PROJECT TITLE :

Sentence Vector Model Based on Implicit Word Vector Expression - 2018

ABSTRACT:

Word vector and topic model can help retrieve data semantically. However, there still are several problems: 1) antonyms share high similarity when clustered through word vectors; two) vectors for name entities can not be absolutely trained, as name entities could appear limited times in specific corpus; and three) words, sentences, and paragraphs, sharing the identical meaning however with no overlapping words, are arduous to be recognized. To overcome the higher than issues, this Project proposes a replacement vector computation model for text named s2v. Words, sentences, and paragraphs are represented in an exceedingly unified manner in the model. Sentence vectors and paragraph vectors are trained along with word vectors. Based on the unified illustration, word and sentence (with completely different length) retrieval are experimentally studied. The results show that data with similar meaning can be retrieved even if the knowledge is expressed with totally different words.