Most automatic methods for identification of expressions of sentiment and opinions in text depend on the availability of a substantial collection of annotated data that can serve as input to machine learning algorithms. But lack of consensus about how subjective text should be classified and the tediousness of collecting reliable training data pose major impediments to the development of large-scale datasets. A partial solution to this problem may reside in a pre-existing, manually created dataset created by an organization that has monitored major news media for over ten years. Trained professional coders collect data regarding statements of opinion expressed by and about prominent political and business leaders. We analyze the relationship between the text and the information about attitudes stored in the database and we consider the potential usefulness of a larger set of this data for 1) sentiment detection and 2) identification and classification of people who express opinions and of people about whom opinions are expressed.
Nina Wacholder is Assistant Professor at Rutgers School of Communication, Information and Library Science. Her research interests are situated at the intersection of computational linguistics and information science, with a focus on content analysis and on evaluation of the usefulness of the output of natural language processing systems for users of information access systems. She is Co-PI, with Michael Lesk, of the Better Book Viewer Project. Prior to coming to Rutgers, she worked at the IBM's T.J. Watson Research Center and at University¡Çs Center for Research on Information Access.
This talk is part of the CLIP Colloquium Series, organized by Jimmy Lin (jimmylin -at- umd .dot. edu). For the complete schedule, please visit http://www.umiacs.umd.edu/research/CLIP/colloq/.