As part of the diploma thesis, we created a recommender system for predicting items in an online shop. The data used come from the Instacart database, which is accessible on the Kaggle website. Recommender systems filter a large amount of data and show users only the content that interests them. The recommender system has been built using a neural network. This approach has not been researched much in this field, but it is yielding promising results. The neural network has been built using collaborative filtering. The testing of results is divided into items that the user has never bought and items that the user already bought in the past. For evaluating the new items, the metric HR@10 was used and for evaluating the items the user bought in the past, the metrics precision and recall were used. The obtained results were compared with the results of predicting the most popular items. Our model performed better when predicting new products and worse when predicting products that the user bought in the past.
|
