Skip to content

Codes and Datasets for our ECIR 2021 Paper: "Reproducibility, Replicability and Beyond: Assessing Production Readiness of Aspect Based Sentiment Analysis in the Wild"

Notifications You must be signed in to change notification settings

rajdeep345/ABSA-Reproducibility

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 

Repository files navigation

ABSA-Reproducibility

Codes and Datasets for our ECIR 2021 Paper: "Reproducibility, Replicability and Beyond: Assessing Production Readiness of Aspect Based Sentiment Analysis in the Wild"

Setup instructions

  • Create a conda environment using the requirements.txt file.
  • Alternately, one can use the ABSA.yml extracted from our conda environment to exactly replicate the environment.
  • Download and unzip the GloVe embeddings into the current folder.

Running experiments

python grid_search.py

Evaluating results

  • Change directory to results/
python process_results.py path [isHard]

path - Select one from [in_domain, contrast_logs, cross_domain, cross_domain_incremental] isHard - [Default: 'False'] is set as 'True' only if you want to evaluate hard set results for in_domain experiments, i.e.

python process_results.py in_domain True

Additional notes

  • We run each experiment with 5 random seeds (1,2,3,4,5).
  • Our experiments were run on a Tesla P100 PCIE, 16GB GPU and CUDA 10.1 and PyTorch 1.1.0.
  • For the incremental cross domain experiments, the --train_dataset argument can be set to crossdomain_indomain_ratio, for instance Laptops_Restaurants_0.1 for evaluting the cross domain combination (Laptops - Train, Restaurants - Test).

About

Codes and Datasets for our ECIR 2021 Paper: "Reproducibility, Replicability and Beyond: Assessing Production Readiness of Aspect Based Sentiment Analysis in the Wild"

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages