Looking for other opinions on approach to classification problem

Question

I'm looking to implement an "opt-out" filter for my company. The input is short, text-message style messages. A few examples of opt-out messages are:

1. "remove me from your list"
2. "remove from list"
3. "please unsubscribe from list"
4. etc.

All other messages are "good", and should not be removed.

My thoughts on approaches:

I was thinking of using a Bayesian classifier here, but not really knowing the solution space (or having much of a background in ML), want to be sure I'm not wasting time on a sub-optimal solution.

I'm fine with not having the most cutting-edge solution, but want to be sure I'm not missing an approach that might be equally as straightforward but more effective.

Answer 1

You should use text classification techniques. The most basic one is multinomial naive Bayes classifier with tf-idf features. for this method, take a look at this: https://scikit-learn.org/stable/tutorial/text_analytics/working_with_text_data.html

If you don’t get enough accuracy (or maybe precision, recall or f-score), you could test more complex techniques e.g. using deep LSTM networks with word embedding. For this method, take a look at this: https://machinelearningmastery.com/use-word-embedding-layers-deep-learning-keras/