<p>Automated Essay Scoring (AES) is a task in which a model automatically assigns a score to a given essay instead of a human grader. Recent AES studies are increasingly utilizing argument structure as a distinctive feature. However, these approaches require high feature extraction costs due to the use of handcrafted features and often consider only limited evaluation traits, making it difficult to provide detailed feedback to writers. To address these limitations, we propose Automated Essay Scoring using Polished Argument Feature Weight (AESPA), a model that uses only argument labels and can assess a wide range of writing traits. AESPA introduces a novel mechanism called trait attention, which allows the model to automatically learn how much the argument structure contributes to each evaluation trait-such as content, organization, or language use-without manually designing features. This enables AESPA to flexibly adapt to traits where argument structure is helpful, and ignore it where it is not. Experiments show that AESPA improves scoring performance across multiple traits and mitigates performance drops in traits where argument structure may otherwise introduce noise.</p>