6. Conclusion
Our study uncovered a number of interesting findings related to the specific nature of online social networking environments. By using temporal information of real OSNs we have found that availability patterns of single individuals has a non-trivial relationships with those of their close friends. We found that the size of the active network of each user is slightly correlated to the time spent online by the ego, such as the average session length (0.36) and the number of daily sessions (0.20). We have shown the extent to which availability patterns of each Dunbar’s circle affect the availability pattern of the user. Namely, social ties on innermost circles not only are stronger in terms of volume of communications, but also show higher similarity of the users availability pattern. We have also measured the average number of times ego and their alters are both online (1,1) or offline (0,0) in the same time slot and the number of times that either ego (1,0) or alter (0,1) are online, separately for each circle. The average percentage of (0,0) matching increases as we consider alters located in the outermost circles. The matching of the form (1,1) has the lowest percentage value for all circles and they increases as we consider users of the innermost circles. Finally, the innermost circle has approximately the same percentage of (0,1) and (1,0) while for the outer circles, the percentage of the (1,0)/(0,1) matching increases/decreases of about 1%. By characterizing the impact of this similarity between the availability pattern of the users and their friends in term of conditional probability, we have shown that users have more probability to be online when at least 10 of their Dunbar’s friends are online . Finally, we have shown that users who happened to be both online during abnormal temporal window (from 0am to 8am) have greater similarity than users connected during the classical time windows (from 8am to 0am, i.e. preceding the sleeping time). These understanding have a significant impact on the success/failure of the OSN and they could lead to many benefits if properly considered and addressed during the design of the designing of the next-generation OSN infrastructures. As future work, we would like to investigate the impact of our finding on content distribution patterns. Answering these questions will let us explore opportunities for efficient content distribution and data replication, as well as advertisement and recommendation strategies. Lastly, based on our results, we plan to build a user churn model able to shape availability pattern of the user behavior by incorporating the most part of our findings, including sessions distribution, tie strength and temporal features.
