Abstract
During the
search for the method to analyze news headlines for Net-tune, I happened to
find the quite interesting software called “CoreNLP” which is a toolkit made by
many members in Stanford University mainly for the core natural language analysis.
(For further information, please refer to this pdf http://nlp.stanford.edu/pubs/StanfordCoreNlp2014.pdf )This Java-based application can evaluate the sentences in many
aspects. One of them is Sentimental Analysis. Sentimental Analysis is a
language evaluation technics that divides the sentences into three categories.
“Positive”, “Negative” and “Neutral”. By using “CoreNLP”, we are able to get the
necessary data to obtain the precise Net-tone figures.
Requirements
As previously
mentioned, “CoreNLP” is a Java-based program. Therefore, “Java SE”, which is a
Java’s system pack for the Java developers, is needed to be installed to launch
this application. In addition, in order to analyze the sentences, which are
written on text files, by using this software, it is also needed to put in some
command lines on command prompt. This experiment was conducted on Windows,
however, the developer team says that it can be run on Mac and Linux as well.
Results for the test
I
analyzed some of English sentences including news headlines from BBC using
“CoreNLP”. There were two types of evaluations. One is a simple judgement appeared
on command prompt. And another one is more detailed data which is output as an
xml file.
(To get simple evaluation result, the command line “java -cp
"*" -mx5g edu.stanford.nlp.sentiment.SentimentPipeline -file input.txt”
is needed to put in. As for the detailed one, “java -cp "*" -Xmx3g
edu.stanford.nlp.pipeline.StanfordCoreNLP -annotators
tokenize,ssplit,parse,pos,sentiment -file input.txt” is used.)
