Most recommendation system focuses on accuracy of recommendation results. However, language styles of suggested contents are rarely discussed for digital reading. This work proposes a recommendation approach where data items are ranked based on their familiarities with user-preferred language styles. Adaptable language models are applied for modeling language styles. Recommended items are rated according to their perplexities of the language models of target users. Experimental results show that dynamic language models can describe continuously changing language styles of users, thus the recommended information satisfies users with their familiar language styles.