Due to the low-barrier of the social media, the Web has been enriched with the opinions of each person. The naturally expressed public opinions have become valuable resources for businesses analyzing the reputations of their products and identifying new chances. In this context, the microblog service Twitter, which has played an important role in social media, has been the main focus in this research. Named entity recognition and sentiment polarity classification are mainly concerned in this research for extracting semantic knowledge for more in-depth opinion mining from Twitter. Because of the characteristics of the microblog, general techniques of opinion mining have limitations to be directly applied to Twitter analysis. This research suggests the automatic construction of the training corpus for the machine learning methods using distant supervision. By training the classifiers with the recently constructed corpus without human effort, the system can be easily kept up to date for processing the trendy data. DBPedia, which is the core of the Linked Open Data, is used as the knowledge base that may empower semantic opinion mining. To verify the suggested approaches, the experiment was made on the specific target domain, mobile devices, which became recently hot with the proliferation of smart phones and tablet computers. The mobile device market is worth of analyzing in that the pace of technology advance is fast and product life cycle is short. The experiment showed the suggested approaches are promising as a starting point of semantic opinion mining from Twitter.