For this project, I analyzed reviews from Ulta.com to examine the trend of ratings for Briogeo's "Don't Despair, Repair!" hair mask. I webscraped data from Ulta.com using python's beautifulsoup and selenium packages. After I used the resulting dataset to examine trends in the product's rating over time. I also conducted a text analysis on the product's reviews using VADER sentiment analysis.
Ultimately, I found that ratings for the mask as well as positive sentiment expressed in the text of the reviews have both decreased since 2020. This aligns with my own research and reading reviews on reddit that customers were disappointed with the mask since they changed the product's formula.
Decrease in ratings
Decrease in positive sentiment
The final analysis product was a dashboard saved on Tableau Public.
I frequently use Briogeo's "Don't Despair, Repair!" deep conditioning mask in my own hair. I have loved the results, but I have heard and read online that the formula for the mask changed. As a result, some customers are not as happy with the mask as they once were. As a "Hair Care Guru" in my spare time and Data Analyst for my day job, I decided to look at the reviews for this product and see if the ratings for this product have changed overtime.
For the analysis, I webscraped reviews for this product from Ulta's website. I conducted exploratory analysis and analyzed the product ratings overtime. Likewise, I also analyzed the sentiment of the written reviews overtime.
In sum, I found the following: the average rating on Ulta.com for this mask has decreased since 2020. Likewise, the positive sentiment of the reviews for this product have decreased since 2020.
- Reviews from Ulta (Webscraped reviews from Ulta's website, n = 2958 reviews, spanning from November 2014 to May 2022)
-
data/briogeo_rating_ulta.csv - N = 2958 reviews. Each review is one row. The following attributes are included: Unique ID for Review (Review_ID), Rating [1-5] for the Review (Rating), Location of customer that submitted the review (Location), Submission date of review (Rating_Submit_Dt), Username of customer that submitted the review (Username), Full text of Review (Review).
-
data/briogeo_ulta_rating_sentiment_scores.csv - Same attributes and number of rows as the briogeo_rating_ulta.csv dataset, but this dataset also includes the addition of the VADER sentiment analysis. The additional attributes are the Postive Score, Negative Score, Neutral Score, and Compound Score.
-
data/briogeo_ulta_rating_wordlist.csv - Long dataset with each Review_ID repeated for the words within each review. Stop words are excluded.
-
data/briogeo_ulta_rating_bigrams.csv - Long dataset with each Review_ID repeated for the bigrams within each review. Stop words are excluded.
-
data/briogeo_ulta_rating_trigrams.csv - Long dataset with each Review_ID repeated for the trigrams within each review. Stop words are excluded.
-
- Ulta Scraper.py - Web scraping script
- briogeo_rating_nltk_analysis.py - VADER sentiment analysis and create wordlists (bigrams & trigrams) for further analysis in Tableau
- briogeo_case_study.twbx - Packaged Tableau workbook that includes the final analysis dashboard
I uploaded the tableau workbook on Tableau Public: here. In the dashboard, you can clearly see that the ratings for the hair mask have decreased since 2020. Likewise, the proportion of reviews deemed as having a negative sentiment based on the VADER analysis has increased in recent years. In addition to looking at ratings and sentinment for the reviews, there is also an option to see the top 10 unigrams, bigrams, or trigrams by year and sentiment classification (positive, neutral, negative).
Overall, this analysis confirmed my suspicions. I still personally love the mask, but other people are not as happy with it in current times. While I am happy with the analysis, there could be improvements especially with the text analysis and playing around with different algorithims. Perhaps in the future, I will update the analysis.
In addition to Ulta reviews, I also webscraped reviews for this product from Makeup Alley. However, I did not end up using them in my analysis. The dataset and code for the Makup Alley reviews are in this repository as: data/briogeo_rating_makeupalley.csv & Makeup Alley Scraper.py.