This system was used for the paper An impact analysis of features in a classification approach to irony detection in product reviews[1].
Download the system with the following command
> git clone https://github.com/kbuschme/irony-detection.git
Download the Sarcasm Corpus by Elena Filatova[2] from its Github repository and place it inside the corpora directory
> git clone https://github.com/ef2020/SarcasmAmazonReviewsCorpus.git corpora/SarcasmCorpus
Alternatively download the Zip archive and place the extracted files in the corpora directory.
Unpack the archive Ironic.rar into a directory corpora/SarcasmCorpus/Ironic and the archive Regular.rar into a directory corpora/SarcasmCorpus/Regular
> unrar e corpora/SarcasmCorpus/Ironic.rar corpora/SarcasmCorpus/Ironic/
> unrar e corpora/SarcasmCorpus/Regular.rar corpora/SarcasmCorpus/Regular/
Download the file opinion-lexicon-English.rar which contains the Opinion Lexicon by Hu and Liu[3] and place it inside the resources directory
> curl -o resources/opinion-lexicon-English.rar https://www.cs.uic.edu/~liub/FBS/opinion-lexicon-English.rar
Unpack the archive opinion-lexicon-English.rar and place the files negative-words.txt and positive-words.txt inside the resources directory
> unrar e resources/opinion-lexicon-English.rar resources/
A working Python 2 installation (tested with version 2.7.14) and the following Python libraries are needed. These can be installed using pip:
-
NumPy (version 1.14.0)
> sudo pip install numpy==1.14.0 -
SciPy (version 1.0.0)
> sudo pip install scipy==1.0.0 -
scikit-learn (version 0.19.1)
> sudo pip install scikit-learn==0.19.1 -
pydot (version 1.2.4) with pyparsing (version 2.2.0)
> sudo pip install pyparsing==2.2.0 > sudo pip install pydot==1.2.4 -
Natural Language Toolkit (NLTK) (version 3.2.5) with PyYAML (version 3.12)
> sudo pip install PyYAML==3.12 > sudo pip install nltk==3.2.5
Additionally NLTK requires the following models:
- Max Entropy Pos Tagger (maxent_treebank_pos_tagger) and
- Punkt Tokenizer Models (punkt)
which can be downloaded with the setup.py script
> python setup.py
or manually with the following steps:
> python
>>> import nltk
>>> nltk.download("punkt")
>>> nltk.download("maxent_treebank_pos_tagger")
>>> exit()
To start the system change the directory to src and run the file main.py which provides a command-line interface:
> cd src
> python main.py
the output should look as follows
> python main.py
usage: Irony Detector [-h] {corpus,feature,interactive,ml,sets} ...
Irony Detector: error: too few arguments
The following commands are available and described in the Manual section below:
corpus,features,interactive,mlandsets
As a first step the sets command should be run. This will create three files inside the corpora/SarcasmCorpus directory. The file shuffled_set.txt is a randomized version of the corpus used for cross-validation. The files training_set.txt and test_set.txt are a training and test set and contain 90% and 10% of the reviews, respectively.
> python main.py sets
Now the machine learning mode of the system can be used to classify reviews. The following example applies 10-fold cross-validation:
> python main.py ml cross-validation
Show a short help message about the available commands:
> python main.py -h
Show a detailed help message about the available commands:
> python main.py --help
This should look like the following message:
usage: Irony Detector [-h] {corpus,feature,interactive,ml,sets} ...
Detects irony in amazon reviews.
optional arguments:
-h, --help show this help message and exit
Commands:
The following commands can be invoked.
{corpus,feature,interactive,ml,sets}
Valid commands.
corpus Show details about the entire corpus.
feature Shows how often each feature is found for ironic and
regular reviews in the training_and_validation_set.
interactive The interactive mode classifies a given sentence using
a saved model.
ml Use the machine learning approach to classify reviews.
sets Divide the corpus into training, validation and test
set.
The corpus mode shows general information about a corpus.
Show all reviews inside the corpus:
> python main.py corpus reviews
Show some statistics about the corpus:
> python main.py corpus stats
The feature mode displays statistics about the specific features or exports all features as Attribute-Relation File Format (ARFF).
Show how often the specific features occur in all reviews:
> python main.py feature show
Export the extracted feature to an ARFF file:
> python main.py feature export
The machine learning mode uses the following classifiers to classify the reviews:
- Naive Bayes,
- Decision Tree,
- Random Forest,
- Logistic Regression and
- Support Vector Machine
Use 10-fold cross-validation:
> python main.py ml cross-validation
Train the classifiers on a training set and classify a test set:
> python main.py ml test
On one hand the sets mode generates a shuffled set for cross-validation and on the other hand divides all reviews into a training and test set by a 90 to 10 ratio.
> python main.py sets
This command creates the following three files inside the directory corpora/SarcasmCorpus:
corpora/SarcasmCorpus/shuffled_set.txt,corpora/SarcasmCorpus/training_set.txtandcorpora/SarcasmCorpus/test_set.txt.
[1] Konstantin Buschmeier, Philipp Cimiano, and Roman Klinger. An impact analysis of features in a classification approach to irony detection in product reviews. In Proceedings of the 5th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis, pages 42–49, Baltimore, Maryland, June 2014. Association for Computational Linguistics.
[2] Elena Filatova. 2012. Irony and sarcasm: Corpus Generation and Analysis Using Crowdsourcing. In Nicoletta Calzolari, Khalid Choukri, Thierry Declerck, Mehmet Uğur Doğan, Bente Maegaard, Joseph Mariani, Jan Odijk, and Stelios Piperidis, editors, Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC-2012), pages 392–398, Istanbul, Turkey, May. European Language Resources Association (ELRA).
[3] Minqing Hu and Bing Liu. 2004. Mining and summarizing customer reviews. In Proceedings of the tenth ACM SIGKDD international conference of Knowledge discovery and data mining, KDD '04, pages 168–177, New York, NY, USA. ACM.
Copyright (C) Konstantin Buschmeier.