Skip to main content

Table 7 Examples of misclassifications

From: Deep learning for pollen allergy surveillance from twitter in Australia

No Example Prediction Prob. Actual
P 1 With this hay fever making me feel like I don’t have enough air in my lungs I hope this helps;-) 2 Informative - generic 0.509 1 Informative - symptom
  Explanation: Rare expression of the symptom, referring to breathing difficulty, and negation ocurrence. Still, the low predictive probability obtained.    
P 2 How can I be awake with hay fever at 2am?! 2 Informative - generic 0.543 1 Informative - symptom
  Explanation: Short length and expressiveness indicators (’?!’) typical for generic class 2. Still, the low predictive probability obtained.    
P 3 @username Same!!! Grass gives me bad rashes and hay fever as well as pollen and it sucksss 2 Informative - generic 0.993 1 Informative - symptom
  Explanation: Repeated characters for added emphasis and expressiveness (’!!!’) typical for generic class 2. New and rare symptom occurred.    
P 4 Dubliner Goran Pavlovic claims started stinging himself with nettles and his hay fever symptoms started to disappear, if suffered would give it a try 1 Informative - treatment 0.527 3 Non-Informative - news
  Explanation: Identified as treatment. The news content is paraphrased by the user, thus classified as personal. Still, the low predictive probability obtained.    
P 5 Could be a dangerous day tomorrow for asthmatic and hay fever sufferers. If we were in summer and not winter we would be looking at a severe fire danger. Those are warm winds. 2 Informative - generic 0.863 3 Non-Informative - warning
  Explanation: Warning content paraphrased by the user - no typical ’warning’/’be careful’ phrase identified.