| dc.description.abstract | This thesis have enlightened the topic of social bots and how to indicate their presence. As a practical part of the work in this thesis, a program for indicating social bots on Twitter were developed.
The presented solution have both a technological and a biological aspect. The technological aspect is the indicator program. The program analyze an account to look for indications of social bot activity. The different indicators used in the program were inspired by several earlier studies on social bots. As a result of analyzing data available through the Twitter API, some of the indicators are also based on new ideas. The program uses analyses of several different features and behaviors. By implementing several analyses, the program should be theoretically able to detect social bots with different behaviors. Instead of having a detection tool, the program analyze an account to look for indications of social bot activity. If any indications are present, the analyzed account is to be further inspected. This is where the biological aspect comes in. Further analyses can be performed by either a dedicated person or by crowdsourcing. Crowdsourcing is a promising approach of analyses. It is effective and spreads the workload over several people. The biological aspect of the solution is included because humans can detect small inconsistencies that is hard to define in algorithms.
The indicator program works in near real-time. Right before analysis, account information and tweets from the account is downloaded. Right after account and tweet data is retrieved, the analyses are performed. The analyses are performed in about one to four seconds (url-analysis excluded). Having a fast-working indication program can be crucial for fast detection of social bots on Twitter. By having fast indication of social bots in place, the impact of the social bots cause is minimized.
Social bots have become very advanced in the recent years. And they continue to develop. Advancements in fields such as AI, makes detection of social bots more difficult. By continuing to develop methods for detecting them, we can be ready for the next types of social bots. To stay ahead of new social bots that come out, we also need to research new approaches for detecting them. | |
