Transparent Neighborhood Approximation for Text Classifier Explanation