100,000 answers per year. Without an automated way to analyze these posts, it is very difficult for Amazon to get a sense of trending topics, pain points and areas to be improved.

SIFT, Seller-Forums Information Filtering Tool, analyzes the Seller Forums using natural language processing to classify the posts into groups clustered around common themes. These clusters identify currently trending topics within the seller forums, thereby helping the Amazon Seller Services team to resolve potential issues for their sellers.

The clustering of posts into topics can be refined by specifying the number of clusters to be created, a date range and other cluster-specific settings.

SIFT’s dashboard displays the current state of trending topics on the Seller Forums. Amazon Seller Services team members can view, search and filter posts related to each cluster.

SIFT is written in Python using the Django web framework. A MySQL database is hosted on Amazon’s Relational Database Service, which is hosted on Amazon’s Elastic Cloud Computing through Amazon Web Services.