One of the automated methods for textual data analysis is topic detection. Eigenspace-based fuzzy c-means (EFCM) is a soft clustering-based method for topic detection. Firstly, EFCM uses truncated singular value decomposition to transform high dimensional textual data into low dimensional textual data. Next, the clustering process is conducted in the lower dimensional space. However, that transformation process may eliminate some important features from the textual data. Therefore, the accuracy may be reduced. In this paper, we use kernel trick to overcome that weakness so that the clustering process is performed in a higher dimensional space without explicitly transforming the textual data to space. Our simulations show that this approach improves the accuracies of EFCM in term of topic recall for the problem of sensing trending topic on Twitter.