Reference Product Search

The reference products can be used as alternatives closely related to a product of customers' interest. In candidates to support downstream modeling tasks and business general, the solutions consist of two stages: a candidate product applications. The search method consists of product representation set is retrieved first, followed by a task-specific ranking model to learning and fingerprint-type vector searching. The product catalog generate the results. Often, research interests focus on a ranking information is transformed into a high-quality embedding of low model built to optimize towards such a business application, but a dimensions via a novel attention auto-encoder neural network, and suitable candidate product set is required to feed the ranking model the embedding is further coupled with a binary encoding vector for and it is not well discussed in the literature.