Please enable JavaScript.

Coggle requires JavaScript to display documents.

Detecting Sarcastic Tweets: A SentiStrength Modeling Approach (Information…

- - - - Applied for tweets that do not have hashtag
      - Works based on the strength level of tweet
      - Found that people used exaggeration with sarcasm. Moreover, a previous study in sentiment analysis works on the impact analysis of the lexical and pragmatic features such as emoticons, interjection, and exclamation marks.
      - We applied these two ideas in the proposed classifier (SSC) for identifying sarcastic tweets based on the strength level of the tweets using and used lexical and pragmatic features.
      - Our most important contribution is to find sarcastic tweets based on the strength level of tweet using lexical and pragmatic features
      - Lexical Features
        
        Two types of lexical features named as positive - negative situation phrases [24] and dictionary-based are used in SSC.
        
        The dictionary-based features were derived from i) SentiStrength dictionary [25], which is used to measure the strength of positive and negative sentiment in short informal texts
        
        SentiWordNet 3.0
        
        List of interjections (e.g., ah, oh, yeah);
        
        List of positive verb phrases and negative situation phrases;
        
        Resource of lexical for sentiment classification
        
        Each of the English words has an emotional value in AFFIN. This value was calculated based on the psychological reaction of a person to a specific word
        
        vi) elongated words (stretched words); vii)
        
        capital words; and viii) punctuations (e.g., !, ?)
        
        To find the strength value of the tweet, at first we need to find the strength value for each of the mentioned features. Each of the words in SentiStrength and AFINN has a value between 0, ±5.
        
        Since, the boundary in SentiWordNet dictionary is between 0, ±1. , so, we multiple each of the values in SentiWordNet to 5. Finally, we have a comprehensive dictionary which has a value between 0, ±5. Furthermore, the strength value for each of the positive verb phrases and negative situation phrases is determined based on the comprehensive dictionary
      - Pragmatic Features
        
        Positive emoticons
        
        Negative emoticons
      - SentiStrength Formula
        
        After determining the value for lexicon features, we identify SentiStrength of the tweet based on the number of Interjection, Elongate words, Capital Words and Question Mark.
        
        Formula
        
        Strength (pos)= ∑ Opw (pos)
        
        Strength (Neg)= ∑ Opw (Neg)
        
        Strength= power(Max (Strength (pos), | Strength (Neg) |), 2No.Interjection + No. Exclamation 2No.questionMark 2No.ElongateWord2CapitalWords*2No.Emoticon)
        
        Equation (1) and (2) calculate the sum of the strength of positive and negative opinions words in a tweet.
        
        After determining the strength of positive and negative opinion words, we are going to calculate the final strength value. Equation (3) demonstrates how final strength value is calculated.
        
        Before determining the SentiStrength, all of the slang words are converted to the normal word such as: “gr8” is converted to “great”.
    - - Consists of two classifiers, SHC and HSSC
      - Sarcasm Hashtags Classifier
        
        For these reasons, in this study, sentiment analysis of the hashtags is considered. In order to identify the polarity of the hashtag(s), at first we need to tokenize them. In our proposed classifier (SHC) we used Norvig algorithm [23] which helps us to tokenize hashtags more accurately. After tokenizing hashtags, we identify the polarity of the tweets and hashtag(s).
        
        Among Twitter users, hashtags are a convention for creating and following a thread of discussion. Popular hashtags and words are used in trending topics. Twitter users also use hashtags for expressing their feelings, so most of the hashtags contains sentiment orientation such as, #bad, #sad, and #love which can flip the polarity of the tweets.
        
        First of all, a list of sarcasm hashtags from a corpus of random tweets is extracted. Then the list is extended by automatically collecting pairs of hashtags where one hashtag contained an existing sarcasm hashtag (either #sarcasm or #sarcastic)
        
        We applied systematic approach used by [24]. Extracting SHI contains three main steps as follow:
        
        Rank the candidate based on the probability and frequency.
        
        Compute the number of times the candidate appear in the tweets and
        
        Extract candidate and score them based on estimating the probability
        
        Works based on the Sarcasm Hashtags Indicator (SHI) and contrast between the orientation of the tweets and hashtag(s)
        
        For example, the following tweets which contain #sarcasm and #sarcastic are used for extracting SHI candidate. From the flowing tweets, #NotReally is added into the list of Sarcasm Hashtags Indicator (SHI) candidate.
        
        If one or more of the SHIs are presented in a tweet, it is considered as a sarcastic tweet. The SHI is then used in Sarcasm Hashtags Classifier (SHC) to facilitate sarcasm identification.
        
        Secondly, after extracting the Sarcasm Hashtags Indicator (SHI) candidates, each candidate from the SHI is scored by estimating the probability.
        
        The score of SHI is computed based on [24] approach i.e. the number of times the Sarcasm Hashtags Indicator (SHI) candidate appear in sarcastic tweets divided by the number of times the candidate appears in all tweets. Candidates that have a frequency < 3 in the tweets collection are discarded.
        
        Finally, the candidates are ranked based on this probability, using their frequency as a secondary key in case of ties. The top candidate with a probability >.90 are added to the Sarcasm Hashtags Indicator (SHI) list. In the proposed model (SDM)
        
        The second part of SHC works based on the contrast between the orientation of the tweet and hashtag(s). Hashtags are words or un-spaced phrases that are followed by hash character (or number sign), “#”, to form a label
        
        Several lexical features such as positive verb phrases and negative situation phrases [24], SentiStrength [25], SentiWordNet dictionary [26] and AFINN [28] are used. Moreover, CMU tagger [29] and some linguistic rule such as negation rules are applied for finding the orientation of the tweet and hashtag(s).
        
        After recognizing the orientation of the tweet and hashtag(s), if there is a contrast between the orientation of the tweet and hashtag(s), we consider the tweet as sarcastic. For instance, the following tweet is sarcastic because it contains negative situation phrases (wake up) followed by positive hashtag (greatstart).
    - - People usually use hashtag(s) to emphasize and attract attention; we use these two facts for proposing a new classifier for identifying sarcastic tweets based on the strength level of the tweet and hashtag(s)
      - Works based on the strength levels of tweet and hashtag(s)
      - The features and methodology for HSSC is the same as SSC where, Eq (1, 2 and 3) is applied for tweet and hashtag(s) separately. If the total SentiStrength value of the tweet is more than ≥ 3, we consider the tweet as sarcastic.