The performance of automatic keyword extraction methods is evaluated by recall and precision criteria. The result of keyword extraction usually improves when the selected keywords get closer to the ones suggested by a person. Since recall and precision has mutual effect on each other, increase in precision leads to recall decrease and vice versa. In this paper, a post-processing phase based on using attention attractive strings is proposed, which evaluates candidate keywords acquired from other keyword extraction methods to find the ones closest to the human point of view. In other words, an innovative keyword evaluation function, which is inspired by total probability theorem, is employed to improve the precision criterion in a Farsi automatic keyword extraction task. The attention attractive strings are selected by a reverse-engineering process from 800 Farsi keyword-assigned documents. Results indicate that apart from extraction method, improving precision criterion is possible using the proposed post-processing phase without noticeable decrement in recall criterion.