ABSTRACT

This paper presents an application of a Fuzzy Bayesian based methodology for task analysis (BMTA) to identify the particular tasks performed by experienced call center agents when responding to calls. A total of 126 customer calls were analyzed pertaining to 55 printer models and 70 software and hardware issues. To develop the model, each exchange of information between customer and agent was first classified into one of 72 task categories by a human expert. This data was then used to train the model to predict which subtask was being performed for a particular exchange of information. Model testing results showed an overall hit rate was above 50%, false alarm rate was less than 1%, and sensitivity value d' was above 2.5, supporting the conclusion that Bayesian methods can serve as a practical tool for identifying some of the tasks performed by a call center agent based on what is said during the conversation with the customer. The findings support the conclusion that Bayesian methods are capable of learning how to classify tasks performed in naturalistic settings based on the results of classifications made by a human expert performing task analysis. This result is important, because identifying the subtasks performed within a job is a difficult, time-consuming process. The most basic application of this Bayesian model would also be used to do off-line analysis of human-computer interaction historical data such as email files, blogs, texting, elearning, transcribed calls and voice mails. The flexibility and mobility of BMTA approach goes in line with the current explosive growing number of mobile computing applications and speech recognition devices.