With its rapid users growth, Twitter has become an essential source of information about what events are happening in the world. It is critical to have the ability to derive the topics from Twitter messages (tweets), that is, to determine and characterize the main topics of the Twitter messages (tweets). However, tweets are very short in nature and therefore the frequency of term co-occurrences is very low. The sparsity in the relationship between tweets and terms leads to a poor characterization of the topics when only the content of the tweets is used. In this paper, we exploit the relationships between tweets and propose intLDA, a variant of Latent Dirichlet Allocation (LDA) that goes beyond content and directly incorporates the relationship between tweets. We have conducted experiments on a Twitter dataset and evaluated the performance in terms of both topic coherence and tweet-topic accuracy. Our experiments show that intLDA outperforms methods that do not use relationship information.