TriMLP: Revenge of a MLP-like Architecture in Sequential Recommendation