You come across a new meme – you send a picture in the application – you get an explanation, a story and similar jokes.
In November, the fourth hackathon from VKontakte took place, in which more than 100 teams participated. One of the applications presented was MemeHunter – it recognizes memes with the help of a neural network and finds their source. spoke with one of the service developers, Dmitry Yermachenkov.
In the team Ermachenkova VLF10 three more people: Andrei Sartison, Alexander Rudykh and Vladimir Glazachev. They performed the task from the Lentach public to create a library of memes. The MemeHunter application should decide what the developers called the “problem of memology”: there are too many memes, and their origin is not always clear. Even more difficult with local memes, whose meaning is known only to a limited circle of people.
In MemeHunter, you can find out the origin and history of an unfamiliar meme by sending them a picture or typing a phrase in the search. Ermachenkov told that the application has several usage scenarios:
- The definition of “protomeme” (source). The application correlates the sent picture or phrase with the base of the “meme families” and offers the most suitable variant with the origin history – “memogenesis”;
- Compilation of collections with different variations of the meme within the same “family”. As an example: I sent a picture with a surprised Pikachu – I received several similar jokes with this template;
- Analytics of “popular” and “hot” memes in social networks. The data is compiled based on manually selected public memes. Among them – “Pictures of category B”, “Dankest Memes”, “Abstract memes for the elite of all varieties” and others.
According to Ermachenkov, the base of origin of memes is still limited to the memepedia library of memes. In the future, the authors plan to use the Know Your Meme English-language library and create an internal platform where users can explain certain memes themselves.
[perfectpullquote align=”full” bordertop=”false” cite=”” link=”” color=”” class=”” size=””]You send pictures with memes to the application, even modified ones, and we classify them. We took the families ourselves from “Memepedia”, pumped out the entire database – about a thousand “Protomemes”. Then they took another neural network and trained on “protomemes”. We also had a script that squeezed out the latest memes from popular public posts on VKontakte.[/perfectpullquote]
On the hackathon, the team presented several functions that may appear in the service in the future. Among them is the native meme constructor for VKontakte.
[perfectpullquote align=”full” bordertop=”false” cite=”” link=”” color=”” class=”” size=””]We thought about how to increase user engagement. Those meme constructors that are external. We decided that it would be cool to combine the constructor and the search for memes. Perhaps there can play a case with TikTok , where the authors of the best memes may eventually become local “stars”.[/perfectpullquote]
The work on MemeHunter is not completed yet – the developers explained this by being busy on the main project. Almost the entire team that came to the hackathon, in everyday life, makes the startup TYLER – they produce a mosaic with any image using an industrial robot. According to Ermachenkov, developers are interested in the development of MemeHunter. “We continue to communicate with Lentach, if we are interested in them, we will cooperate,” he added.