Challenges
The client, an e-commerce affiliate, wanted to increase customer engagement through content. So far, their content had trivial items like product rank and product features. These were not enough to generate traffic. Also, e-commerce affiliate is a competitive market, and it became imperative for the client to provide a unique proposition to its customers. The information supplied needed to be informative, factually correct, and based on current trends.
Solution
We used a template-based approach. The template was filled with analysis, while the analysis consisted of sentiment, feature, and comparison.
Creating engaging templates and data preparation was a challenge. The template was designed keeping in mind the content produced by analytics. The Data Modelling required sample data, and this sample data was prepared in-house using 1 million reviews available to us generated by the client’s previous listings. This enabled us to achieve a high model accuracy. For feature extraction, the model did not have high accuracy, and thus a hybrid approach was used. The output of the model was manually verified to ensure the correctness.
Outcome
The in-house sentiment model developed very well, and the feature extraction model worked very well for the client and helped them achieve their target of engaging with their target customers better.