<p>The processing of large-scale datasets is complex and requires high efficiency. The database needs to be scanned multiple times by traditional Apriori algorithms to generate candidate itemsets, resulting in significantly reduced efficiency, but also have too simple a support threshold setting, which fails to fully consider the importance and differences of different itemsets in practical business, thereby affecting the accuracy of frequent item mining. Therefore, a multi-stage frequent item mining method for accessing data based on association rules is proposed. By applying the concepts of clustering analysis and Boolean vector operation, the Apriori algorithm for association rules is improved. The matrix is compressed through clustering and counting of the same transactions to reduce the size of multi-stage database access. Only one database scan is required, and frequent itemsets are directly generated without generating candidate itemsets, greatly improving the efficiency of the algorithm; Introducing weights into the improved Apriori algorithm and designing a weighted support function, by assigning different weights to different itemsets, more accurately reflects the importance and differences of itemsets in practical applications, thereby improving the accuracy of frequent item mining. Experiments have shown that this method can effectively mine frequent items in multi-stage access data, and the Pearson coefficient of frequent items mined by this method is close to 1, that is, the mining accuracy of frequent items is high; at different minimum weighted supports, this method has multi-stage comprehensive evaluation function value of mining frequent items in access data is higher, that is, the mining effect of frequent items is better.</p>