A Session-Based Recommendation Approach with Word Embeddings
Ahmet Tuğrul Bayrak · 2022
Recommender systems have become prominent in past years. Today, many service providers use recommendation systems. Considering the number of products and customers, it is crucially important and necessary to meet the right products to the right customers. Various methods have been applied for recommendation systems. In our study, it is aimed to examine the purchasing relationship between the products and recommend alternative products. To achieve this, product spaces have been created by applying Word2Vec and FastText models, which extract the relationship between words according to the frequency of occurrence of items, on the products in the same basket (Item2Vec). In addition, pre-trained models are applied to product names to measure the similarity and they are ensembled with Item2Vec models with different weights. The results obtained regarding the use of word embedding methods in product recommendation are reassuring and may be applied in following projects.