[DSSM]Learning Deep Structured Semantic Models for Web Search using Clickthrough Data

Architecture

Loss function

Trick: Word Hashing

Given a word (e.g. good), we first add word starting and ending marks to the word(e.g. #good#). Then, we break the word into lettern-grams(e.g. letter trigrams: #go, goo, ood, od#). Finally, the word is represented using a vector of letter n-grams.

paper