Homepage News
Interesting facts

Memes and virals: what will gain popularity on the Internet? An algorithm will tell

31.07.2017 Technology, Interesting facts

Photo: Fotolia

Can you predict whether a material will gain popularity on the web or not? Maybe yes! Researchers from Warsaw University of Technology are working on an algorithm that will distinguish between the materials that have the chance to become a hit and those with less potential for mass audience.

Does the advertising that the marketing agency is working have a chance to catch the public's attention? Which version of the video will appeal to the mass audience? Which picture to choose for the text to increase the "clickability"? These are the dilemmas that businesses and institutions face everyday. Perhaps soon they will not have to rely solely on the subjective assessment of employees in the evaluation of these materials. Because artificial intelligence may help in making decision...


Dr. Tomasz Trzciński from the Faculty of Electronics and Information Technology at the Warsaw University of Technology is working on predicting the popularity of content on the Internet. Information about his research has appeared on the university website.


Researchers from Dr. Trzciński's team are working on a new algorithm. It analyses the content that has gained the highest popularity on the Internet and learns what features these materials have. This will help it assess whether the next material it analyses has a chance to appeal to large audiences.


Researchers use machine learning techniques in their research. Algorithms learn on hundreds of thousands of already published films. "We process data about viewing content on the Internet, especially the social networks" - explained Dr. Trzciński. "We use deep neural networks to do this. We adapt the structures of these networks to get answers to our questions" - he added.


For now, researchers focus mainly on visual and textual aspects. "We take into account the beginning of the video or the elements we see at a glance, such as the title and the teaser. We know that if something catches the attention of the recipient, he or she will stay with the content for a longer time" - said Trzciński.


The algorithm does not show which features the material determine its popularity. "Interpretation of results remains our responsibility" - said Dr. Trzciński. He explained that thanks to the experiments conducted with the use of the algorithm, users are able to identify certain patterns of popularity or anti-popularity.


Their research has already led to certain conclusions "Our algorithms indicate that it is difficult to +sell+ a centred face with a uniform background" said Dr. Trzciński. "Mainly because it looks like the face of a politician or a scene of a party speech, and people generally get tired looking at that. In turn, things such as food, videos showing the preparation of food, attract attention. Warm colours are more effective: red, yellow, brown. Materials with a predominance of colours considered cool (green or blue) are less popular, as are those with a lot of text".


It all started with a collaboration with an American media company on a project carried out with Tooploox, where Dr. Trzciński leads the research team. "The videos that the company publishes on social networks have a combined total of 9 billion views a month" - said the researcher from Warsaw University of Technology. "The company wants to know in advance whether the content will be popular" - he added.


Now that the research of Warsaw University of Technology scientists is well underway, it attracts interest of further entrepreneurs. "We receive inquiries from media houses, marketing agencies, people creating press releases on Facebook, specific brands that want to promote effectively" - said Dr. Trzciński. "We do not want to eliminate or limit the role of a human. We are not saying that we know better than people who make videos or take pictures. We know different things. We give access to knowledge that these people did not have. Now they can use it effectively" - he added.


Dr Trzcinski has involved students from the Faculty of Electronics and Information Technology in the work on the project. This has resulted in several theses, papers and web applications that allow users to upload their video and check whether it has a chance of gaining popularity on the internet.


In the framework of their work on the algorithm, researchers have also co-operated with other teams from Warsaw University of Technology - researchers from the Faculty of Physics participating in the project RENOIR, as well as with scientists from the Faculty of Mathematics and Information Science.


PAP - Science and Scholarship in Poland


lt/ ekr/ kap/


tr. RL

Share this
Rating: 0 votes


Forgotten password


Comments: 0
Add comment See all comments  

Most popular sites