Enhancing matrix factorization-based recommendation systems: A comparative study on trip recommendation systems

As the impact and usefulness of recommendation systems continue to grow, their importance in various applications is becoming more and more pronounced. Thus, it is crucial to design and implement recommendation systems that are both efficient and highly accurate to meet the increasing demands and expectations. This study focuses on a model awarded first place in a travel forecasting recommendation system competition. The objective is to investigate whether improvements, such as the correlation of the model used with the dataset and the impact of the reduced dataset applied on the success rate, affect the model's performance. Changes have been made to the dataset to reduce resource utilization. Unusually, the dataset has been reduced to one-fifth using random and selective reduction methods, and the results have been observed. While random reduction of the dataset has led to a decrease in the success rate, methodical reduction of the dataset has significantly increased the success rate. Instead of the deep learning algorithms used by the original model, GRU, another algorithm using the same principles, LSTM, is utilized. The effects of Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) on the data set are also investigated. For these datasets, the GRU algorithm has produced more accurate results than LSTM. New models have been developed on the embedding layers, and the results have been observed. Furthermore, the optimizer used by the model is changed and the performance of other optimizers is evaluated. Optimizers have shown a wide range of effects on hardware. The original model achieved a success rate of 0.5664, while the experiments conducted in this study reached the highest and most reliable success rate of 0.6654. We argue that changes applied to models, optimizers, and feature engineering can be useful for maintaining effective recommender systems.