WORD EMBEDDING VECTOR INTEGRATION DEVICE, WORD EMBEDDING VECTOR INTEGRATION METHOD, AND WORD EMBEDDING VECTOR INTEGRATION PROGRAM

WIPO Patent Application WO/2020/255912

A1

The present invention enables efficient learning of word embedding vectors for the words in two corpora.　A base vector correspondence determination unit 22 determines a correspondence relationship of a base vector obtained from the word embedded vector of each word generated from a corpus A and a base vector obtained from the word embedded vector of each word generated from a corpus B. On the basis of the determined correspondence relationship a word embedded vector integration unit 24 changes the word embedded vector for each word included in the corpus B so as to sort the word embedded vector elements for each word included in the corpus B.

HIRAO TSUTOMU (JP)
NAGATA MASAAKI (JP)
HAYASHI KATSUHIKO (JP)

PCT/JP2020/023370

December 24, 2020

June 15, 2020

Click for automatic bibliography generation  

NIPPON TELEGRAPH & TELEPHONE (JP)

G06F40/216

ISHIWATARI, SHONOSUKE ET AL.: "Translation of vector representations for wordsusing language resources on the web", PROCEEDINGS OF WEBDB FORUM 2014, 12 November 2014 (2014-11-12), ISSN: 1882-0840
WENPENG YIN; SCHÜTZE HINRICH: "Learning Word Meta- Embeddings", PROCEEDINGS OF THE 54TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS, vol. 1, 7 August 2016 (2016-08-07), pages 351 - 1360, XP055590416, DOI: 10.18653/v1/P16-1128
LUMINOSOS BEAUDOIN: "commonsense/conceptnet5", GITHUB.COM, 6 June 2018 (2018-06-06), pages 1 - 4, XP055773676, Retrieved from the Internet [retrieved on 20200805]

TAIYO, NAKAJIMA & KATO (JP)

View/Download PDF      PDF Help

Previous Patent: SEMICONDUCTOR DEVICE AND OSCILLATION CIRCUIT

Next Patent: STARTING MATERIAL FOR CHEMICAL VAPOR DEPOSITION COMPOSED OF ORGANIC MANGANESE COMPOUND, AND CHEMICAL...