Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Selecting TV News Stories and Newswire Articles Related to a Target Article of Newswire using SVM

Yoshimi Suzuki, Fumiyo Fukumoto, Yoshihiro Sekiguchi

Department of Computer Science and Media Engineering Yamanashi University, Takeda, Kofu, Japan

This paper describes a method for selecting TV news stories and newswire articles related to a target article of newswire by using a machine learning technique called SVM (Support Vector Machines) We used selected antecedents of overt pronouns, compound nouns in the experiments. The results of experiments showed that the use of antecedents of overt pronouns and compound nouns for SVM is effective. And SVM is more effective than term weighting methods such as word density, TF*IDF, χ2, a method based on entropy or a method based on standard deviation for event detection.


Full Paper

Bibliographic reference.  Suzuki, Yoshimi / Fukumoto, Fumiyo / Sekiguchi, Yoshihiro (2000): "Selecting TV news stories and newswire articles related to a target article of newswire using SVM", In ICSLP-2000, vol.2, 668-671.