youtube_comments_sentimental_analysis

Initial sentiment analysis steps

1. Scrape youtube comments form shape of you <<youtubeCommentScraper.py>>
2. Preprocessed the scraped data <<preprocessing.ipynb>>
3. Labelled the sentiments using textBlob package <<Labeling_with_textBlob.ipynb>>
4. Model traing with 68k features <<Model_training>>
5. Observed model performrance <<ConfusionMaterix.ipynb>> <<initial_model.png>>

The model did not perform good for neagtive comments.

Limit feature size

1. Limit the feature size to 25k <<Model_training>>
2. Observed the confusion matrix and found it to be biased with positive values <<25k_features.png>>

The model perforemd worse for the neagtive labels and became mopre bias towards the positive labels

Change the sample size, make occurance of all polarity data equal (i.e., each 859)

1. Changed code in labelling notebook as labelled_data_limit_datapoints.csv
<<Labeling_with_textBlob.ipynb>>
2. Trained the model <<Model_training.ipynb>>
3. Observed model performance <<ConfusionMatrix.ipynb>>

The model performs good for each catagory espicially the negative polarity which was seen privously not as good