arruw / fri-1920-nlp-project

Aspect-based sentiment analysis

Aspect-based sentiment analysis

• Authors: Žan Jaklič, Iztok Ramovš, Matjaž Mav
• Draft preview: here

Folder structure

code : IPython Notebooks and Python files
data : required datasets and saved pickle DataFrames for faster execution
models : saved Neural Net Models
old : deprecated, unrunnable files from the project's 1st phase

File order

Run notebooks in the given order:

parsing.ipynb
context_extraction.ipynb
feature_expansion.ipynb
modelling.ipynb

For evaluating saved FFN model run following:

python code/nn_from_feature_vector.py

To retrain FFN model run following:

python code/nn_from_feature_vector.py -train

Installation help

# Download Anaconda (Python 3.7) for your OS
https://www.anaconda.com/products/individual

# Install PyTorch
conda install pytorch torchvision -c pytorch

# Install Stanza
conda install -c stanfordnlp stanza

# Open Python in Anaconda Prompt and download Slovene Stanza modelling
import stanza
stanza.download('sl')

# Install Keras and Tensorflow (gpu or cpu, doesnt't matter)
conda install -c conda-forge tensorflow
conda install keras

# Install h5py for saving and reading Keras models to disk
conda install h5py 

# Install Pandas
conda install -c anaconda pandas