Indonesia's Question Answering System With Utilizing Of Indonesia Natural Language Inference (IndoNLI): An Experiment Of Natural Language Inference For Indonesian.

This is a repository code accompanying my final thesis and soon-to-be paper TODO. The experiments can be found under the experiments directory:

• main_fine_tuning_qas_dataset.py: this file contains experiment regarding to Sequence Classification task with use various kind of IndoNLI; like: Basic --the one that available in Hugging Face, Translated --the one that available translate_train.tar.gz in this repo, and lastly: Augmented --this is a concatenation from Basic & Translated version above.

• main_training_indonli.py: this file contains experiment regarding to Question Answering task with use two kind of experiments; without Intermediate Task Transfer Learning and with Intermediate Task Transfer Learning. You can read further by reading the README.md in experiments folder.

• filtering_nli_experiment.py: this file contains experiment regarding to filtering QAS system based on NLI with use three kind of experiments; QAS filtering system, smoothing type, and maximum search iteration. You can read further by reading the README.md in experiments folder.

The experiment code can be found under experiment directory, please check the related README file.

The utilites can be found under the utilities directory, utilities consist all of the support all of the experiment things. Such as: Exploratory Data Analysis of various of Question Answering datasets, code to augmented basic & translated IndoNLI, etc. Pplease check the related README file.

All of my experiment --in my local UI campus server, there are also several runs on the MBZUAI campus server on the @afaji (one of my supervisor) account; not the @muhammadravi251001 account-- is also available in Hugging Face (https://huggingface.co/muhammadravi251001).