LING 575 - Sentiment, Subjectivity, and Stance Spring 2014
Final Project Topics

Project types

• Annotation: Perform annotation of a small corpus for some aspect of sentiment or subjectivity. This can involve a novel type of data for a more common phenomenon or an established dataset for a new factor relating to sentiment, subjectivity or stance.
• Analysis: Perform a substantive analysis of some linguistic phenomenon in the context of sentiment/subjectivity/stance. Use an existing corpus to assess the relationship of the phenomenon to expression of subjectivity. (Suggested for those taking the course for linguistics elective credit.)
• Sentiment task: Perform a standard sentiment/subjectivity task. Implement and evaluate an approach. Attempt to improve on a standard method and evaluate the results.
• Existing data, new task: Use existing data in a novel way.
• Application: Employ sentiment, subjectivty, or stance in a downstream application, e.g. subjectivity related question-answering or stance-based summarization.

Project Examples

• Compare different machine learning models for polarity classification of movie reviews.
• Apply a domain adaptation technique to different products in the Amazon product review database. Assess the improvement in different domains.
• Perform subjectivity-based summarization on a dataset like email conversations or blogs.
• Perform automatic sentiment lexicon extraction on two (or more) different corpora/domains. Compare the resulting lexicons to each other and to generic lexicons like MPQA or General Inquirer.
• Implement a system to automatically recognize frustration in human-computer dialogs.
• Build a baseline system to detect stances in debates. Employ richer syntactic or discourse information to improve your results.
• Most systems use low order (1,2) n-grams over part-of-speech tags as part of sentiment classification. Using a POS-tagged corpus, analyze the relationship of higher order n-gram patterns (3,4,..7) to sentiment polarity. (Ling elective example.)
• Using Socher's corpus of fine-grained sentiment over parsed sentences, analyze the effect on overall sentence polarity of different subordinating and coordinating conjunctions. (Ling elective example.)