The association rule mining (ARM) is an important method to solve personalized recommendation problem in e-commerce. However, when applied in personalized recommendation system in mobile ecommerce(MEC), traditional ARMs are with low mining efficiency and accuracy. To enhance the efficiency in obtaining frequent itemsets and the accuracy of rules mining, this paper proposes an algorithm based on matrix and interestingness, named MIbARM, which only scans the database once, can deletes infrequent items in the mining process to compressing searching space. Finally, experiments among Apriori, CBAR and BitTableFI with two synthetic datasets and 64 different parameter combinations were carried out to verify MIbARM. The results show that the MIbARM succeed to avoid redundant candidate itemsets and significantly reduce the number of redundant rules, and it is efficient and effective for personalized recommendation in MEC.