A bayesian based model application to the prediction of task accuracy – case study on call center’s agents dialog

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review

Abstract

This paper investigates whether an application of a hybrid Bayesian based semi-automated task analysis model is able to learn and predict subtask categories from the narrative telephone conversations between agent and customer. A total of 126 customer calls were collected from a call center and were transcribed into word document. These data were then classified into 72 subtask categories by a human expert and were verified by 22 other persons. The data was used to train a Bayesian based model to predict the subtask categories. Hit rate, false alarm rate, and sensitivity value are used to test if the Bayesian based machine learning tool is able to learn or predict subtask categories. The preliminary results show that the Bayesian based model has overall hit rate of 57.21%, false alarm rate of 0.64%, and sensitivity value d' of 2.67. These results indicate that the model is able to learn subtask categories from the agent/customer narrative telephone conversations and to predict them as well. Further investigations in this study will focus on the comparison of prediction accuracy among thirteen various word combinations, ranging from single word, two-word combination, three-word combination, four-word combination, single-pair combination, single-three combination, single-four combination, pair-three combination, pair-four combination, three-four combination, single-three-four combination, pair-three-four combination, to single-pair-three-four combination. The aim is to examine which combination(s) might fit best for different datasets that include training set, testing set, and a combination of both. The preliminary analyses show that the single-pair-three-four words combination has the highest hit rate of 54.60% compared to other combinations, while the four words combination has the least hit rate of 26.34% among others. A further investigation is needed to check whether the single-pair-three-four word predictors have the highest accuracy to predict the subtask categories.

Original languageEnglish
Title of host publicationAdvances in Physical Ergonomics and Safety
PublisherCRC Press
Pages452-461
Number of pages10
ISBN (Electronic)9781439870594
ISBN (Print)9781439870389
DOIs
Publication statusPublished - 1 Jan 2012
Externally publishedYes

Keywords

  • Bayesian model
  • Call center agent
  • False alarm
  • Hit rate
  • Task analysis

Fingerprint

Dive into the research topics of 'A bayesian based model application to the prediction of task accuracy – case study on call center’s agents dialog'. Together they form a unique fingerprint.

Cite this