vOtpuskSam.ru – is an AI powered daily feed of YouTube travel videos on a map, based on NLP and Named Entity Recognition. The project is founded and built by BeWebi, and shows current geo position of travel bloggers, their videos, and allows to watch the videos on the platform.
The main problem of modern society is information overload. Today you can find literally anything on YouTube, especially when it comes to travel – the number of videos is off the scale and it is very difficult to find useful videos on the necessary geographical objects in this mass. Moreover, it is extremely difficult to track the appearance of new videos in given countries.
vOtpuskSam.ru is a unique intellectual system that makes it possible to systematize YouTube travel videos and make it easier for the user to find the right one to watch without spending tremendous amount of time typing countries and cities names in YouTube search.
Videos locations are automatically recognized by their description and then are plotted on a map, which greatly simplifies the search for new videos in many geolocations at once, especially when you need to find exactly relevant and fresh information for a given country/city.
Building a system for recognizing geolocations and allocating YouTube travel videos on a map is not a trivial task, since it requires collecting all the needed travel channels and relevant videos, videos location recognition, its coordinates, and relative interactive mapping tools to place the video on a map. The solution tools are the corresponding NER libraries and API systems for interactive data display on the map, and the process itself is as follows:
- All the relevant travel channels are constantly and automatically searched through YouTube and added to the project.
- As soon as a travel video appears on a YouTube travel channel, it is automatically picked up by the project system via YouTube API and placed in the project database.
- After the video is placed in the data warehouse, a special service is launched to recognize the location of the video. The service is based on the NER algorithm (NLP) of machine learning, which extracts the location directly from the title and description of the video.
- Next, the service associates the extracted city or country with coordinates and places the video on the map (see Fig. 1).
- All of the videos are clustered for each geo location for convenience.
The process is fully automated and does not require human intervention. Thousands of travel videos are processed every day so that visitors can watch new videos in different and unknown places.
The project provides a video tape from which the user can play a specific video from a selected location on the map (see Fig.2). The map can be zoomed in and out to display more detailed video locations collected by the system.
