Blibli Future Program Batch 5 — Data Track Phase 2: Data Analysis
About the project
In this project, we are required to extract reviews (comments) from Google Play Store and Apple Store. With these reviews, we will then create a dashboard and perform topic modeling using deep learning.
The reviews considered in this project are the top 10 e-commerce apps cited from Selular.id, which are Tokopedia, Shopee, Bukalapak, Lazada, Blibli, Orami, Ralali, Bhinneka, JD.ID, and Zalora.
Data Analysis
In the previous post, we have extracted all the data from the source layer until the datamart layer. Now, let’s see what we can analyze with these data.
We will go through the charts one by one. However, let’s see what is the problem statement, hypothesis, and constraints for this analysis.
Problem statement
Blibli.com has been one of Indonesia’s e-commerce marketplace for 10 years. Hence, the number of reviews written by its users is supposed to be a lot. Reviews might contain important information that reflects how the users’ experience while using this app. Therefore, the analysis of these reviews to draw insights is needed.
With this analysis, we will figure out how do Blibli.com users rate the Blibli.com application?
Hypotheses
We created three hypotheses to help with our analysis:
1. Blibli.com has average ratings greater than half of the top ten e-commerce apps.
2. The percentage of positive reviews is greater than negative reviews.
3. The number of reviews created is increasing each day.
Constraints
- Only data from the second half of 2021 is considered to compare the e-commerce apps. To analyze Blibli.com’s rating, a one-year data is available.
- The source of the data is extracted from Google Play Store.
In our database, we have a one-year review except for Shopee (started in May). This is the reason why we used the data from the second half of 2021 to compare the e-commerce apps.
Analysis
First, we can see that there are 1,2 million reviews in the second half of 2021. Bukalapak, with a rating of 4.39, has been the e-commerce with the highest average rating. Blibli.com itself ranks eighth with an average rating of 3.99. This shows that our first hypothesis is rejected. How about the number of reviews per app? Shopee with a total of 647 thousand reviews, ranks first being the e-commerce with the highest number of reviews.
We have seen how Blibli.com is in terms of rating and number of reviews compared to other e-commerce apps. Next, let’s see how Blibli.com is performing for the entire 2021.
In 2021, Blibli.com has 16 thousand reviews, with 74.25% of positive reviews, 20.63% of negative reviews, and 5.12 % of neutral reviews. Since the percentage of positive reviews is greater than negative’s, our second hypothesis is accepted. Lastly, the percentage of reviews replied to is 97.18%. Wow! Almost all of the reviews have been replied by Blibli.com’s operators!
Next, we have a line chart that shows the average rating on the first day of 2021 versus the last day of 2021. Note that this average is a cumulative average, which means that the last day of 2021 averages all the reviews from January 1st until December 31st, 2021. Blibli.com has been dropping from 4.16 to 4.03.
We have another line chart, the total created reviews per day. Since the number of reviews created is not increasing smoothly each day, we can conclude that our third hypothesis is rejected. The highest number of reviews created is on December 7th and the lowest number of reviews created is on June 15th.
At this point, we would like to suggest that Blibli.com should think of a way to increase its number of reviews. Recall back to these charts.
By considering 8 thousand reviews, Blibli.com has an average rating of 3.99. If we compare the average rating from Shopee, Shopee achieves an average rating of 4.36 with a total review of 647 thousand reviews. With such a great amount of reviews, we are more confident that our app’s rating is in such a number. Blibli.com needs to gather more reviews to know what the average rating is. How can Blibli.com increase its number of reviews? Let’s see the reviews created date chart of Shopee.
There is a sharp increment of reviews created from December 3rd until December 15. Specifically, it is on December 2nd and December 12th (12–12 — Double date). Another sharp increment of reviews created can be seen in the second week of November. Specifically, it is on November 11th (11–11 — Double date) and November 14th. There is a unique pattern here, in which the number of reviews created tends to increase when it is on the double date. Ok, Lazada has a great total of reviews too! Let’s see its total created reviews per day.
In here, we can more clearly see the sharp increment of reviews created date. It is on November 11th and December 12th — once again, a double date. There are a lot of promos during these unique dates. What if Blibli.com makes good use of these dates to gather reviews as much as possible? Perhaps, those who write a review on these dates will receive a point or voucher to purchase anything with a discounted price? Woo who hates discounts? However, a more thorough analysis should be performed to fully decide what is the best way to gather more reviews.
Up until this point, we have analyzed the total reviews, the percentage of positive, neutral, and negative reviews, the percentage of reviews replied, average rating in the second half of 2021, the average rating at the beginning versus the last day of 2021, and the number of reviews created each day. In short, we have just analyzed the characteristics of the sentiments/reviews. We haven’t known what topics the users are talking about? Is it about the promotion? Delivery fee? or application? No worries, because we are going to analyze the topics of the reviews in the Data Science section next! I hope you enjoy this analysis and see you in the next section ;D.
