A Method for Handling Negative Similarities in Explainable Graph Spectral Clustering of Text Documents -- Extended Version