Opinion retrieval through unsupervised topological learning

Opinion Mining is the field of computational study of peopel´s emotional behavior expressed in text. The purpose of this article is to introduce a new framework for emotion (opinion) mining based on topological unsupervised learning and hierarchical clustering. In contrast to supervised learning, the problem of clustering characterization in the context of opinion mining based on unsupervised learning is challenging, because label information is not available or not used to guide the learning algorithm. The algorithm described in this paper provides topological clustering of the opionon issued from the tweets, each cluster being associated to a prototype and a weight vector, reflecting the relevance of the data belonging to each clsuter. The proposed framework requires simple computational techniques and are based on the double local weighting self-organizing map (dlw-SOM) model and Hierarchical Clustering. The proposed framework has been used on a real dataset issued from the tweets collected during the 2012 French election compaign.