In this assignment, you will develop a text classifier to classify tweets into two classes:
Class '1': If the tweet mentions the tweeter doing a moderate to high physical activity or how they feel about the mentioned activity
Class '0': If the tweet does not refer to the tweeter doing a moderate to high physical activity
For this task, all tweets will contain at least one word / term that refers to a moderate to high physical activity. However, the tweet may use the activity term in non-activity context (e.g. "star wars marathon") or in the context of someone else actually doing the activity or (e.g. the tweeter is watching a basketball game). In such cases, the label should be '0'.
Dataset
On Canvas, you are provided with a training file of 4000 tweets, and a test file with 1000 tweets. You are expected to train a classifier on train data, and predict labels for each tweet in the test file. The training and test datasets are imbalanced, i.e. the proportion of tweets in the positive and negative classes are not equal in both training and test datasets. You should follow best practices, such as splitting the train file into train and validation sets, to evaluate out-of-sample performance.
Acknowledgements
The dataset is derived from a current research project on identifying sedentary and physical activity mentions in tweets.