Abstract. Despite playing many important roles in society, the news media have been frequently criticised for failing to represent a wide range of viewpoints. Online news systems have the potential to allow readers to add additional information and perspectives. However, due to the simplicity of the filtering mechanisms typically employed, these systems can themselves be prone to over-promoting popular viewpoints at the expense of others. Previous research has attempted to diversify news comments through the use of content similarity, sentiment analysis, named entity recognition, and other factors. In this paper we propose the use of a commonly used community detection algorithm on a network of voting data to identify sentiment groups in news discussion threads, with the eventual goal that these groups may be used to present diverse content. In a controlled experiment with 154 participants, we verify that the Louvain Community Detection algorithm is able to group users with accuracy comparable to an average human. This produces groups containing users who share similar sentiment on a given topic. This is an important step towards ensuring that each group is represented, as by using this method future news systems can ensure that more diverse views are represented in open comment threads.