[ANLP] Team Billie-Newman at SemEval-2023 Task 5: Clickbait Spoiling

This is the repository for the course 'Advanced Natural Language Processing' for the study 'Digital Sciences' at the University of Applied Sciences Cologne.
It contains the project code for the participation in the Clickbait Challenge proposed at SemEval-2023

• Task 1 Spoiler Classification: RoBERTa model with NER and custom components
• Task 2 Spoiler Generation: RoBERTa SQuAD2.0 model with rule-based approach
• Dataset: Webis Clickbait Spoiling Corpus 2022

---

Structure of this repository

• doc\: Contains the project presentation and project report
• task1_anlp_deploy\: Code and Docker File of Task 1
• task2_anlp_deploy\: Code and Docker File of Task 2

File description

filename	description
EDA.ipynb	Code for pre-processing the WEBIS Clickbait Spoiling Corpus 2022
simple_transformer_task1.ipynb	Code for training the RoBERTa model for multi-class classification
run_task_1.py	File for running the spoiler classifcation
Reformat_to_SQuAD.ipynb	Code for reformatting the spoiler questions into the SQuaD2.0 format
Training_model.ipynb	Code for training the RoBERTa-SQuAD2.0 model for the downstream task for spoiler generation
run_task_2.py	File for running the spoiler generation. Arguments: --apply_rule_base v1 / --apply_rule_base v2

Docker Images

The docker images can be pulled from these dockerhub repositories:
[Task 1 Dockerhub Repo] | [Task 2 Dockerhub Repo]

Task 1 Command

docker run --rm -d >>>CONTAINER_NAME<<< --input >>INPUT_DATA<<<.jsonl --output output.jsonl --apply_ner=yes

Task 2 Command

Without rule-based approach:

docker run --rm -d >>>CONTAINER_NAME<<< --input >>INPUT_DATA<<<.jsonl --output output.jsonl --apply_rule_base=v1

With rule-based approach:

docker run --rm -d >>>CONTAINER_NAME<<< --input >>INPUT_DATA<<<.jsonl --output output.jsonl --apply_rule_base=v2

Language Models

Due to their size, the language models can be downloaded separately from these sciebo links:
[Task 1 Model] | [Task 2 Model]

Rename the respective root folder to saved_models and put it in either task1_anlp_deploy/saved_models or task2_anlp_deploy/saved_models respectively for usage.

---

Sources

SemEval-2023 Task 5
Webis Clickbait Spoiling Corpus 2022