Semi-supervised LDA by labelling words

We propose a new semi-supervised learning technique, which is called Words labelled Semi-Supervised Latent Dirichlet Allocation (wssLDA) by labelling words for large text collections analysis. The model incorporates supervision with Latent Dirichlet Allocation by adjusting weights of topic words chosen by users. Results with perplexity for documents and F-measure for clustering show the improvements for the topic learning and document analysis tasks.