Social media is a service that is very supportive for government activities, especially in providing openness and community-based government. One form of its implementation is the Semarang City government through the Center for Community Complaints Management (P3M), whose task is to manage community complaints that enter one of the communication channels namely social media twitter. The number of public complaints that enter every day is very varied. This is certainly quite difficult for managers in categorizing complaints reports according to the relevant Local Government Organizations (OPD). This paper focuses on the problem of how to conduct clustering of community complaints. The data source comes from Twitter using the keyword "Laporhendi". Text document data from community complaint tweets was analyzed by text mining methods. A number of pre-processing of text data processing begins with the process of case folding, tokenizing, stemming, stopword removal and word robbering with tf-idf. In conducting cluster mapping, clustering algorithm will be used in dividing the complaint cluster, namely the k-means algorithm. Evaluation of cluster results is done by using purity to determine the accuracy of the results of grouping or clustering.