Extraction and clustering of arguing expressions in contentious text

摘要

This work proposes an unsupervised method intended to enhance the quality of opinion mining in contentious text. It presents a Joint Topic Viewpoint (JTV) probabilistic model to analyze the underlying divergent arguing expressions that may be present in a collection of contentious documents. The conceived JTV has the potential of automatically carrying the tasks of extracting associated terms denoting an arguing expression, according to the hidden topics it discusses and the embedded viewpoint it voices. Furthermore, JTV's structure enables the unsupervised grouping of obtained arguing expressions according to their viewpoints, using a proposed constrained clustering algorithm which is an adapted version of the constrained k-means clustering (COP-KMEANS). Experiments are conducted on three types of contentious documents (polls, online debates and editorials), through six different contentious data sets. Quantitative evaluations of the topic modeling output, as well as the constrained clustering results show the effectiveness of the proposed method to fit the data and generate distinctive patterns of arguing expressions. Moreover, it empirically demonstrates a better clustering of arguing expressions over state-of-the art and baseline methods. The qualitative analysis highlights the coherence of clustered arguing expressions of the same viewpoint and the divergence of opposing ones.