<p>In the context of recommendation systems, addressing the multi-behavior recommendation problem has become increasingly vital for online platforms seeking to comprehend users’ dynamic behavior over time. Recent models employ various techniques, such as graph neural networks and attention mechanisms, to learn users’ behaviors. However, similar to the sequential recommendation, preserving sequential dynamic patterns in multi-behavior recommendation remains crucial. Existing models encounter challenges in learning deep behavioral representations within a given sequence while simultaneously maintaining the temporal order of items. To address this issue, we introduce a novel inter- and intra-attention mechanism for multi-behavior recommendation (IIARec). Specifically, our approach applies intra-self-attention to items of the same behavior, followed by inter-self-attention across all behaviors. Additionally, we propose historical behavior indicators to encode the historical frequency of each item’s behavior in the input sequence. Furthermore, the IIARec model operates in a multitask setting, allowing it to learn item behaviors and their associated scores concurrently. Extensive experimental results on four real-world datasets demonstrate that our proposed model outperforms state-of-the-art methods, illustrating its efficiency in addressing the multi-behavior recommendation challenge.</p>