EmotiKLUE is a submission to the WASSA 2018 Implicit Emotion Shared Task. The aim of the shared task is to predict one of six emotions (anger, disgust, fear, joy, sadness, surprise) from the context of masked emotion words as in the following example:

My dad was [#TRIGGERWORD#] that I washed his car so he gave me money to buy snacks 😢

EmotiKLUE tackles this task by learning independent representations of the left and right contexts of the masked emotion word and by combining those representations with an LDA topic model.

The system is described and evaluated in greater detail in Proisl et al. (2018).

For information on how to train, retrain or test a model or on how to use it for prediction, see the help messages of the corresponding subcommands:

emotiklue.py {train,retrain,test,predict} -h

• Proisl, Thomas, Philipp Heinrich, Besim Kabashi, and Stefan Evert. 2018. “EmotiKLUE at IEST 2018: Topic-Informed Classification of Implicit Emotions.” In Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis (WASSA), 235–242. Brussels: Association for Computational Linguistics. PDF.

  @InProceedings{Proisl_et_al_IEST:2018,
    author    = {Proisl, Thomas and Heinrich, Philipp and Kabashi, Besim and Evert, Stefan},
    title     = {{EmotiKLUE} at {IEST} 2018: {T}opic-Informed Classification of Implicit Emotions},
    booktitle = {Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis},
    year      = {2018},
    pages     = {235--242},
    address   = {Brussels},
    publisher = {Association for Computational Linguistics},
    url       = {https://aclweb.org/anthology/W18-6234},
  }