Paul Alpar, Daniel Ohliger
We analyze user generated content in social media to discover and summarize qualitative information about a commercial bank’s business customers who applied for or received a loan. This information is meant to supplement the quantitative information provided by the customers, e. g., balance sheets, and qualitative information gathered through interviews with the management. We extracted about 74,000 user generated contributions related to three companies from the same industry for seven months and developed their risk profiles. Text mining was used for the creation of risk profiles. Specifically, we apply the expectation-maximization algorithm to automatically identify text clusters. The sentiment expressed in the text clusters reveals whether the observed companies enjoy a positive or negative reputation in social media. In one case, for example, there is appreciation for the company’s products but criticism of the management. We summarize these sentiments in form of risk profiles that can be used to benchmark the reputation of a company against its industry or to adjust the conventionally calculated customer credit score.