Message classification in the call center

Abstract

Customer care in technical domains is increasingly based on e-mail communication, allowing for the reproduction of approved solutions. Identifying the customer's problem is often time-consuming, as the problem space changes if new products are launched. This paper describes a new approach to the classification of e-mail requests based on shallow text processing and machine learning techniques. It is implemented within an assistance system for call center agents that is used in a commercial setting.

Comparing Manual Text Patterns and Machine Learning for Classification of E-Mails for Automatic Answering by a Government Agency

“…The research area of automatic e-mail answering is a rather novel research area, but work has been carried out for example by Busemann et al [1], who constructed an automatic mail answering system for a German call centre. They used 4,777 e-mails that were manually divided into 47 categories with at least 30 e-mails in each.…”

“…A number of machine learning techniques were used to train the system. The best performance was given by SVM (Support Vector Machines); SVM-light obtained 56.2 percent accuracy and a top five accuracy of 78.2 percent [1]. The classification tool described in Busemann et al [1] was included in an e-mail client where categorised messages were assigned a standard answer that could be further edited by a human.…”

Accurate Stemming of Dutch for Text Classification

“…On the basis of her experiments for English text classification, Riloff (1995) concludes that "stemming algorithms may be appropriate for some terms but not for others" and that classification systems would benefit from using all available information, including morphological variants. Busemann et al (2000), on the other hand, have shown that morphological analysis increases performance for a series of classification algorithms applied to German email classification. Spitters (2000) compares, among others, the performance of two machine learning algorithms for topic classification of Dutch newspaper articles, using both unstemmed text and text stemmed with the Dutch Porter stemmer.…”

“…If so, the corresponding answer text can be inserted automatically. An accurate classification system can help improve the efficiency of agents by almost a factor of two (Busemann, Schmeier and Arens 2000). Statistical text classification systems compute the most likely class for a text by computing how likely the words and n-grams in the text are for any given class.…”

Re-ranking Google search returned web documents using document classification scores

“…Two issues that will be addressed in this paper are classification and ranking of search engine returned web documents. Document classification techniques have been applied to many areas such as spam filtering, [7] email routing, [8] and genre classification. [9] Widely used classifiers include knearest neighbours (kNN), [10,11] support vector machine (SVM), [10] and linear discriminant analysis (LDA).…”