Aspect sentiment model for short text
This code was used in the experiments of the research paper
Reinald Kim Amplayo and Seung-won Hwang. Aspect Sentiment Model for Micro Reviews. ICDM, 2017.
To run the code, several parameters are needed to be set. Refer to our paper to determine the recommended values:
file
: The file containing the short text, one per line.seedDir
: The directory containing the sentiment seed words, containing two (or more if you have more sentiments, in the paper we only use positive and negative as sentiment labels) files named0.txt
and1.txt
. These files should contain one sentiment lexicon per line. For a sample seed word lists, refer to thedata
folder.noOfTopics
: The number of topics/aspects.noOfSentiments
: The number of sentiments.noOfPseudodocs
: The number of clusters.noOfIters
: The number of iterations.alpha
: The Dirichlet prior on the aspect-sentiment distribution.beta0
,beta1
,beta2
: The different Dirichlet priors on the word distribution that depends on the sentiment seed words given.gamma
: The Dirichlet prior on the sentiment distribution.delta
: The Dirichlet prior on the document distribution.window
: The context window to create aspect-sentiment pairs.
An example run would be:
java MicroASM data.txt seed/ 15 2 500 1500 0.1 0.01 0.1 0 1 0.1 5
To cite the paper/code, please use this BibTex:
@inproceedings{amplayo2017aspect,
Author = {Reinald Kim Amplayo and Seung-won Hwang},
Booktitle = {ICDM},
Location = {New Orleans, LA},
Year = {2017},
Title = {Aspect Sentiment Model for Micro Reviews},
}
If you have questions, send me an email: rktamplayo at yonsei dot ac dot kr