Nowadays it is so important for almost every medium/big sized e-commerce companies to create personalized experience on their websites to increase the interaction with users. Personalized experience can be increased by different kind of ways. However, recommender systems are commonly utilized because of their very handful and effective implementation. It uses users’ transactional or behavioral data to suggest the products that are likely related to users’ historical data. These recommendations help users but also companies to increase product variety in the basket and make them reach the growth that they aim. For this reason, the application part of this study aims to develop a product recommender system for an e-commerce company. Recommender systems can be developed either Memory based, or Model based. While Memory based approaches mostly work better and give good results with small data, Model based approaches work better with big and sparse data and perform well in production. Since the data has been used is a very big transactional data and the model has been developed will be used in production, we should use an approach which overcomes these both situation in future. Therefore, Model based Collaborative Filtering approach and Alternating Least Squares (ALS) Matrix Factorization method have been applied and the results have been compared to pick the best model. In addition, to be able to apply those two models to our data, we firstly focused on the transition of transactional data to explicit data. In final, RMSE results from both models are observed and the best model is extracted to integrate into the main recommendation engine. Based on RMSE results and the product categories have been recommended for users, we decided to make use of ALS Matrix Factorization method in production.
Anahtar Kelimeler: Recommendation Systems, Model-based Collaborative Filtering, ALS Matrix Factorization Method, Transactional Data, Explicit Data
|