AIC - Sentiment Analysis

Setup

To start the server a twitter properties file with the credentials for the Twitter API must be provided in the resources directory (src/main/resources/twitter.properties):

consumerKey=YOUR_CONSUMER_KEY
consumerSecret=YOUR_CONSUMER_SECRET
accessToken=YOUR_ACCESS_TOKEN
accessTokenSecret=YOUR_ACCESS_TOKEN_SECRET

For compiling the sources and starting the Tomcat server simply type the command gradlew bootRun on your shell. You'll see some log messages during startup, you know that the server is ready when something like this shows up:

2014-10-17 17:37:25.509  INFO 1151 --- [           main] s.b.c.e.t.TomcatEmbeddedServletContainer : Tomcat started on port(s): 8080/http
2014-10-17 17:37:25.519  INFO 1151 --- [           main] com.aic.Application                      : Started Application in 4.915 seconds (JVM running for 5.496)

When everything went ok, you can reach the server at http://localhost:8080. Currently the server offers two REST endpoints, one for registering a new company and one for querying the sentiments for a company during a specific time period. The registration endpoint can be reached at http://localhost:8080/register and expects name as POST parameter. The sentiment for a company can be retrieved by contacting the endpoint http://localhost:8080/sentiment and takes the four GET parameters name, classifier, startDate, endDate. The latter two parameters represent dates and therefore have to conform to the pattern yyyy-MM-dd. The parameter classifier can be set by one of the followed values: SVM_C_SVC, SVM_NU_SVC, SMO, NAIVE_BAYES, BAYES_NET or KNN. An example query may look like this: http://localhost:8080/sentiment?name=Microsoft&classifier=KNN&start=2014-09-08&end=2014-09-12.

It is also possible to use the REST service via a AngularJS Frontend which can be reached at http://localhost:8080/.

Environment

Programming Language: Java (Tomcat Container)

Framework: Spring

Library: Twitter4j

Architecture: Client-Server

Preprocessing

For preprocessing purposes the Stanford NLP implementation is used. The part-of-speech tagger makes use of the Penn Treebank, the corresponding tagset can be found here. We used the GATE Twitter part-of-speech tagging model for extending the tagging informations especially for twitter data (HT (hashtags), UH ("lol", "ikr"), USR (username mentions), RT (retweet signifiers), URL (URLs)). The Penn Treebank Tagset can be found here.

Classification

For the classification part we used the data mining framework WEKA and LibSVM as a support vector machine.

Team

Twitter API: Florian Taus

Preprocessing: Dominik Pichler, Christian Hotz-Behofsits

Classifier: Matthias Reisinger, Thomas Schmidleithner