Multi-Class Text Supervised Classification on Romanian Financial Banking Reviews

Abstract

Nowadays, with the proliferation of social networks, forum discussions, blogs, and other forms of expression, more and more users have engaged in financial banking communities to gather information about financial banking products or services, to share experiences or to interact with other users. Reviews expressed by users on social media and Internet search activity are a valuable source of information. Text Classification is a growing research area used in social media analysis. The usefulness of Text Classification is manifested in different areas for various needs. For instance, the ability to automatically classify a review into its right financial banking product or service category would be appreciated by users when looking for a specific product or service as well as financial banking institutions to analyze the customers' feedback. Most of the research papers are focusing on exploring Text Classification methods for English-based text. Further, supervised machine learning methods for Text Classification require a huge data volume. In this paper, we focus on implementing supervised machine learning methods for multi-label text classification of Romanian financial banking reviews. We investigate the efficiency of using various supervised machine learning algorithms such as Random Forest, Support Vector Machine, Multinomial Naïve Bayes and Logistic Regression to classify text applied to a limited amount of Romanian financial banking reviews. We analyze the corpus, features and the results data. The baseline features are unigrams and bigrams weighted by TF-IDF. The results illustrate that the SVM classifier outperforms in classification of text into 30 categories of financial banking products and services from Romanian reviews.