Comparative Analysis of Recommendation Algorithms in the C# Environment

Abstract

The article is devoted to a comparative analysis of the effectiveness of recommendation system algorithms in the .NET ecosystem. The problem of information overload and the importance of personalization are considered. The implementation of two key approaches is described: classical matrix factorization (MF) using the ML.NET framework and the modern neural network architecture Neural Collaborative Filtering (NCF/NeuMF) based on SciSharp Stack (TorchSharp). To compare the accuracy of predictions and the quality of rankings, a series of experimental studies were conducted on publicly available MovieLens and Amazon Reviews datasets. The RMSE, MAE, and NDCG@K metrics were analyzed. The results show that, on the data studied, the matrix factorization model consistently outperforms the neural network approach in terms of accuracy, robustness to data volume, and speed. The study confirms the conclusions of global analyses that classic, well-tuned models are often more effective than more complex neural network architectures