Nearly everybody once faced this situation: you visit an online shopping site, head over to the search box and start typing. Next, you'll get a dropdown list of search suggestions – but many times these suggestions are of poor quality. On the other hand, search suggestions are a helpful feature and add value to the search user experience. In addition, they can also increase your shop performance w.r.t. indicators such as 'add to basket' and 'order value'. But what most people don’t expect: bad suggestions can potentially harm your shop performance.
We've observed during A/B tests on our top tier e-commerce site that poor search suggestions may have a negative impact on our business performance indicators. Based on this observation we decided to start a project to iteratively improve the quality of our search suggestion engine.
In this talk, we will share our iterative methodology and our approach to solving the suggestion problem using Elasticsearch, Lucene and machine learning techniques.