Opinion mining is important in learning human behaviors and different personality traits by extracting information from all possible instances such as hidden emotions, unidentified representative, big texts, and even the use of urban language. The fastest growing social media forum such as the Facebook allows all its users to contribute information without borders; sharing opinions about current world issues, and even their attitude and views towards life. Thus, opinion mining is one of most wellknown and important fields of study nowadays. This report analyses different approaches used for opinion mining. It explains the stages of analysis and the tools used for it. This study demonstrates partially-supervised Word Alignment Model for sentiment classification of text reviews. Using this model mining opinion relations between words under constraints can be developed. The partially-supervised word alignment model identifies opinion relations as an alignment process. The proposed model obtains better accuracy because of the use of partial supervision as compared to the supervised or unsupervised alignment model.