Multi-interest sequential recommendation uses multiple representations to capture diverse interests of users to predict the next item by exploring a user historical interactions. Although existing methods have achieved promising successes, they ignore temporal dependencies and the explicit rating feedback for representing a user preference. In this paper, we focus on how to fully exploit them to capture the personal preference embedding adjustment by extracting temporal semantic scoring paths. Specifically, we propose a novel framework for Temporal semantic Scoring path aware Multi-embedding sequential Recommendation named TSMR, which not only comprehensively explores sequential dependencies and explicit rating feedback to capture adjustment embeddings by extracting temporal semantic scoring item-item and user-user paths, but also can be aware of global presentation of the entire user-item interactions. Our model extracts temporal scoring semantic paths and continuously learns the personal preference adjustment embeddings on objective item or user. That is beneficial to achieve personal user and item adjustment embeddings to better match the user personal preference and item feature. Finally, we use the refined representations of users and items to predict the next item that a user is most likely to interact with. To the best of our knowledge, this is the first attempt to explore personal preference adjustment embedding by extracting temporal scoring semantic paths for better sequential recommendation. Extensive experiments on three benchmark datasets demonstrate that TSMR outperforms several state-of-the-art methods.