End-2-End Product Search using Deep Learning

Research Project by Duncan Blythe (ex-member), Han Xiao (ex-member)

Product search is one of the key components in an online retail store. A good product search can understand a user’s query in any language, retrieve as many relevant products as possible, and finally present the results as a list in which the preferred products should be at the top, and the less relevant products should be at the bottom.

At Zalando Research we are investigating whether a deep-learning model can be trained to perform this task mapping the user query as a sequence of symbols (represented by integers) to a representation of the products (represented as multidimensional arrays of floating point numbers). Our approach is illustrated in the figure below:

Our approach is illustrated in the image below:


Our approach uses as components:

  • Recurrent neural networks
  • Convolutional neural networks
  • Deep learning trained using a ranking objective function

The prototype system has several attractive qualitative features displayed in the sample queries below:


“kleid rotes lang” – compounding of attributes understood


“kleifd roteeews lang” – severe mispelling implicitly corrected