Sentiment Analysis

AG
by apala guha0 views·Updated Jan 16, 2017
Powered by AI

Need to present on this topic?

Create your own polished presentation in minutes. Just describe what you need.

Free to try, no credit card required

Transcript

Sentiment Analysis

Building the Pipeline

Linear

Regression

docs =>

word count vectors

words =>

hash indices

- count the number of occurrences of each possible string in each doc => term frequency (TF)

- what are the problems with this approach?

- each word is a string

- hash each string to a number

- what are the potential problems?

- large vocabularies lead to large vectors => store as sparse vectors

- mapping word to a vector position is inefficient => use hashing => collision potential

- frequent words with less meaning dominate e.g. as, there etc. => scale down count with a measure of ubiquity => inverse doc frequency (IDF)

- semntically equivalent words are not grouped together

Predict Sentiment

Tuning

Text preprocessing

docs => average word vectors

what other preprocessing could be useful?

Words => vectors

Logistic Regression

- word semantics taken into account

- dense word representation (word2vec)

word vectors => clusters,

docs => average cluster vectors