Objective Aiming at the large amount of redundancy generated during the mining process of frequent item sets in datasets with high similarity and frequent occurrence among transactions this paper analyzes the deficiencies of some existing association rule algorithms in mining frequent item sets and proposes an association rule mining algorithm that uses an attention mechanism for pruning. Methods This algorithm combined the advantages of the vertical data format and the attention mechanism. It sequentially calculated the attention weights in frequent k-item sets used these weights to filter out the redundancy in intersection calculations generated simplified frequent item sets and finally merged the simplified frequent item sets with the attention weights to obtain the frequent item sets. Results The experimental results showed that on the Data dataset the proposed algorithm improved time efficiency by up to 94. 6% 73. 0% and 95. 5% compared with Eclat Apriori and FP-Growth respectively it also saved up to 61. 991 MB of space compared with FP-Growth. On the Accident dataset it achieved maximal time efficiency improvements of 93. 4% 69. 0% and 85. 7% over the same three baselines respectively with a maximum space saving of 58. 786 MB against FP-Growth. Conclusion By introducing the attention mechanism the generalization ability and stability of the algorithm are enhanced the redundancy generated during frequent item set mining is reduced the algorithm speed is improved and the memory overhead during the mining process is lowered. Especially for datasets with a large AFI index and low support this algorithm has more advantages in terms of time and space consumption.