Ir para o conteúdo

Software livre Brasil

 Voltar a Blog
Tela cheia

AppRecommender - Last GSoC Report

22 de Agosto de 2016, 16:44 , por Luciano Prestes Cavalcanti - 0sem comentários ainda | Ninguém está seguindo este artigo ainda.
Visualizado 1410 vezes
My work on Google Summer of Code is to create a new strategy on AppRecommender, where this strategy should be able to get a referenced package, or a list of referenced packages, then analyze the packages that the user has already installed and make a recommendation using the referenced packages as a base, for example: if the user runs "$ sudo apt install vim", the AppRecommender uses "vim" as the referenced package, and should recommend packages with relation between "vim" and the other packages that the user has installed. This work is done and added to the official AppRecommender repository.
During the GSoC program, more contributions were done with the AppRecommender project helping the system to improve the recommendations, installation and configurations to help Debian package.
The following link contains my commits on AppRecommender:
During the period destined to students get to know the community of the project, I talked with the Debian community about my project to get feedback and ideas. When talking to the Debian community on the IRC channels, we came up with the idea of using the popularity-contest data to improve the recommendations. I talked with my mentors, who approved the idea, then we increased the project scope to use the popularity-contest data to improve the AppRecommender recommendations.
The popularity-contest has several privacy political terms, then we did a research and published, on the Debian Planeta post that explains why we need the popularity-contest data to improve the recommendations and how we use this data. This post also contains an explanation about the risks and the measures taken to minimize them.
Then two activities were added to be made. One of them is to create a script to be added on popularity-contestThis script is destined to get the popularity-contest data, which is the users' packages, and generate clusters that group these packages analyzing similar users. The other activity is to add collaborative data into the AppRecommender, where this will download the clusters data and use it to improve the recommendations.
The popularity-contest cluster script was done and reviewed by my mentor, but was not integrated into popularity-contest yetThe usage of clusters data into AppRecommender has been already implemented, but still not added on official repository because it is waiting the cluster cript's acceptance into the popularity-contest. This work is not complete, but I will continue working with AppRecommender and Debian community, and with my mentorshelp, I will finish this work.
The following link contains my commits on repository with the popularity-contest cluster script's feature, as well as other scripts that I used to improve my work, but the only script that will be sent to popularity-contest is the
The following link contains my commits on repository with the AppRecommender collaborative data feature: 
Google Drive folder with the patch:

0sem comentários ainda

Enviar um comentário

Os campos são obrigatórios.

Se você é um usuário registrado, pode se identificar e ser reconhecido automaticamente.