抄録
Analysis technology of document data on customer review has been received broad attention because it can be reflected in improving products and services. It is difficult to read all of the text data, and classification by learning model is effective for acquiring knowledge from the voices of customers. In constructing a learning model for large-scale data, it is important to make effective use of data without labels. In this situation semi-supervised learning has gained a lot of success. The purposes of this research are to reduce the cost of labeling document data by human and to improve prediction accuracy of learning models. In this research, we focused on classification task of customer review data with limited labeling. Through extending the random forest to semi supervised learning, we achieved improved classification accuracy of customer review data.
