Predicting the Cost of Goods on Marketplaces Based on Description

In this paper, a machine learning model is built to predict prices based on text descriptions. The data for training the model was collected from one of the well–known marketplaces, limited to one category - electronics, in which, in turn, there are many subcategories. This limitation allowed us to work with a relatively small amount of data at the initial stage. Different methods were used for text processing: TF-IDF, Bag-of-Words, Word2Vec and BERT. The resulting numerical representations of the texts were then used as features to solve the regression problem of price prediction. In addition, the subcategory of the product was used as a feature. The following machine learning models were used: linear regression, random forest and various boosts. As a result, the optimal method of processing text descriptions was selected, which, in combination with the selected model, gives the best metrics, making minimal mistakes in predicting prices based on the text description.