Please enable JavaScript.

Coggle requires JavaScript to display documents.

Text Sentiment Analysis (Emotion Mining (Emotion models (Discrete:…

- - - - Problem Levels
        
        Sentence Level
        
        Aspect Level: opinion polarities about a specific aspect of a product or service
        e.g. Restuarant reviews have two aspects: food and cleanliness
        
        Extraction of aspects from text
        
        polarity classification of the extracted aspects
        
        Document Level
      - Feature extractions
        
        Annotated Data
        
        Presence-Based and Frequency-Based Features: one-hot vector; Frequency is popular in information retrieval
        
        Unigram and N-gram Features:
        A group of adjacent words as features
        Preserve the order
        
        Part of Speech:
        Adjectives are important when determining the sentiment, thus extracted as features
        
        Syntax: use the dependency-based rules to select features
        e.g. adj. or adv. around the keywords
        
        Negation: add the negative word (e.g. not) to flip the polarity
        
        Topic-oriented features: the same word has different polarity towards different topics
        e.g. "fast" in car review (positive) againist "fast" in movie review (negative)
        
        Unannotated Data
        
        lexical or dictionary-based method
        corpus-based lexicon expansion: automatic create the lexicon words and assign the degree of polarity by machine
        e.g. the words near the "excellent", "but" and "and" between pairs of words
        
        Dictionary-based lexicon expansion: assign the polarity ofnew words with respect to their relationship to the seed words
        
        Domain Adaptation: training a classifier over a labeled dataset from one domain or topic, called the “source,” and use it to label the unlabeled data from another domain
        
        Label adaptation: the polarity of some features are different between source and target domain
        
        Instance adaptation: some word frequency is remodelled from source to target domain
        
        Label transferring: find the most informative samples of the target domain by means of a classifier that is trained over the source domain and then label those informative instances to train a brand new classifier over them.
        
        Bootstrapping: use an initial pre-trained classifier on another dataset to label the target dataset and then use this newly labeled dataset to train a new classifier.
        
        Belief network modeling: Lin and He [2009] add a layer of sentiment to the structure of a famous probabilistic document model called “Latent Dirichlet Allocation” (LDA) [Blei et al. 2003] to find the polarity of words inside a set of documents with respect to each topic.
        
        Combining lexical and machine-learning methods: Lexical and learning methods can be combined to compensate for the disadvantages and drawbacks of each other.
      - Publicly available lexicons and dataset for opinion mining
        
        Lexicons
        
        Harvard General Inquirer:
        11,790 words
        positive, neutral,or negative
        
        Opinion Lexicon:
        6786 words
        positive, negative
        
        Multi-perspective Question Answering (MPQA):
        8222 words
        how subjective the word is
        how strong its subjectivity is
        the prior polarity of the word, which can be positive, negative, both, or neutral
        whether the word is stemmed
        
        WPARD:
        1402 words
        from -6 (very negative feeling) to +6 (very positive feeling), with 0 being a neutral feeling
        
        SentiWordNet 3.0
        
        NRC
        14,182 unigrams (words) => sentiments(negative, positive): 0 (not associated) or 1 (associated)
        ~25,000 senses => emotions (anger, anticipation, disgust, fear, joy, sadness, surprise, trust): not associated, weakly, moderately, or strongly associated
        
        Dataset
        
        Amazon
        2000 reviews for items sold on Amazon
        Rated from 0 to 5 stars
        Half pos (4~5) Half Neg (0~2), discard 3 stars
        
        IMDB Movie Datasets
        
        Blogs:
        Size: 252
        Technology blogs: positive, negative, neutral,or irrelevant
        political blogs: positive or negative
- - - - Ekman: anger, disgust, fear, joy, sadness, surprise (Most popular in CS field)
      - Shaver: anger, fear, joy, love, sadness, surprise
    - - Plutchik: anger, anticipation, disgust, fear, joy, sadness, surprise, trust
      - Lovheim: anger, disgust, distress, fear, joy, interest, shame, surprise
  - - - WordNet Affect:
        4787 words
        a hierarchy of emotions
      - LIWC:
        5000 words
        affective or not, positive, negative, anxiety, anger, sadness
      - NRC:
        14182 words
        anger, fear, anticipation, trust, surprise, sadness, joy, disgust
      - NRC hashtag:
        32400 words
        anger, fear, anticipation, trust, surprise, sadness, joy, disgust
      - clean balanced emotional tweets(CBET):
        24000 words
        anger, fear, joy, love, sadness, surprise, thankfulness, disgust, guilt
      - Chinese Lexicon:
        ...
    - - International Survey On Emotion Antecedents And Reactions (ISEAR):
        7666 crowd written paragraphs
        joy, fear, anger, sadness, disgust, shame,and guilt
      - fairy tales:
        15000 sentences from children's stories
        anger, disgust, fear, happiness, sadness, positively surprised, negatively surprised,or neutral
      - SemEval 2007:
        1250 news headlines
        anger, disgust, fear, joy, sadness,and surprise
      - TEC:
        21051 tweets
        anger, disgust, fear, joy, sadness,and surprise
      - CBET:
        81163 tweets
        anger, fear, joy, love, sadness, surprise, thankfulness, disgust,and guilt