Nowadays, social recommendation is an important task to enable online users to discover friends
to follow, pages to read and items to buy. Big companies such as Facebook and Amazon heavily
rely on high-quality recommendations to keep their users active. There exist numerous
recommendation algorithms --- from the simple one based on common neighbor counting, to
more complicated one based on deep learning models such as Graph Neural Networks. The
common setup for those algorithms is that they represent the social network as a graph, and
perform various operations on the nodes and edges.
In this project, you will implement a simple application to generate customized
recommendations based on user queries. Specifically, consider that Facebook users in different
countries want to follow new friends. They would send their queries to a Facebook server (i.e.,
main server) and receive the new friend suggestions as the reply from the same main server.
Now since the Facebook social network is so large that it is impossible to store all the user
information in a single machine. So we consider a distributed system design where the main
server is further connected to many (in our case, two) backend servers. Each backend server
stores the Facebook social network for different countries. For example, backend server A may
store the user data of Canada and the US, and backend server B may store the data of Europe.
Therefore, once the main server receives a user query, it decodes the query and further sends a
request to the corresponding backend server. The backend server will search through its local
data, identify the new friend to be recommended and reply to the main server. Finally, the main
server will reply to the user to conclude the process.
