With the dramatic growth of web's popularity, the number of freely available on-line news, reviews, blogs has rapidly increased. This project focuses on mining opinions and their sentiment polarity from these valuable on-line resources. We explore multilevel sentiment polarity analysis method to better capture finer-grained polarity of opinions. Meanwhile, in order to reduce the efforts of opinion mining, we further explore an effective domain adaptation method to overcome the data distribution difference among various domains and resources. This project may provide an effective opinion-oriented information-seeking platform for government intelligence and business intelligence. Based on the multi-aspect sentiment polarity analysis and summarization, governments could better monitor hostile or negative news reports to further enhance their service. Companies also may better track public viewpoints, perform reputation management and trend prediction in sales or other relevant data.