from dataclasses import dataclass
from enum import Enum

@dataclass
class Task:
    benchmark: str
    metric: str
    col_name: str


class Tasks(Enum):
    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
    Task1_Title_Search_Rate = Task("(Task1) Title Search Rate", "(Task1) Title Search Rate","(T1) Title Search Rate (%)")
    Task1_Precision = Task("(Task1) Precision", "(Task1) Precision","(T1) Precision (%)")
    Task1_Overlap = Task("(Task1) Overlap", "(Task1) Overlap","(T1) Overlap (%)")
    Task1_Precision_First_Author = Task("(Task1) Precision (First Author)", "(Task1) Precision (First Author)","(T1) Precision (First Author) (%)")
    Task1_Overlap_First_Author = Task("(Task1) Overlap (First Author)", "(Task1) Overlap (First Author)","(T1) Overlap (First Author) (%)")
    # Task2
    Task2_Similarity = Task("(Task2) Similarity", "(Task2) Similarity", "(T2) Similarity (%)")
    Task2_Entail_TRUE = Task("(Task2) Entail (TRUE)", "(Task2) Entail (TRUE)", "(T2) Entail (TRUE %)")
    Task2_Entail_GPT_4o = Task("(Task2) Entail (GPT-4o)", "(Task2) Entail (GPT-4o)", "(T2) Entail (GPT-4o %)")
    Task2_ROUGE_1 = Task("(Task2) ROUGE-1", "(Task2) ROUGE-1", "(T2) ROUGE-1 (%)")
    Task2_ROUGE_2 = Task("(Task2) ROUGE-2", "(Task2) ROUGE-2", "(T2) ROUGE-2 (%)")
    Task2_ROUGE_L = Task("(Task2) ROUGE-L", "(Task2) ROUGE-L", "(T2) ROUGE-L (%)")

    # Task3
    Task3_Precision = Task("(Task3) Precision", "(Task3) Precision", "(T3) Precision (%)")
    Task3_Title_Search_Rate = Task("(Task3) Title Search Rate", "(Task3) Title Search Rate", "(T3) Title Search Rate (%)")
    Task3_Overlap = Task("(Task3) Overlap", "(Task3) Overlap", "(T3) Overlap (%)")
    Task3_KPR = Task("(Task3) KPR", "(Task3) KPR", "(T3) KPR (%)")
    Task3_ROUGE_1 = Task("(Task3) ROUGE-1", "(Task3) ROUGE-1", "(T3) ROUGE-1 (%)")
    Task3_ROUGE_2 = Task("(Task3) ROUGE-2", "(Task3) ROUGE-2", "(T3) ROUGE-2 (%)")
    Task3_ROUGE_L = Task("(Task3) ROUGE-L", "(Task3) ROUGE-L", "(T3) ROUGE-L (%)")

# Your leaderboard name
TITLE = """<h1 align="center" id="space-title">LLM-Based Automated Literature Review Evaluation Benchmark</h1>"""

# What does your leaderboard evaluate?
# Which evaluations are you running? how can people reproduce what you have?

INTRODUCTION_TEXT = """
This leaderboard evaluates Large Language Models (LLMs) on their ability to perform automated literature review tasks, including reference generation, abstract writing, and review composition.<br>
It is based on the study: <b>Large Language Models for Automated Literature Review: An Evaluation of Reference Generation, Abstract Writing, and Review Composition.</b><br>
The leaderboard measures how well different models perform in references generation, factually consistent, and stylistically appropriate academic texts.<br><br>
"""


# <table>
# <tr>
#     <td align="center">
#         <img src="https://huggingface.co/spaces/XuemeiTang/LLM4LitReview_Benchmark/resolve/main/llm_litReview_images/acc_score.png" width="150"><br>
#         Reference Generation: Precision
#     </td>
#     <td align="center">
#         <img src="https://huggingface.co/spaces/XuemeiTang/LLM4LitReview_Benchmark/resolve/main/llm_litReview_images/t2_true_entailment.png" width="150"><br>
#         Abstract Writing: True
#     </td>
#     <td align="center">
#         <img src="https://huggingface.co/spaces/XuemeiTang/LLM4LitReview_Benchmark/resolve/main/llm_litReview_images/acc_score_t3.png" width="150"><br>
#         Review Composition: Precision
#     </td>
#     <td align="center">
#         <img src="https://huggingface.co/spaces/XuemeiTang/LLM4LitReview_Benchmark/resolve/main/llm_litReview_images/kpr_score.png" width="150"><br>
#         Literature Review Writing: KPR
#     </td>
# </tr>
# </table>
# """

EVALUATION_QUEUE_TEXT = """"""
# Which evaluations are you running? How can people reproduce what you have?
LLM_BENCHMARKS_TEXT = """
## Introduction

The **LLM4LitReview Leaderboard** is dedicated to evaluating the capabilities of large language models (LLMs) in automating academic writing tasks, specifically literature review generation.

We focus on three subtasks:
1. **Reference Generation** – accuracy and validity of citations.
2. **Abstract Writing** – semantic coverage and factual consistency.
3. **Review Composition** – accuracy and validity of citations and semantic coverage and factual consistency.

This benchmark provides a structured and reproducible framework for assessing how close LLMs are to human-level academic writing quality.


---

## Evaluation Dataset

Our evaluation dataset includes:
- 1015 academic papers sampled from open-access journals across multiple disciplines.
- For each paper, models generate:
  - A reference list according to the given title and key words
  - An abstract summarizing according to the given title and key words
  - A short literature review paragraph based on the provided title, keywords, and abstract

---

## Metrics Explained
- **Precision** – Precision of references that correspond to real, verifiable academic papers.
- **Title_search_rate** – Whether the generated paper can be searched by title in Semantic Scholar.
- **Overlap_rate** – LLM-cited vs human-cited references.
- **Similarity** – Semantic similarity between model and human-generated texts (human-written and LLM-generated abstract).
- **Entailment (TRUE %)** – Factual consistency between model and human-generated texts (Use NLI model TRUE as the evaluator).
- **Entailment (GPT-4o%)** – Factual consistency between model and human-generated texts (Use GPT-4o as the evaluator).
- **ROUGE-1, ROUGE-2, ROUGE-L** – Standard metrics for evaluating text generation quality based on n-gram Overlap with human-written texts.
- **KPR (Key Point Recall)** – Measures how well key points from source documents are captured in the generated text.
---

"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""
@misc{tang2025largelanguagemodelsautomated,
      title={Large Language Models for Automated Literature Review: An Evaluation of Reference Generation, Abstract Writing, and Review Composition}, 
      author={Xuemei Tang and Xufeng Duan and Zhenguang G. Cai},
      year={2025},
      eprint={2412.13612},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.13612}, 
}"""