E-commerce has revolutionized retail experience through multiple insights that it provides customers and retailers. Customer' reviews are considered as feedback by retailers about their product and services. These feedback allow them to understand the industry better and enhance the efficiency and quality of their product.

A thorough analysis of a review can provide insights to retailers and E-commerce platforms about improvements that are required for better customer satisfaction. Analysis of such million reviews are handled by the E-commerce team manually and it is a very time-consuming task. This impacts the business of retailers as they get delayed feedback.

Your task is to create a simplified Machine Learning and/or Deep Learning approach to analyze the feedback & identify the key problems that are highlighted. You are also required to curate creative visualization reports that can be shared with retailers to aid the understanding of potential improvement areas for sellers and the E-commerce team.

Task: Your tasks are as follows:

1. Predict the ‘topic’ label for reviews under the test.csv file
2. Share your documented analysis of data, along with the source code, in Jupyter notebook file (.ipynb)

Data description

Data volume:

• Train.csv: 5960 x 3
• Test.csv: 2553 x 2
• Submission.csv: 2553 x 3 (Refer 'Sample Submission.csv' here)

Submission guidelines: You are required to upload:

1. The submission.csv file that contains your prediction labels such as Upload Prediction File 
2. A .zip file that contains your .ipynb or Jupyter notebook file under the Upload Source Files labels 
   • The notebook file must contain a well-documented analysis of data, logic, and reasons 

Evaluation metric: 

Machine Learning models: The evaluations will be based on the top-n accuracy metrics.

The top-n accuracy in this case is calculated by checking if the expected 'topic' is present in the predicted 'topics'. For each unique 'review text', 'review title', n is the expected number of topics per unique review. 

To know more about this evaluation metric, click here. 

$score = {(Top-n\ accuracy\ per\ review) / n}$

Data analysis:

• Algorithms, approaches, and logic used during the selection of models
• Creative in-depth researches (visualizations and reports) that are targeted at the following attributes:
  • Key aspects of product and services that customers require
  • Users’ underlying intentions and reactions concerning those aspects