Home

Update (Nov 26, 2014): The proceedings of the workshop are online! Check here.
Update (Oct 8, 2014): The workshop is over, some of the slides are already online, check out the workshop program!
Update (Sep 14, 2014): The annotated data for task 2 is now available.
Update: Our introductory paper is now published! JLCL 1, 2014

Overview

Accurate opinion mining requires the exact identification of the sources and targets of opinions. To evaluate diverse tools, the research community relies on the existence of a gold standard corpus covering this need. Since such a corpus is currently not available for German, the Interest Group on German Sentiment Analysis decided to create such a resource and make it available to the research community in the context of a shared task. 

We offer two different tracks, Maintask 1 on mining political speeches, and Maintask 2 on product reviews.

Maintask 1 deals with Sentiment in political debates. In the illustration below you can see an annotated sentence. The term "Jobmotor" (job producer) provides a positive evaluation of the NP "der Ausbau der erneuerbaren Energien" (the construction of renewable energy). Maintask 1s goal is now to detect such relations automatically!




In Maintask 2, the goal is to reproduce information like depicted as follows (but in German!). Training is provided for a set of product classes (like "camera" in this example). For evaluation, a set of annotations from a different domain needs to be produced.

If you are interested in participation, we encourage you to register in our mailing list at https://groups.google.com/forum/#!forum/iggsa-sentiment-analysis-shared-task.


Timeline


   Maintask 1   Maintask 2
 Now             Release of full training data
 May 1st, 2014 Release of trial data    
 July 15th, 2014    Registration deadline  
 August 1st, 2014 Release of evaluation data for main subtask    
 August 10th, 2014 Submission of results for main subtask    
 August 11th, 2014 Release of data for subtasks   Release of evaluation data for all subtasks
 August 21st, 2014 Submission of results for subtasks   Submission of results
 September 1st, 2014    Results to the teams  
 September 22nd, 2014    Working note submissions  
 October 7th, 2014
  Workshop at KONVENS in Hildesheim
 
 October 30th, 2014    Camera ready submissions  

How can I get the data?
Please refer to the respective Maintask pages for more information on this topic!

Do you have any information to get me started?
Yes,  we do! We have a page with resources and tools that you might find relevant.

Will the Working Notes be published?
No. The working notes will have restricted access and only participants of the Maintask will be offered access. Please note, that in case of a missing camera ready submission, working notes might be published instead.

How will the Camera Ready Paper be published?
The Camera Ready papers will be part of the proceedings of Konvens 2014. They will be published only after the workshop.

What is GESTALT?
It is just an abbreviation, however, we like that it has a meaning which is (very) roughly related to sentiment analysis, in the sense of gestalt psychology.

Acknowledgment

We are happy to acknowledge the finanical support that the GSCL (German Society for Computational Linguistics) has granted us for annotating data for Maintask 1.