On the unsupervised analysis of domain-specific Chinese texts