System to categorize, evaluate and moderate web content for new services and advertising selection (KOMTUR)

The project aims to develop a new technology using machine learning to better classify and moderate content on websites. As a result, the quality of content on websites will be improved and advertising services will become more attractive, which in turn will increase advertising revenues.

Wykop.pl will use the KOMTUR system to assess the credibility, type, and tone of Polish-language content, which will increase the effectiveness of their advertising module. In addition, operators of Polish co-created websites will be able, through the API, to assign published content to a category and rate it.

The results of the project have a triple application. First – in line with the Fast Track funding scheme – the beneficiary is the project leader, Wykop.pl, which, through the use of the solutions developed in the project, will increase its advertising revenues, as well as improve the user experience by improving moderation. Secondly, thanks to the VaaS (Verification as a Service) API offered commercially, other editors, especially – but not only – of portals relying on user-provided content, will gain a new tool to quickly assess various types of web content, obtaining warnings about controversial or drastic content, low credibility, negative overtones, etc. Thirdly, ordinary users of the Polish Internet will benefit from the implementation of the developed solutions, and in two ways: indirectly, through improved moderation and selection of content on Wykop.pl and other media using VaaS, and directly, by making available to the public a tool with functionality similar to VaaS, allowing manual checking of individual pages through a web form.

An important advantage of the project is its deep embedding in the Polish Internet. The classifiers created are trained on data from a Polish portal, thus taking into account the cultural peculiarities of this community, and, above all, analyzing the Polish language in the textual layer. So, while many elements of the declared functionality of the solution can be found on the market for English, in Polish conditions KOMTUR fills an important gap.

Summarizing the approaching end of the research phase of the project, it must be said that we have achieved very interesting results, both in image, video and text classification. In particular, the research on the classification of comments from the wykop.pl website has yielded a number of interesting conclusions, which will soon appear in publications currently in preparation. The models used for classification, after more than a year of development work, have achieved high performance. We are thus looking forward to the impending inexorable implementation, hoping for high usability of the developed solutions. The test implementation of the integrated system will take place later this year, 2023, and early next year the system will be deployed under target conditions, it will begin to benefit wykop.pl, and the VaaS service will become available to users as a commercial API and public web service.