Entropy Guided Extrapolative Decoding to Improve Factuality in Large Language Models