search

Blubberli / ConllSharedTask

0 stars 0 forks source link

neural network #1

Open Blubberli opened 7 years ago

Blubberli commented 7 years ago

learn how to use tensorflow which kind of network do we use word vector issues

Blubberli commented 7 years ago

Link for word vectors trained on wikipedia (300 dimensions) https://github.com/facebookresearch/fastText/blob/master/pretrained-vectors.md

Which network do we use?

Blubberli commented 7 years ago

Neue Idee: Zusätzlich zu WordEmbeddings, auch POSembeddings benutzen (und evtl. Labelembeddings)

Blubberli commented 7 years ago

Hier ein Link, wie man POS embeddings erstellen kann. https://radimrehurek.com/gensim/models/word2vec.html

In tensorflow: https://github.com/tensorflow/models/blob/master/swivel/prep.py mit Hilfe vom svivel algorithm

TobiasElssner commented 7 years ago

Hier ist der Link zum Stanford-Paper: http://cs.stanford.edu/people/danqi/papers/emnlp2014.pdf

Blubberli commented 7 years ago

Morphological Embedding: https://pdfs.semanticscholar.org/3e8c/10a3b0158f15c3e47ffee1478e332cc2bd23.pdf

Blubberli commented 7 years ago

skip gramm

Blubberli commented 7 years ago

WORD VECTORS: train morphological, pos embeddings. use the three types of embeddings as features. we can try how the results change if we use only one of them as a feature or if we concatenate them and use them as one feature.

Blubberli commented 7 years ago

which NN do we use? 1) connection weights: feed forward 2) label: recurrent

Blubberli commented 7 years ago

Daniels Folien:

http://www.sfs.uni-tuebingen.de/~ddekok/dl4nlp/#schedule