文件名称:
斯坦福CS224 NLP课程-课件lecture02/cs224n-2017-lecture2
开发工具:
文件大小: 2mb
下载次数: 0
上传时间: 2019-03-03
详细说明:斯坦福CS224 NLP课程-课件lecture02 深度学习与NLP专栏地址:https://blog.csdn.net/qq_34243930/column/info/319581. How do we represent the meaning of a word?
Definition meaning(Webster dictionary)
the idea that is represented by a word, phrase, etc
the idea that a person wants to express by using
words, signs etc
the idea that is expressed in a work of writing, art, etc
Commonest linguistic way of thinking of meaning
signifier 6 signified ( idea or thing denotation
How do we have usable meaning in a computer?
Common answer Use a taxonomy like WordNet that has
hypernyms(is-a relationships and synonym sets
from nltk corpus import wordnet as wn
panda wn synset( panda n. 01)
hyper lambda s: s hypernyms()
list(panda closure(hyper))
Chere, for good
[ Synset(procyonid n. 01)
S: adj) full, good
Synset(carnivore. n.01)
S: (adj)estimable, good, honorable, respectable
Synset(placental n.01)
S: (adi) beneficial, good
onset(mammal n. 01),
S: ( adj) good, just, upright
SSss
Synset(vertebrate n. 01),
S: (adj)adept, expert, good, practiced,
onset(chordate n. 01)
proficient, skillful
onset(animal n. 01),
S: (adi) dear, good, near
Synsetcorganism. n. 01)
S: adj) good, right, ripe
Synset(living thing. n. 01)
Synset(whole n02")
S: adv) well, good
Synset(object. n.01)
S: adv) thoroughly soundly, good
Synset('physical entity. n.01)
S: (n) good, goodness
Synset(entity. n. 01 )
S: (n) commodity trade good, good
Problems with this discrete representation
Great as a resource but missing nuances, e. g
synonyms
adept, expert, good, practiced, proficient, skillful?
Missing new words (impossible to keep up to date)
wicked, badass, nifty, crack, ace, wizard, genius, ninja
Subⅰ ective
Requires human labor to create and adapt
Hard to compute accurate word similarity
Problems with this discrete representation
The vast majority of rule-based and statistical nlp work regards
words as atomic symbols: hotel, conference, walK
In vector space terms, this is a vector with one 1 and a lot of zeroes
Dimensionality: 20K (speech)-50K(PTB)-500k (big vocab)-13M(Google 1T)
We call this a representation
It is a localist representation
From symbolic to distributed representations
Its problem, e.g for web search
If user searches for [Dell notebook battery size, we would
like to match documents with" Dell laptop battery capacity
If user searches for [seattle motell, we would like to match
documents containing Seattle hotel
But
moteL [o ooooooooo1000oT
hoteL [oooooo10000000]=0
Our query and document vectors are orthogonal
There is no natural notion of similarity in a set of one-hot vectors
Could deal with similarity separately
instead we explore a direct approach where vectors encode it
Distributional similarity based representations
You can get a lot of value by representing a word by
means of its neighbors
You shall know a word by the company it keeps
(.R.Fith1957:11)
One of the most successful ideas of modern statistical nlp
government debt problems turning into banking crises as has happened in
saying that Europe needs unified banking regulation to replace the hodgepodge
n These words will represent banking i
Word meaning is defined in terms of vectors
We will build a dense vector for each word type, chosen so that
it is good at predicting other words appearing in its context
those other words also being represented by vectors. it all gets a bit recursive
0.286
0.792
-0.177
-0.107
linguistics
0.109
0.542
0.349
0.271
Basic idea of learning neural network word
embeddings
We define a model that aims to predict between a center
word w and context words in terms of word vectors
p(context w,
Which has a loss function e g
J=1-p(W-tlw.
We look at many positions t in a big language corpus
We keep adjusting the vector representations of words
to minimize this loss
(系统自动生成,下载前可以参看下载内容)
下载文件列表
相关说明
- 本站资源为会员上传分享交流与学习,如有侵犯您的权益,请联系我们删除.
- 本站是交换下载平台,提供交流渠道,下载内容来自于网络,除下载问题外,其它问题请自行百度。
- 本站已设置防盗链,请勿用迅雷、QQ旋风等多线程下载软件下载资源,下载后用WinRAR最新版进行解压.
- 如果您发现内容无法下载,请稍后再次尝试;或者到消费记录里找到下载记录反馈给我们.
- 下载后发现下载的内容跟说明不相乎,请到消费记录里找到下载记录反馈给我们,经确认后退回积分.
- 如下载前有疑问,可以通过点击"提供者"的名字,查看对方的联系方式,联系对方咨询.