Automatic multilabel categorization using learning to rank framework for complaint text on Bandung government

Abstract

Learning to rank is a technique in machine learning for ranking problem. This paper aims to investigate this technique to classify the responsible agencies of each complaint text of LAPOR, which is our government complaint management system. Since this categorization problem is multilabel one and the latest work using learning to rank for multilabel classification gave promising result, we work on experiment to compare the typical classification solution with our proposed approaches on this multilabel categorization problem. The experiment results show that LamdaMART, which is listvvise approach in learning to rank, is the best algorithm for classifying the primary agency and the secondary agencies for complaint text.