TY  -  
T1  - Low-rank approximations of second-order document representations 
SN  -  /  
UR  - http://hdl.handle.net/10138/309458 
T3  -  
A1  - Lagus, Jarkko; Sinkkonen, Janne; Klami, Arto 
A2  - Bansal, Mohit; Villavicencio, Aline 
PB  - ACL 
Y1  - 2019 
LA  - eng 
AB  - Document embeddings, created with methods ranging from simple heuristics to statistical and deep models, are widely applicable. Bag-of-vectors models for documents include the mean and quadratic approaches (Torki, 2018). We present evidence that quadratic statistics alone, without the mean information, can offer superior accuracy, fast document comparison, and compact document representations. In matching news articles to their comment threads, low-rank representations of only 3-4 times the size... 
VO  -  
IS  -  
SP  -  
OP  -  
KW  - 113 Computer and information sciences 
N1  -   
PP  -  
ER  -