| Summary: | 碩士 === 大同大學 === 資訊經營研究所 === 92 === “Telephone Center” has become enterprises’ major service window and important information source for customer relationship management (CRM). Customer service operators are dealing with similar questions for most of time, and by agents could improve their service quality and efficiency. if automatic agents could apply to replace part of their complicate works, they could deal with more specialized problems and reduce manpower to improve process efficiency. Therefore, for agents’ capability to undertake information, this thesis researches the improvement of generally applied vector model’s retrieval efficiency. This research applies word bigram relation model on vector model by three methods, namely, Mutual Information, Association Norm and Conditional Probability, to strengthen words constriction and increase similarity comparison. General customer service questions could compile as enterprise’s FAQ, therefore our experiment object is the FAQ in Chunghwa Telecom’s website, and undertake inside test to tune the best parameter as outside test’s reference, and evaluate retrieval performance by precision rate, recall rate and recall at 11 levels’ corresponding precision rate. The research results is that Mutual Information’s average precision rates under standard recall level increases 41.9% in inside test, and Association Norm increases 8.14% in outside test, so it is indeed could be a segment of customer service agent.
|
|---|
