/Comment-Delegation-Propagation-Simulation

In decentralized social networks, comment handling with user propagation. Discrete event simulation application.

Primary LanguageJava

Comment Delegation Propagation Simulation

This simulation developed for an academic journal which is in review

In this simulation, we aimed to find out the best way to handle comments in a decentralized social network application while a user goes offline. During this offline time, the user gives delegation to a friend for handling comments below posts, and this delegation goes on until user come back or one of the previously delegated friends come back. We evaluated following three delegations heuristics along with an ideal scenario where an oracle tells which of the currently online friends will remain online the most;

  1. a random friend
  2. friend who has recently been online the most
  3. friend who has recently been online the most while the user if offline (i.e., most disjointly online).

We simulated these scenarios using the online timing of each user in the circle of Facebook volunteers data. We assumed the online timing of oneof the volunteer’s friends as the online timing of the main user as long asthe friend appeared online for at least a minute. We performed a timedsimulation of when the user goes offline, for how long the user is offline,and the delegation chain in the user’s absence. We recorded for how long aparticular chain length is observed while the user is offline.

Results

Figure 12 presents the chain length probability of each simulation case of the 2,266 timing samples from 16 users’ friend circles where probabilities are independently ranked for clarity. We determined the most disjointly online and most online persons based on the 5 previous day’s data assuming a circular time to use on the first five day’s calculations.

We also analyzed the average likelihood of each chain length for all data sets. Figure 13 shows the performance of each approach in linear and logarithmic scales. The offline probability of each friend – 74% and the likelihood of having no online friend – 7.3% are the same for all models. We observe on the average chain length of 1 is achieved by ideal with 61% of cases and both most online and most disjoint achieve it with 51%. However, random delegation can achieve minimal chains with 29% and performs much worse than others.