Modeling Risk Cluster based on Sentiment Analysis in Bahasa Indonesia for SME Business Risk Analysis Documents
Kusuma, Wisnu A
MetadataShow full item record
Currently, there are two risk analysis models that commonly used for business financing in banking industry, namely, the quantitative model and the qualitative model. The quantitative model are mostly implemented as an credit scoring system that consists of several accounting formulation that calculates the financial statement and business performance to determine the feasibility of bank customers in accepting loan. The second model, namely qualitative model, emphasizes the risk analysis, opinion, and mitigation from risk analyst to support the decision makers in accepting loan proposal. From the observation through the Standard Operating Procedure (SOP) the quantitative model has some drawbacks in measuring the acceptance criteria, since the data are originated from the customer itself and vulnerable to have a manipulation, especially when the financial statement has no any inspection from external auditor. The second drawback is that the quantitative model tend to be subjective since the credit scoring calculation are performed by the marketing staff that stand sides to the customer. Hence, the qualitative model are deployed to overcome these drawbacks, where the analysis is objectively proceed by risk analysts. However, the implementation of qualitative model are not a hundred percent perfect, since the qualitative model neither has decision criteria nor risk measurement. Another issue is that the risk analysis documents that consist of risk opinion and mitigation from previous analysis, are not well managed. Actually, these documents are useful for the risk analyst to evaluate and relearn from the previous analysis. In this research, the opinion or sentiment analysis against the risk analysis documents is conducted by modeling the risk cluster to help the risk analyst in refine the analysis. There are three tasks that have been conducted, those are clustering the risk analysis documents based on the term presence. Secondly is quantify the risk level within each cluster by measuring the term importance and sentiment score using TF-IDF and SentiWordNet 3.0 respectively. The task is eventually finished by evaluating the cluster quality using Silhouette function and examining the most frequent terms by its importance and sentiment. We also develop a prototype that enables risk analysts to retrieve the risk analysis documents by entering query terms and presents the level of risk from each document. The results has been shown that sentiment mining technique is effective and could be utilized to model risk cluster. This could be seen in how relevant is the cluster model with the 5Cs Credit criteria that commonly used in banking industry. In discussion, there are also some suggestions for the management on what criteria they should resharpen in conducting the qualitative model. By giving the risk score in each cluster, it is expected that the model also could be used as a benchmark to qualify the submitted loan proposal and assist the decision maker to have a better decision making.