Mobile Apps Price Recommendation

There are over 2 billion apps available on Google Play, around a third of them are paid. Nevertheless the task of finding the right price of an app is not easy and it is highly dependent on category of the app, the implemented features, how polished the app is, how much work went into developing it. Mobile apps developers do not have the necessary skills to correctly determine the price of an app, but setting a good price from the beginning is very important: a too high price means the app is not downloaded and will fail from the start, a too low price means the developer won’t make enough money to cover the development and the future maintenance costs.

The goal of this project is to build an intuitive web UI that based on meta-data from mobile apps marketplaces (e.g. Google play) and machine learning models is able to recommend a price for a new app based on different features of the app (description, size, images, etc.). The tool should support two scenarios: new apps that are launched as paid or existing apps for which a free version is available and the developer wants to also launch a paid one.

The main tasks of the project are:

• Gathering data:

  • crawl metadata of paid apps from Google Play;

  • find paid apps that are available as open source apps on the F-Droid repository or Github to obtain a dataset with source code of paid apps;

• Build machine learning models for price recommendation:

  • Experiment with different machine learning algorithms (Gradient Boosted Trees, Nearest Neighbors, etc.);

  • Experiment with different sets of features to recommend the correct price (likely use price ranges, not absolute values);

  • Use ML to find the most similar apps and use these to make a price recommendation but also explain to the user how the recommendation was made, experiment with different feature for finding the most similar apps (name, description, size, etc.);

  • Extract from the description of an app only the part that describes its features (use some heuristics), does this lead to better price recommendations?

  • Investigate weather code metrics improve the price recommendation(use only the apps for which the source code is available);

• Build an intuitive UI that uses the defined models to make different price recommendations

• Evaluation:

  • precision/recall/f1-scores per category from a test set of the crawled apps metadata;

  • based on similar apps as they are recommended from Google Play.

• Scope: 2 students.

Posted: 15.03.2017

Contact: Adelina Ciurumelea