DSSM
- introduction
- two line for extending latent semantic models
- clickthrought data
- deep atuo-encoders
- text hashing. This hashing is not for security, while word embedding or sentence embedding.
- translation model for IR, directely translate doc into query. suboptimal
- two line for extending latent semantic models
- Question
- target is for re-construction of document term vectors, rather than differentiating the relevant docs from irrelevant ones for a given query.
- small vocabulary size, only 2000 frequent words
- loss
- loss funcion
is a smoothing factor in the softmax fomulatation, it is for the held-out data set.
MIX
Question:
- after uni-conv, bi-conv and tir-conv, the sentence is transformed into a one-demension embedding?
- are there 9*3 feature maps in weight channels layer?