Amazon scientists are prototyping algorithms that use crowdsourcing to identify product data, the company reports in a blog post. The researchers believe these algorithms could be used to predict human judgments of product quality on Amazon, which might improve people’s shopping experience by matching only high-quality products to search queries.

The work is something of a follow-up to a study Amazon published in early January that examined why Amazon customers buy seemingly irrelevant products while shopping for specific items. In an analysis, a team of Amazon researchers found that customers are partial to products that are broadly popular or cheaper than products relevant to a given search query. Additionally, their results suggested people are more likely to buy or engage with irrelevant products in categories like toys and digital goods than in categories like beauty and groceries.

In this latest study, which is scheduled to be presented next week at the ACM SIGIR Conference on Human Information Interaction and Retrieval (CHIIR) in Vancouver, the researchers presented crowd workers with images of pairs of related products, along with product information supplied by both sellers and customers. The researchers then asked the crowd workers which products were of higher quality and which terms extracted from the product information best explained their judgments.

Amazon product recommendations

VB Transform 2020 Online - July 15-17. Join leading AI executives: Register for the free livestream.

Each product pair in the study included one product that was actually purchased and one that was clicked on but not purchased during the same customer search query. Products also shared the most fine-grained classification available in the Amazon.com product classification hierarchy (e.g., Electronics, Home, Kitchen, Beauty, Office Products), and the terms presented to the crowd workers were chosen based on how frequently they appeared in texts associated with these categories.

The team found that while perceived quality wasn’t a good predictor of customers’ purchase decisions, it was highly correlated with price, such that customers generally chose lower-quality products if the items were correspondingly priced. Furthermore, the terms that best described the crowd workers’ judgment criteria came from the public customer-supplied information — that is, customer reviews and question-and-answer sequences in which customers answered other shoppers’ product-related questions — as opposed to the seller information.

“Existing research on product recommendation has mainly focused on modeling purchases directly, without attempting to find the reasons behind customer decisions. We believe that understanding the processes that underlie customers’ purchasing decisions will help us make better product recommendations,” wrote study coauthors Jie Yang, Rongting Zhang, and Vanessa Murdock. “This work represents one of several steps we’re taking in that direction.”