Automated Complaint System Using Text Mining Techniques

Abstract

Complaints System is the system that manages the process of how organizations handle, manage, respond and report to client’s complaints. Manual organizing for large number of requests is extremely difficult, time consuming, error prone, expensive and often not feasible. Results also may differ according to the variety of expert’s judgments. Not forgetting that there would be many questions that already been answered before. For example organization such as UNRWA, receive many complaints each day and make categorization for each request manually based on the contents of the message, forwarding the request to the responsible person according to its category to get the answer. The problem of increasing the cost and efforts required to manage the complaints manually leads to the need to develop automated solutions to handle this problem by including text-mining techniques to substitute the human part. The solution will deal with Arabic content that is different from English which makes data analysis a complex task. Little researches have been conducted on Arabic corpuses mainly because it is highly rich and requires special treatments such as verbs order and morphological analysis. In our work, we propose a new solution to overcome the manual system limitations that consists of three phases. First, we analyze the text message contents, categorize it by using text categorization algorithms and try to decide where to direct the question request automatically to the right person in order to get it answered. Then, we will use text similarity techniques to suggest the answers automatically. Finally, system will use summarization techniques to update the FAQ library with the most asked questions. As a result, the automated complaints system will improve the quality of answering questions by speeding the process and minimizing the required time and effort. We found that the process is efficient and effective. According to results analysis for the classification part, the developed classifier by SVMs achieved the highest average accuracy (74.69%). Also for the answers suggestion part, we obtained best F-Measure (72.45%) at similarity score (0.50). For Summarization part, we obtained the best results at compression rate =0.3, the best F-Measure was 71.56%.