As research on large language models (LLMs) progresses, LLM-based evaluation has surfaced as an economical and viable substitute for human assessments in the comparison of an expanding array of models. Rather than concentrating exclusively on theoretical evaluation techniques, such as the indicator system, this article advocates for a pragmatic strategy by developing and deploying an LLM evaluation platform that utilizes the big data framework, Hadoop, to manage extensive evaluation datasets and streamline the cleansing of various data sources. Additionally, the proposed system accommodates tailored algorithms and rules, allowing for the export of results to designated databases, which significantly alleviates the burden on data cleaning personnel. Building on the system architecture and theoretical validation outlined in this paper, the author has established a robust LLM evaluation platform grounded in big data infrastructure. The standard data cleaning procedure illustrates that data cleansing can be effectively executed, and user interactions can be optimized in accordance with the theoretical framework proposed herein.