This paper addresses the task of extracting opinions from a given document collection. Assuming that an opinion can be represented as a tuple (Subject, Aspect, Evaluation), we propose a computational method to extract such tuples from texts. In this method, the main task is decomposed into (a) the process of extracting Aspect-Evaluation pairs from a given text and (b) the process of judging whether an extracted pair expresses an opinion of the author. We apply machine-learning techniques to both subtasks. We also report on the results of our experiments and discuss future directions.