In this article, a novel approach to selecting tourist destinations based on textual data is presented, utilizing a proprietary glial network architecture. The model achieved a classification accuracy of 82.41% and a Top-2 accuracy of 85.23% in the task of recommending travel countries. An additional advantage of the proposed solution is the ability to modify and simplify the convolutional neural network (CNN) structure after training, allowing model optimization without loss of prediction quality. Experiments used a real dataset containing customer reviews of hotels in various countries. Input data included reviewer nationality, stay rating, hotel address, and a list of tags describing the purpose and form of travel. After initial filtering (removal of low ratings \(\le 6\) ), the data were appropriately transformed and encoded. The hotel country was extracted from the address and served as the class label. The model was trained on combined textual information such as nationality and travel tags, with country labels numerically encoded. The proposed solution was designed in the context of a international tourism portal, but it can also be applied in intelligent recommendation systems on tourism portals, enabling personalized destination selection based on user profiles and preferences.