You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Here are the things I did today:
-Coded 450 more tweets
-Added all new articles to google drive and started updating their descriptions. I still have a handful more to read through.
-I haven't been able to find any corpora that defines whether or not a text contains emotion so I tried classifying from the tweets I have hand coded (only 105 are relevant so this is a really small test case) Initially I'm getting poor results, but it kind of makes sense. In the training data, only 275/989 tweets are labeled with emotion, and my classifier labels most of the data as not containing emotion. However, in the new data set I marked 62/105 with emotion. (the code is posted in github under src/EmotionDetection.ipynb and is tied to my github issue)
From here on the emotion detection task I will play around with removing stop words and some other parameters to see if I can get the accuracy up. I'm also going to look into some alternative ways to test the classifier. One of the papers I read used tweets with emojis to reflect ones as subjective and tweets from popular newspapers such as NYT to reflect objective tweets. I might try to gather similar tweets so I have a more robust corpora for testing my emotion detector.
Activity
syd-shelby commentedon Feb 19, 2019
Find the best way to filter tweets into emotion/no-emotion categories. Apply sentiment analysis on emotion tweets and see results.
audrism commentedon Feb 21, 2019
Older datasets have only relevant tweets
newer tweets mark 0 when irrelevant
model d2v has 87%
#6 progress from last week
#6 Today's updates: verifying on new data
syd-shelby commentedon Feb 26, 2019
Here are the things I did today:
-Coded 450 more tweets
-Added all new articles to google drive and started updating their descriptions. I still have a handful more to read through.
-I haven't been able to find any corpora that defines whether or not a text contains emotion so I tried classifying from the tweets I have hand coded (only 105 are relevant so this is a really small test case) Initially I'm getting poor results, but it kind of makes sense. In the training data, only 275/989 tweets are labeled with emotion, and my classifier labels most of the data as not containing emotion. However, in the new data set I marked 62/105 with emotion. (the code is posted in github under src/EmotionDetection.ipynb and is tied to my github issue)
From here on the emotion detection task I will play around with removing stop words and some other parameters to see if I can get the accuracy up. I'm also going to look into some alternative ways to test the classifier. One of the papers I read used tweets with emojis to reflect ones as subjective and tweets from popular newspapers such as NYT to reflect objective tweets. I might try to gather similar tweets so I have a more robust corpora for testing my emotion detector.
#6 updating emotion detection and testing against news articles and f…
#6 more or less final emotion detection