Sentence semantic similarity algorithms based on HowNet ignored the fact that different words have different contribution weight to sentence similarity value, and therefore, the similarity result is not quite reasonable. In order to solve this problem, we proposed an improved algorithm based on word frequency. The algorithm calculates the similarity between words based on HowNet, both considering the distance and the height of primitives. Then, a frequency function of words in corpus as a weight factor is embedded into the sentence semantic similarity algorithm, which reduces the proportion value that the high frequency words devote to sentence similarity calculation. The sentence semantic similarity experiment results show that the improved algorithm is much better in rationality as well as in matching with people's subjective judgment.