Enhancing comment Feedback classification using text classifiers with word centrality measures

Autor: Phayung Meesad, Sunantha Sodsee, Watchreewan Jitsakul
Rok vydání: 2017
Předmět:
Zdroj: 2017 2nd International Conference on Information Technology (INCIT).
DOI: 10.1109/incit.2017.8257879
Popis: This paper presents a novelty of item's feedback classification in e-commerce systems. This proposed work is developed based on a combination between a text classifier and word centrality measures. Herein, the item's feedback means comments written by customers to the purchased items, which are classified into positive or negative comments. In this work, the suitable text classifier is selected from four major types of classification: Rule-based, Tree structure-based, Probability-based, and Learning-based, which are Conjunctive Rule, Random Forest, Bayesian Logistic Regression, and Support Vector Machine, respectively. In this work, the classifiers are used for identifying the feedbacks in the probability distribution value [0, 1]. On the other hand, items' feedbacks are also represented by a graph, which is presenting a relationship among words. As well as, centrality measures are applied to determine each contained word centrality, and finalize to a probability centrality in [0, 1]. Both probability distribution and probability centrality, here, are applied to classify the item's feedback to positive or negative comments. The simulation results showed that the proposed classification method was efficient to classify three benchmark datasets, compared to other existing approaches with an average of classification accuracy 80.9%.
Databáze: OpenAIRE