The review is a form of text in which a consumer evaluates and gives feedback on a product. Opinion summarization is the process of extracting and compressing reviews to form a short text that summarizes the core information of the reviews. Aspect-level opinion summarization focuses on identifying sentences relevant to a specific aspect to generate a summary related to that aspect. Currently, aspect-level opinion summarization methods primarily focus on sentence-level information extraction, but often disregard sentences that briefly mention a specific aspect, resulting in summaries that fail to adequately reflect the evaluative feedback on that aspect. Compared with sentence information, opinion phrase information exhibits finer-grained properties and more accurately captures information related to a specific aspect. Therefore, this paper proposes an aspect-level opinion summarization method that integrates an opinion phrase masking task (MTAOS), effectively addressing the information omission problem in traditional methods by deeply mining fine-grained information from reviews. The experimental results show that on both the SPACE dataset and the OPOSUM+ dataset, the method significantly improves the ROUGE metrics compared to the baselines, generating more comprehensive and accurate aspect-level opinion summary.