akshaykapoor347/NLPClothingReview

I have worked on the analysis of reviews of an ecommerce clothing website where I have performed EDA and sentimental analysis. For sentiment analysis, I performed cleaning on it like removing the punctuation and the stop words from it, then tokenizing and like removing words which were not important like which have length less than 3. I performed analysis such as finding the most common words used in a review. (dress, size, love, like, top) Then made use of text blob to find the sentiment of the reviews and created a list of most commonly used words in positive review and a negative review. Then used a classification algorithm like naïve Bayes to train the model to rate to a review and tested it on the new data. Count vectorizer Results: 1) Reviews with 3 and 4-star rating had the longest reviews. 2) Users shopped for tops 60 percent more than bottoms 3) Got 85 percent accuracy in the naïve bayes model.

Jupyter Notebook