Abstract: This study proposes an approach to Automatic Keyphrase Extraction depending on the semantic coherence between the phrases in a document. The approach uses a probabilistic index added to existing measures to predict the occurrence of keyphrases before the processing. The approach is suitable for both single and multiple word phrases and uses the measures of semantic coherence and baseline metrics to calculate the relative weights for the phrases as well as using a probabilistic measure introduced to improve performance. This probabilistic measure is intended to reduce the complexity of identifying a keyphrase considering its likeliness to occur in an index position.