词向量开山之作2_Distributed Representations of Sentences and Documents.pdf

Manymachinelearningalgorithmsrequirethe inputtoberepresentedasafixed-lengthfeature vector.Whenitcomestotexts,oneofthemost commonfixed-lengthfeaturesisbag-of-words. Despitetheirpopularity,bag-of-wordsfeatures havetwomajorweaknesses:theylosetheorderingofthewords