Before launching new products, companies engage in market testing. It's crucial for them to thoroughly understand and analyze customer sentiments expressed in reviews, particularly regarding health and personal care products The motivation behind this project stems from the observation that customers often churn or disengage from products/services when their sentiments or concerns expressed in reviews are not adequately addressed by companies. By analyzing these sentiments, we aim to uncover valuable insights into customer pain points, satisfaction levels, and areas for improvement. Addressing these sentiments can lead to enhanced customer retention, loyalty, and ultimately, improved business performance Who: The stakeholders involved in our project include: Reviewers: Customers who provide feedback , reviews Product Companies: Manufacturers and sellers of health and personal care products who are interested in analysing customer feedback to improve products and services. E-commerce Platforms: Online platforms where customers leave reviews and purchase. They benefit from improved customer satisfaction and engagement
Data Sources We obtained our data from an open data source of Stanford Large Network Dataset Collection (link) Collecting Data In the future, we can collect data from product reviews across multiple online e-commerce platforms Features Our data contains review text, ID of the reviewer, product ID, time of the review, helpfulness of the review, overall rating, summary of the review Building Models We created machine learning models like logistic regression, SVC, and random forest to understand the sentiments. We have tested our model on a size of 0.2. We can update the models every 3 months or when there is a new product to be launched
The predictions provide insights into how customers perceive the products. This can further be used to make decisions related to product launches and improvisations. It also helps prevent customer churn by staying on top of customer feedback and making changes quickly
Before deployment, we conducted an in-depth analysis using metrics such as confusion matrix, ROC AUC, precision, and recall scores. These metrics helped us assess the model's fit with the data and its accuracy in predicting sentiments.