Thursday, March 18, 2010

AppBrain launches automatic filtering of "spam" apps

The Android market contains over 30,000 apps. Many of them are very cool additions for your phone, but there's also a fair share of low quality ones. Because it's so easy to launch apps on Android, some developers flood the market with lots of nearly identical apps which have little functionality.

When looking for good apps, these spam apps can really get in the way. I was recently searching for an app to track tennis game results. A number of results I got were by a developer who put 580 applications in the market. These apps are probably just meant to make a quick buck, as they don't have much functionality and all have low ratings. 

On AppBrain these results are now just shown as light grey lines. If you're interested in them, you can still click them to get the full search result.

This improvement makes searching for apps on AppBrain even more enjoyable and lets you find the best apps faster. 

Mathijs for the AppBrain team.

How we did it
We calculated statistics on all developers in the AppBrain database, which covers all apps that are available in the US. In total there were 28,936 apps created by 7,429 developers. Even though the average number of apps per developer was just 3.9, there were 32 developers who launched more than 100 apps in the Android Market. The number of developers grouped by their number of apps launched is shown in the figure below.

We used the ratings given by users to categorize apps as good (rating over 4 stars), average (between 3 and 4 stars), bad (lower than 3 stars) or too few ratings to decide.
The following figure shows apps broken down by quality for each developer output category:

Even though the high-output developers are few in number, they produce a giant proportion of all apps. These apps are often too unpopular to have a rating at all, and if they have, they're less likely to be rated well. However, not all high volume developers are spammers. Two counterexamples are for instance Better Android and K-J Themes. They both have published more than 100 apps, and they're almost all rated quite well.

After some experimentation we let both the fraction of bad apps and the fraction of very unpopular apps determine whether a developer is regarded as spamming the market. The 359 developers that were marked as spammers published over 10,000 apps. There are almost no good apps among those 10K, this means that the overall percentage of very good apps is bumped up from nearly 14% of all apps to well over 20% in our non-filtered set.

- We analyzed all 28,936 apps created by 7,429 developers.
- 25% of all apps are produced by only 32 app developers (0.4 % of all developers).
- Removing 359 spammers (who in total made over 10,000 apps) boosts the fraction of good apps in the market by 52%.

About AppBrain
AppBrain ( aims to provide the easiest and fastest way to get and share Android apps. The website is paired with an Android app (AppBrain Market Sync) for easy syncing between website and phone. AppBrain provides personalized app recommendations, fast and relevant search and browsing of other users’ installed apps.
AppBrain is developed by AppTornado GmbH, a leading Android developer team based in Zürich, Switzerland.

Friday, March 12, 2010

Android users download most tools, but spend most money on productivity apps

Just a month ago, we launched AppBrain, a website and Android app to discover and install great apps. We received fantastic feedback from our users, and since then we have launched many additional features, such as personalized app recommendations, new sharing options, a new app list browser, continuous updates to the app database, and many more.

Today we analyzed some of the interesting data that is available in the lists of apps users store on the AppBrain servers. Recently, others have shown that the number of free apps in the Android Market is relatively high [Reuters, Techcrunch], and that games seem unpopular in the Android Market [Techcrunch]. Interestingly, these conclusions were drawn from indirect data (availability of apps in the market and traffic statistics of a website showing Android apps), therefore it would be interesting to look at actual install figures.

We extracted a representative sample of application installations from the AppBrain database. This data is uploaded by the "AppBrain Market Sync" application and thus directly reflects what people have installed on their phones. (However, there are still limitations of this data. As the application lists are uploaded by a special applications for app lovers, our population probably is biased towards people who are more interested in apps in general. Therefore they'll likely have more apps installed than the overall population.)

Phone models
The data we present comes from 5,000 applists, which were randomly selected from our database of uploaded app lists. The phones from which these were uploaded were running Android 1.5 in 35% of cases, 1.6 in 34%, 2.0 in 21% 2.0 and 2.1 in 10%. This is quite similar to the official numbers from early January listed on the Android developer site (except for the Nexus which didn't exist yet in early Jan.).
We will present our stats broken down by Android version, as later Android versions are better and might reflect behavior of newer users more than the older versions. Especially for gaming the later Android versions (and the better hardware they run on) are much better than the earlier ones.

Installations of games and paid apps
First we look at the number of games installed per phone. Figure 1 shows in light green (left axis) the average number of games per device and the dark green bars (right axis) show the fraction of all apps that are games. The number of games s around 6.5 across the board. However, in later Android versions it makes up a lesser percentage as people are installing more apps in total.

Figure 2 shows install numbers for paid apps. About 70% of all phones in our sample had at least one paid app, and this number seemed to rise with the Android version. The mean number of paid apps installed is about 5, and total money spent on them around $18.25, without a clear trend across Android versions.

Installed apps and money spent per category
To get a better idea of what apps Android users spend their money on we analyzed the number of installed apps per market category. In figure 3, we have plotted each category with on the X-axis the mean number of apps installed per phone in that category, and on the Y axis the mean number of dollars spent on those apps. The dark bubbles are game categories. The size of each bubble represents the total number of available paid apps in that category.

It is interesting to see that our users install very many Tools apps. However, this could be a bias in our sample as only Android users that have installed the AppBrain Market Sync app (which is a tool) occur in this study. Still it is interesting that even though users install very many tools apps, it is not the category they spend the most money on.
It is striking to note the difference in the graph between the "Entertainment" and "Productivity" categories. Even though users have about the same number of those apps on their phones (4.5 vs. 4.2), they spend much more on the productivity apps (on average $4.48 vs. $0.55). And as the size of the bubble shows, there are more paid applications competing for the small entertainment pie than for the bigger productivity one!
With regard to games, the Arcade & Action category seems to be a fine choice to develop a paid application for. It is #5 in terms of money spent ($1.61), with clearly less paid applications than the Tools or Travel categories. The Brain & Puzzle category seems less attractive, Casual even less so, and Cards & Casino is truly tiny.

Happy app browsing,

Mathijs, Software Engineer