Presentation 2015-06-23
Corpus and Topic Scalable Topic Model
Soma Yokoi, Issei Sato, Hiroshi Nakagawa,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) It is known that topic model with high dimensional topics improves IR performance like search engines and online advertisements, because it helps to model long-tail words in large scale corpora. However, high dimensional topics with large corpora cause 2 problems: computational performance and memory requirement. For the fundamental topic model, LDA, SGRLD LDA is proposed to scale to large corpora and AliasLDA to accelerate computing topics. In this paper, we propose a method for both topic computation and data scalability, by combining these techniques. Also careful calculation of gradients reduces required space to expectations. Experiments demonstrate that our method is scalable for both corpus size and topic dimension, also achieve faster runtime speed compared to the existing approach, especially 10+ times faster on high dimensional topics setting.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) topic modeling / Langevin MCMC / alias method / scalability
Paper # IBISML2015-5
Date of Issue 2015-06-16 (IBISML)

Conference Information
Committee NC / IPSJ-BIO / IBISML / IPSJ-MPS
Conference Date 2015/6/23(3days)
Place (in Japanese) (See Japanese page)
Place (in English) Okinawa Institute of Science and Technology
Topics (in Japanese) (See Japanese page)
Topics (in English) Machine Learning Approach to Biodata Mining, and General
Chair Toshimichi Saito(Hosei Univ.) / Masakazu Sekijima(東工大) / Takashi Washio(Osaka Univ.) / Hayaru Shouno(電通大)
Vice Chair Shigeo Sato(Tohoku Univ.) / / Kenji Fukumizu(ISM) / Masashi Sugiyama(Tokyo Inst. of Tech.)
Secretary Shigeo Sato(Kyushu Inst. of Tech.) / (Kyoto Sangyo Univ.) / Kenji Fukumizu(京大) / Masashi Sugiyama(お茶の水女子大) / (OIST)
Assistant Hiroyuki Kanbara(Tokyo Inst. of Tech.) / Hisanao Akima(Tohoku Univ.) / / Koji Tsuda(Univ. of Tokyo) / Hisashi Kashima(Kyoto Univ.)

Paper Information
Registration To Technical Committee on Neurocomputing / Special Interest Group on Bioinformatics and Genomics / Technical Committee on Infomation-Based Induction Sciences and Machine Learning / Special Interest Group on Mathematical Modeling and Problem Solving
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Corpus and Topic Scalable Topic Model
Sub Title (in English)
Keyword(1) topic modeling
Keyword(2) Langevin MCMC
Keyword(3) alias method
Keyword(4) scalability
1st Author's Name Soma Yokoi
1st Author's Affiliation The University of Tokyo(UTokyo)
2nd Author's Name Issei Sato
2nd Author's Affiliation The University of Tokyo(UTokyo)
3rd Author's Name Hiroshi Nakagawa
3rd Author's Affiliation The University of Tokyo(UTokyo)
Date 2015-06-23
Paper # IBISML2015-5
Volume (vol) vol.115
Number (no) IBISML-112
Page pp.pp.27-31(IBISML),
#Pages 5
Date of Issue 2015-06-16 (IBISML)