2

I want to calculate score of urgency for a text fragment, Like SentiWordnet provides scores for words for sentiment polarity. I want to tag text as High, Medium or Low on the basis of how urgent the request is. It seems classification will not serve the purpose as it is not something Black or White. I want some algorithm that will give a score for urgency and then i may use threshold for reaching different categories.

I investigated that i can make use of adjective grading. Is there any already built theasure for it, like SentiWordNet for sentiment.

Can i use SentiWordNet's polarity score value for all adverbs and use them to calculate score.

Thanks in advance.

Jyoti Gupta
  • 79
  • 4
  • 1
    It's not black or white, but given you want high, medium, low, it looks like you want black or gray or white :) Why not then a classification task? The classifier would apply the thresholds for you and select the most appropriate category to fit your training data categorized with those 3 labels. – Pascal Soucy Oct 13 '16 at 15:57
  • i understand your point. There is an order between the classes here i.e. Black => Grey => White. Infact, the number of classes may also vary when want more granular level tagging. Will classification at core solve this problem in correct way? If i am not wrong, the confidence value used in classification is about how sure the model is about predicting a label, instead of intensity of the class. – Jyoti Gupta Oct 14 '16 at 03:31
  • 2
    The prediction "score" depends on the algorithm used, it is not always a confidence score. Actually is rarely is, even a Naive Bayes classifier does not provide an estimate of the probability. So you really have two options: either you use a linear regression type of approach (outcome is a continuous value) or you use a logistic regression (categorical outcome: High, Neutral, Low) . I suggest you read about Sentiment Analysis, which is very similar to your problem (the goal often is classifying between Positive, Neutral and Negative) – Pascal Soucy Oct 14 '16 at 12:49
  • Thank you. I am researching ordinal classification. R has already developed library for it. will post once i get finalized strategy for same. – Jyoti Gupta Oct 18 '16 at 10:00

0 Answers0