Research Projects

NLP Lab @ Stony Brook

• Words and Pictures
  
  With Prof. Tamara Berg and Alex Berg, we are investigating how to compose natural language descriptions from images [KPDLCBB11, LKBBC11].
  
  
• Stylometric Analysis (Authorship, Gender, Deception Detection)
  
  Language is a window into people's minds, and stylometric analysis can be used as a tool for understanding the intent of individual writers, as well as societal characteristics reflected in human language. So far we have looked at gender attribution [SGC11], deceptive opinion spam detection [OCCH11], and Wikipedia vandalism detection [HHSJC11].
  
  
• Opinion & Sentiment Analysis
  
  (1) Lexicon Induction and Adaptation for Sentiment Analysis: Lexical resources are one of the key ingredients to many approaches for sentiment analysis. However, as noted by many researchers, word meanings in a specific domain often do not align well with dictionary senses. Our work in [CC09] casts the lexicon adaptation problem as a constraint optimization problem, and converts a general-purpose polarity lexicon into a domain-specific one. We also investigated Connotation Lexicon exploiting the selectional preference of Connotative Predicates [FBC11].
  
  (2) Analysis in light of Compositional Semantics: A lot of problems in natural language understanding have compositional nature, which motivates the need to incorporate theories from compositional semantics into statistical models. As two gentle attempts, We explored compositional inference rules for sentiment analysis [CC08], and the use of compositional semantic vectors for extractive summarization to improve sponsored search [CFGJMP10].
  
  (3) Semantic Negators: Semantic negators (e.g., "prevent", "fail") negate the polarity of their arguments (e.g., "prevent cancers"), constituting one of the key challenges in fine-grained opinion analysis. We have investigated the discovery of semantic negators [CC09], and the polarity inference rules dealing with negators [CC08].