Collaborative　Filtering　(CF)　is　an　important　building　block　of　recommendation　systems.　Alternating　Least　Squares　(ALS)　is　the　most　popular　algorithm　used　in　CF　models　to　calculate　the　latent　factor　matrix　factorization.　Parallel　ALS　on　Hadoop　is　widely　used　in　the　era　of　big　data.　However,　existing　work　on　the　computational　efficiency　of　parallel　ALS　on　Hadoop　have　two　defects.　One　is　the　imbalance　of　data　distribution,　the　other　is　lacking　the　fine-grained　parallel　processing　on　the　rating　data.　Aiming　on　these　issues,　we　propose　an　integrated　optimized　solution.　The　solution　first　optimizes　the　rating　data　partition　with　the　consideration　of　both　the　number　of　involved　data　records　and　the　partitioned　data　size.　Then,　the　multithread-based　fine-grained　parallelism　is　introduced　to　process　rating　data　records　within　a　map　task　concurrently.　Experimental　results　demonstrate　that　our　solution　can　reduce　the　overall　runtime　of　Hadoop　ALS　by　82.17%　by　maximum.　©　Springer　Nature　Switzerland　AG　2020.