That thetrending topic or most popular Twitter topics to spread “junk” content(spam) has become a fairly common practice. To solve this, UNED researchers have developed astatistical method that allows detecting junk content on this well-known microblogging platform. To demonstrate their effectiveness, they have analyzed 20 million messages, 34,000 trending topics and six million web addresses.
Of all the content analyzed, the toolcorrectly classified 93.7% of malicious messagesand 89.3% of valid messages. “Only 6.3% of valid tweets were erroneously classified as garbage,” explains Juan Martínez-Romo, researcher at the Department of Computer Languages and Systems of the UNED and co-author of the work.
The research, which is published in the journalExpert Systems with Applications, uses language models to detect spam and, unlike other methods,does not analyze user profiles but tweets. “The creation of new users is very simple, which makes the elimination of spammers profiles an unsuccessful task, because immediately or in parallel, the same person has other user accounts from which to send junk content,” he says. Martínez-Romo.
The language models used by the researchers arestatistical tools that define the structure of the languageand that are very useful in applications with complex syntax or semantics, such as Twitter. Since trending topics deal with specific topics, researchers have built a model with the messages that refer to that topic. “In this way, we can detect new messages that do not correspond to this model and, therefore, are spam,” says Lourdes Araujo, co-author of the study.
To further refine the tool, as some malicious messages appear to be valid, the methodalso analyzes the content of the web pages that are linked in the message. For example, if a popular topic is David Bowie, and a tweet’s website link leads to a pharmaceuticals portal, that message would be classified as spam.
The method can be implemented in real time and in a simple way on Twitter’s servers, something that researchers are open to, in case the North American company wants to include its algorithm.
