How to improve product recommendations for customers visiting dell online product store and increase the sales figures ensuring customer satisfaction.
- We gather customer's activity data from different sites example, flipkart, amazon and google using a browser extension (or Ad Networks if available).
- The gathered data is used to train a machine learning algorithm which then predicts the preference/priority value for all dell products in the store inventory.
- We further improve the predictions by implementing refinement algorithms based on dell's requirements to target the user with specific categories of products
- The final recommendations are shown to the user. Maximum customer satisfaction is achieved as the products are of high specifications in minimum customer budget.
- Now we track the user's interaction with the recommended products and refine the recommendations further based on the actions taken like - product added to cart or products checked out.
- Interactions with a product on dell.com are used to track product recommendation conversion ratio. This ratio is used to manipulate the recommendations and display popular products.
- The customer can provide a feedback in the form of a comment. This comment is run through a sentiment analysis algorithm and the factor is used to improve the predictions.
- We provide a dashboard for the marketing and analysis team to visualise the performance of our recommendation engine and its outcomes.
The following diagrams show a high level view of data flow in the solution design.
Following are some important technologies and libraries used.
| Technology | Version |
|---|---|
| python | 3.6.5 |
| Django | 3.0 |
| nltk | 3.4.5 |
| numpy | 1.17.4 |
| pandas | 0.25.3 |
| psycopg2 | 2.8.4 |
| scikit-learn | 0.22 |
| scipy | 1.3.3 |
| sklearn | 0.0 |
| tensorflow | 1.13.1 |
| tflearn | 0.3.2 |
| postgreSQL | 12.1 |