Analyzing User Sentiment & Insights from DeepSeek iOS App Reviews
DeekSeek has been making the top app charts for last few weeks since the app was released on both Apple app store and Google play store.
We thought to analyze Deepseek’s user reviews using AppStoreGenie.
We used the AppStoreGenie review tracker to download every new review that appeared on the Apple app store for Deepseek and later used the data report feature to generate the graphs.
Firstly, lets get word counts for all the reviews and plot that against rating value and against review date.
There are a bunch of 1 star reviews with lots of text and that might indicate some issue with the app.
There is also definitely an uptick in review text count after 30 jan 2025; that could indicate either some fraudulent reviews trying to sink the review score, or on other hand, it can just be due to app gaining more popularity.
We can check if there was any flooding of reviews with highly positive or highly negative reviews by plotting the review star rating against time.
This seems pretty well balanced except that it seems that in the initial days of the app (pre 23 Jan 2025), it mostly had neutral and positive reviews, however, the week of January 29, the reviews per day really ramped up and as we can see, it started getting a lot of 1 star review (blue area) during that week.
Let us look at overall rating distribution or the histogram that is available on the app details page.
Rating Distribution (Overall) shows combined rating from users that have left a detailed text review (like the one we are analyzing here) and just a star rating.
Let us look at only users who actually left a text review.
So our sample contains more 1 star reviews than what the overall ratings indicate.
Let us look at how the overall rating score has been dropping since last 6 weeks due to a bunch of 1 star reviews.
The blue line is the average rating till date. So every new review that is published changes the overall rating score; this curve visualizes the overall impact on the rating value with each new review.
If we see a string of “bad” reviews, than the average rating of this recent reviews sample (green line) will plunge below the overall rating score (orange line) and vice versa.
This could mean that user satisfaction in recent past is less than overall user sentiments across all app versions.
This decrease in the average rating till date does have a negative effect on the overall rating score and we can see that here (the orange line).
It seems that there was a strong onslaught of negative reviews around the week of Jan 30 (2025) making the blue line dipping below green and it has led in some decrease in overall rating, but now the overall rating has stabilized for now.
Let us look at wordcloud of the review text itself.
Its better if we plot just the 1 star reviews. Most one star reviewers are concerned about deepseek being a Chinese app, about their data or information being secure and lastly, about server being busy. Tiananmen square and propaganda also features in this wordcloud.
Let us instead look at 4 and 5 star reviews
In this case most reviewers seem to have compared deepseek to chatgpt, and it seems like the model is good or great.
The next step will be to analyze the Android version of the DeepSeek app and see if the user reviews are saying the same thing or not.
All of the pltos and data are generated on our AppStoreGenie platform.
