This project represent a code for paper
Methods of Lexicon Integration with Machine Learning for Sentiment Analysis System
and describes the application of SVM
classifier and Neural Networks
classifiers for sentiment classification of Russian Twitter messages in the
banking and telecommunications domains of SentiRuEval-2016 competition.
All dependencies described in Makefile
and could be installed as follows:
make install
For research purposes. Use run/Makefile
to run workflow for
certain task (bank
or tcc
) and classifier (svm
, lr
), for example:
cd run && make svm_sre15_bank_w2v_bal
returns F-macro/micro result for SentiRuEval-2015 bank dataset using w2v-based embedding model for balanced test collection.
All embedding classifier settings presented in data/embedding
folder.
- SentiRuEval-2015 contest data;
- SentiRuEval-2016 contest data & results of this approach (participant #1) in comparation with the other participants.
This work has been formed into AIDT Journal 2017/2 paper publication.
The latter was presented at RUSSIR conference in a form of the following posters: RUSSIR-2017, RUSSIR-2016;
Early publication could be found here: Dialog-2016;
@article{rusnachenko2017methods,
title={Методы интеграции лексиконов в машинное обучение для систем анализа тональности},
author={Русначенко, Николай Леонидович and Лукашевич, Наталья Валентиновна},
journal={Искусственный интеллект и принятие решений},
number={2},
pages={78--89},
year={2017},
publisher={Федеральное государственное учреждение" Федеральный исследовательский центр~…}
}