Due　to　the　increasing　amount　of　opinion　data　on　the　internet,　opinion　mining　has　become　a　hot　topic,　in　which　extracting　opinion　targets　is　a　key　step.　The　state-of-the-art　approaches　only　use　direct　dependency　relation　patterns　to　extract　opinion　targets　and　the　indirect　dependency　relation　patterns　have　not　been　used.　In　this　paper,　the　dependency　relations　between　opinion　target　and　opinion　word　are　defined,　and　direct　and　indirect　dependency　relation　patterns　are　designed.　Then,　a　bootstrapping　approach　is　used　to　extract　and　evaluate　both　candidate　patterns　and　opinion　targets.　The　experimental　results　show　that　in　formal　text,　the　approach　improves　the　performance　compared　with　the　state-of-the-art　approaches　for　opinion　target　extraction.　©　Springer　International　Publishing　AG　2018.