Aspect Extraction for Opinion Mi

Sandro Cavallari, PhD Candidate, NTU

• Aspect Extraction

  • based on frequent words (occurrences and co-ocurrences of noun and noun phrases)
  • by exploiting opinion and target relations
  • based on topic modelling
  • based on supervised learning

• Deep convolutional belief network

  • composition of the unsupervised RBM
  • trained layer wise to maximise a energy function

• Sentic-net: extract the linguistic pattern as additional information

• Dataset

  • Word Embedding
    • Random
    • Google word2vec embedding
    • Our amazon embedding(~34 reviews)
  • Evaluation copra
    • Aspect-based sentiment analysis [Qui et al.]
    • SemEval 2014
  • Postag
    • Stanford tagger to detect 6 basic part of speech