Recommendations in Marketplace

此文章不再受維護,內容可能已經過時。

我們的志工正在翻譯這篇文章。在完成之前,也許英文版可以先幫上忙。若您也想協助翻譯文件,請點擊此處

Recommendations are a customized list of apps provided to users. These recommendations are algorithmically calculated using anonymous installation data from registered Marketplace users.

app recommendations

Who will see app recommendations?

App recommendations in the Firefox Marketplace are provided to users who:

  • have a Firefox Account and logged in to it when using Marketplace
  • have not opted out of receiving recommendations.

How to turn this feature on or off

You can choose to opt in or out at any time. Opting out will also prevent your data from being used (anonymously) to calculate recommendations.

  1. When you've logged in, click on the settings icon, then choose Edit Account Settings.
    account settings marketplace
  2. Put a check mark next to Yes, show me recommendations based on my installed apps. Uncheck the box to opt out.

How does the recommendation system work?

The recommendation system used in the Firefox Marketplace is developed by Telefonica Research Labs. The system provides better and more personalized recommendations as you install more apps. Before you install any apps the system recommends apps based on their popularity since the algorithms do not have enough data to discover personalized tastes.

The main technique used in the recommendation system is called collaborative filtering (CF). It is a recommendation technique which generates personalized recommendations for a user using opinions of other users who have similar taste.

The specific CF technique used is called Matrix Factorization, which originated here. The exact algorithm used is described in this research paper. We use only anonymized user app installations data and don’t use other information described in this paper such as home vs work location.

For Developers: The (open source) code for the algorithm implementation is located at: https://github.com/grafos-ml/test.fm/blob/master/src/testfm/models/tensorcofi.py. The backend server code (named Frappe) of the recommender system is located at: https://github.com/grafos-ml/frappe. .

How is data collected and used?

When you log in to the Marketplace and install an app, Marketplace records the application, the country of installation, date and time of installation, and language preference. It does not know if you actually like the app, used the app, or removed the app.

User data used in calculating recommendations include:

  • apps installed and date of installation
  • country setting of user
  • language setting of user

If you install apps while you are not logged in, Marketplace has no record of your app installation and does not record any information other than the fact that the app was installed by someone in a particular country.

All the data is stored and processed within the Mozilla Virtual Private Network. We never share any of this information with the app developer and adhere to the terms of the Marketplace Privacy Policy where we use data to improve the user experience.

這些好人幫助我們撰寫了這篇文章:

Illustration of hands

成為志工

在此回答問題並幫助我們改善知識庫內容,與其他人一起切磋琢磨專業能力。

了解更多