Publication

Abstract : Text classification on product reviews has long been a challenging task due to the rapid growth of Web usage that has resulted in a huge volume of unstructured data. Recently, Opinion mining has been emerged as an important discipline to process the unstructured data. Although several opinion mining approaches addressed the problem of dealing with unstructured data, further research opportunities are available due to the issues like class imbalance, and complexity in text data analytics that affects the performance of opinion learning. Further, the manual text classification consumes a lot of time while identifying useful information. Also, the existing approaches for classifying texts based on majority category are not enough for realistic scenarios specifically in large scale applications. This paper proposes a prediction approach which focuses on obtaining useful information by using keyphrase and category labels. In this paper, we first investigate existing machine learning techniques to classify customer opinions with respect to multiple categories. Moreover, we propose keyphrase based multiclass text classification that finds insights from opinions of various customers on financial products and services. The result of our experiment shows that our dependency-directed opinion learning can show significant improvement over precision, recall, and F1-measure.