Presentation | 2007/7/17 A dialogue segmentation method via uterance based HMM Yasuhiro TAJIMA, Daizo KITADE, Michiko NAKANO, Koji FUJIMOTO, Tomo NAKABAYASHI, Yoshiyuki KOTANI, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We propose a dialogue segmentation and topic structure finding method via Hidden Markov Model (HMM). HMM has been applied for this problem in previous studies and its advantages have been shown. Nevertheless, the length of the dialogue must be restricted about a hundred words because of computational errors, i.e. the occurrence probability of a dialogue which has a thousand words tends to be less than 10^<-1000> and we fail to construct HMM because of lack of computational precision. In this paper, we propose a new approach for this problem by HMM whose state outputs a symbol of an utterance. Every utterance is classified into some symbols of a segment by a Bayesian classifying method, then we construct an HMM for the target dialogue. The HMM in our method can handle a long dialogue whose length is about 1500 words for 1000 kinds of words. We used 62 dialogues by 68 testee and evaluate our method. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | dialogue structure / text segmentation / Hidden Markov Model / naive Bayes |
Paper # | NLC2007-2 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2007/7/17(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Natural Language Understanding and Models of Communication (NLC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | A dialogue segmentation method via uterance based HMM |
Sub Title (in English) | |
Keyword(1) | dialogue structure |
Keyword(2) | text segmentation |
Keyword(3) | Hidden Markov Model |
Keyword(4) | naive Bayes |
1st Author's Name | Yasuhiro TAJIMA |
1st Author's Affiliation | Tokyo University of Agriculture & Technology() |
2nd Author's Name | Daizo KITADE |
2nd Author's Affiliation | Transcosmos Inc. |
3rd Author's Name | Michiko NAKANO |
3rd Author's Affiliation | Transcosmos Inc. |
4th Author's Name | Koji FUJIMOTO |
4th Author's Affiliation | NIWS-FEG Inc. |
5th Author's Name | Tomo NAKABAYASHI |
5th Author's Affiliation | Tensor-Consulting Co. Ltd. |
6th Author's Name | Yoshiyuki KOTANI |
6th Author's Affiliation | Tokyo University of Agriculture & Technology |
Date | 2007/7/17 |
Paper # | NLC2007-2 |
Volume (vol) | vol.107 |
Number (no) | 158 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |