The first option involved a better selection of images (of higher quality). The second is search ranking optimization. Both were tested. 

The former solution focused on optimising the miniature images appearing to users, to potentially improve CTR with their attractiveness and subsequently the conversion rate. For this purpose several approaches were tested, including those based on regression and classification. 

The second option’s goal was optimising the ranking of the search results. The best features for the problem solution were selected by our data scientist and in delivery NLP was utilised together with a deep network to obtain optimal results. 

The second solution proved significantly more effective, therefore to most efficiently support conversion rates, we continued with this project. The goal was to optimise it for the company’s search options.

The implementation process went as follows: At first the focus was on getting a clear understanding of the problem being faced as well as the solutions that were already tried and tested. Part of the available data was provided for investigating this matter. Once the knowledge was acquired, the planning phase started. Result oriented options only. Our data scientist was providing the solutions and monitoring results delivered by the models. Once the solutions were investigated, the selected model was actively investigated. The data scientist monitored what went wrong and how to improve the model.