The problem is to identify the fake news detection, As this is text classification problem . Constraints are basically that we cannot use traditional machine learning and deep learning approaches.
If we move towards Machine learning then we can easily sort out this problem, using Naive Bayes or Logistic Regression etc but we cannot use this. I want to take yours suggestion that using cosine similarity can we done this , Take out the text, apply the feature embedding techniques like Glove or Word2vec or Doc2vec and returns it similarity score. Is it possible , if not kindly suggest your solutions ?