Inferring Shared Interests from Tweets

Social networks have intruded into the every day’s life. Users contribute with uploading content and sharing it with friends and public. From this information knowledge on who know whom, who communicates with whom, who meets whom and who shares the shame interests with whom can be extracted. These four aspects of human behavior can provide valuable insights while designing networking algorithms and protocols for opportunistic networks. Having efficient algorithms for exchanging information can help in the decentralization of the current centralized approached used to store the information. The distribution of the information could be quite useful in cases where the infrastructure is absent and thus fails to provide services. So far research on the first three topics and the relations between them has be done. In this thesis an effort on the introduction of how sharing the same interests correlates with the first three topics is done. In order to be able to do such a correlation, a measure on assigning similarity scores between users with respect to their shared interests needs to be created. In this thesis the problem of text comparison is investigated. A frequency of words based approach is chosen for the comparison of the texts posted on the social network called "Twitter". Five metrics are created based on special characteristic of twitter texts. The validation of the metrics is done in a set of 250 users. The results show that it is feasible to assign similarity scores between twitter users and thus provide a tool for measurements with the ultimate goal to deepen the understanding of how a pairs of users sharing same interests correlates with knowing each other, meeting each other or communicating with each other.