Sentiment Analysis on Extracted Entities

Various ops for handling several entities in a document, performs anaphora resolution, clustering, etc..

Requirements

• Python 2.7
• CharLSTM model

Directory Structure

This code assume you have the following directory structure, where CharLSTM is taken from this page.

- Main
-- CharLSTM
-- entity-sentiment-analysis
--- stanford-corenlp-full-2016-10-31

Exemple Usage

First, download the Stanford-CoreNLP server files at this page. And type the following commands in a terminal to start it.

# Start the Stanford-CoreNLP server
cd entity-sentiment-analysis/stanford-corenlp-*
java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServer -port 9000 -timeout 15000

Then, you can use this script to perform anaphora resolution, entity extraction & sentiment analysis on any sentence by modifying parse_doc.py. As an example, here's the result by running this script on the sentence: "Jean is really sad, but Adam is the happiest guy ever".

python entity-sentiment-analysis/parse_doc.py

>>> Processing sentence: Jean is really sad, but Adam is the happiest guy ever
>>> Sentence: 0,Jean is really sad , yielded results (pos/neg): 0.35512/0.64488, prediction: neg
>>> Sentence: 0,Adam is the happiest guy ever , yielded results (pos/neg): 0.97269/0.02731, prediction: pos
>>> Entity:  Jean -- sentiment: -0.2897535
>>> Entity:  Adam -- sentiment: 0.94538456

You can read the full blog post here.