https://github.com/hamedbabaei/author-profiling-pan2023
Symbol Team model for PAN@AP 2023 shared task on Profiling Cryptocurrency Influencers with Few-shot Learning
https://github.com/hamedbabaei/author-profiling-pan2023
author-profiling contrastive-learning encoder-decoder-model few-shot-learning flan-t5 large-language-models
Last synced: 6 days ago
JSON representation
Symbol Team model for PAN@AP 2023 shared task on Profiling Cryptocurrency Influencers with Few-shot Learning
- Host: GitHub
- URL: https://github.com/hamedbabaei/author-profiling-pan2023
- Owner: HamedBabaei
- License: mit
- Created: 2023-06-03T17:50:18.000Z (over 2 years ago)
- Default Branch: main
- Last Pushed: 2024-01-17T13:24:19.000Z (about 2 years ago)
- Last Synced: 2025-09-05T04:43:02.302Z (5 months ago)
- Topics: author-profiling, contrastive-learning, encoder-decoder-model, few-shot-learning, flan-t5, large-language-models
- Language: Python
- Homepage:
- Size: 1.48 MB
- Stars: 1
- Watchers: 2
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Citation: CITATION.cff
Awesome Lists containing this project
README
# Leveraging Large Language Models with Multiple Loss Learners for Few-Shot Author Profiling
This repository contains the code and data for the paper "Leveraging Large Language Models with Multiple Loss Learners for Few-Shot Author Profiling" by `Hamed Babaei Giglou, Mostafa Rahgouy, Jennifer D’Souza, Milad Molazadeh Oskuee, Hadi Bayrami Asl Tekanlou, and Cheryl D Seals`. The paper was presented at the 14th International Conference of the CLEF Association (CLEF 2023).
The objective of author profiling (AP) is to study the characteristics of authors through the analysis of how language is exchanged among people. Studying these attributes sometimes is challenging due to the lack of annotated data. This indicates the significance of studying AP from a low-resource perspective. This year at AP@PAN 2023 the major interest raised in profiling cryptocurrency influencers with a few-shot learning technique to analyze the effectiveness of advanced approaches in dealing with new tasks from a low-resource perspective.
## Architecture of Proposed Framework
## How to use this repository
### Directories
- `dataset/`: Contains the datasets used in the paper.
- `assets/`: Contains the model checkpoints used in the paper.
- `visualization/`: Contains the code for the visualizing experiments in the paper.
- `results/`: Contains the results of the experiments.
### Requirements
- Python 3.9 or higher
- PyTorch 1.9.x or higher
- Transformers 4.3.x or higher
### Usage
1. Clone the repository:
```
https://github.com/HamedBabaei/author-profiling-pan2023
cd author-profiling-pan2023
```
2. Install the required packages:
```
pip install -r requirements.txt
```
3. Run the experiments:
1. Inference fsl
```bash
bash inference_fsl.sh
```
2. Inference fsl
```bash
bash inference_fsl_biencoder.sh
```
3. Baseline (random)
```bash
bash random_baseline.sh
```
4. Baseline (Zero Shot)
```bash
bash zero_shot_baseline.sh
```
5. Train & Test SBERT
```bash
bash train_test_runner_sbert.sh
```
6. Train & Test flanT5
```bash
bash train_test_runner_flan_t5.sh
```
## Citation
If you use this code in your research, please cite the following paper:
```bib
@InProceedings{giglou:2023,
author = {Hamed Babaei Giglou, Mostafa Rahgouy, Jennifer D’Souza, Milad Molazadeh Oskuee , Hadi Bayrami Asl Tekanlou and Cheryl D Seals},
booktitle = {{CLEF 2023 Labs and Workshops, Notebook Papers}},
month = sep,
publisher = {CEUR-WS.org},
title = {{Leveraging Large Language Models with Multiple Loss Learners for Few-Shot Author Profiling}},
year = 2023
}
```