app.py

import gradio as gr
from huggingface_hub import InferenceClient
import re
import datetime
from urllib import request
from lxml import etree


url_prefix_mapping = {
    'acl': 'https://aclanthology.org',
    'emnlp': 'https://aclanthology.org',
    'naacl': 'https://aclanthology.org',
    'tacl': 'https://aclanthology.org',
    'nips': 'https://papers.nips.cc',
    'icml': 'https://papers.nips.cc',
    'iclr': 'https://iclr.cc',
}

mlr_mapping = {
    ('icml', 2020): 'v119',
    ('icml', 2021): 'v139',
    ('icml', 2022): 'v162',
    ('icml', 2023): 'v202',
    ('icml', 2024): 'v139',
}

def get_paper_home(venue, year):
    if venue in ['acl', 'emnlp', 'naacl']:
        return f'https://aclanthology.org/events/{venue}-{year}'
    
    elif venue == 'nips':
        return f'https://papers.{venue}.cc/paper_files/paper/{year}'
    
    elif venue == 'icml':
        return f'https://proceedings.mlr.press/{mlr_mapping[(venue, year)]}'
    
    elif venue == 'iclr':
        return f'https://iclr.cc/Downloads/{year}'


def check_keywords(ele, keywords):
    s = ''.join(ele.itertext()).lower()
    url = ele.get('href')
    for i in keywords:
        match = re.search(i, s)
        if match:
            return True

    return False

def check_keywords_icml(ele, keywords):
    s = ''.join(ele.find('.//p[@class="title"]').itertext()).lower()
    url = ele.get('href')
    for i in keywords:
        match = re.search(i, s)
        if match:
            return True

    return False


def search(keywords, venues, min_year, max_year):
    keywords = [keyword.strip() for keyword in keywords.split(",")]

    year_range = list(range(min_year, max_year))
    
    search_venues = []
    if "NeurIPS/ICLR/ICML" in venues:
        search_venues.extend(['nips', 'iclr', 'icml'])
    if "*ACL" in venues:
        search_venues.extend(['acl', 'emnlp', 'naacl', 'tacl'])
    if "CVPR/ECCV/ICCV" in venues:
        search_venues.extend(['nips', 'iclr', 'icml'])

    results = []
    for venue in search_venues:
        for year in year_range:
            print(venue, year)
            paper_home = get_paper_home(venue, year)
            url_prefix = url_prefix_mapping[venue]
            if venue == 'icml':
                url_prefix = paper_home

            try:
                response = request.urlopen(paper_home)
            except:
                continue

            html = response.read().decode()
            tree = etree.fromstring(html, etree.HTMLParser())
            
            if 'acl' in venue:
                paper_tag_on_html = ".//a[@class='align-middle']"
            elif venue == 'iclr':
                paper_tag_on_html = ".//a[@class='Poster']"
            elif venue == 'nips':
                paper_tag_on_html = ".//a[@title='paper title']"
            elif venue == 'icml':
                paper_tag_on_html = ".//div[@class='paper']"
            
            elements = tree.findall(paper_tag_on_html)
            for element in elements:
                if venue == 'icml':
                    if check_keywords_icml(element, keywords):
                        paper_url = element.find('.//p[@class="links"]').find('a').get('href')
                        results.append([paper_url])
                else:
                    if check_keywords(element, keywords):
                        paper_url = url_prefix + element.get('href')
                        results.append([paper_url])
                        
            print(len(elements))
            print()

    return results


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
current_year = datetime.datetime.now().year

# demo = gr.Interface(
#     search,
#     inputs=[
#         gr.Textbox(lines=2, placeholder="Keywords of the paper title. Supports ReGex."),
#         gr.CheckboxGroup(["NeurIPS/ICLR/ICML", "*ACL", "CVPR/ECCV/ICCV"], label="Choose Venues to Search", value=["NeurIPS/ICLR/ICML", "*ACL", "CVPR/ECCV/ICCV"]),
#         gr.Slider(minimum=2020, maximum=current_year, value=[2020, current_year], label="Year Range", step=1)
#     ],
#     outputs=gr.DataFrame(headers=["Paper Link", "Title", "Authors"])
# )
def test_search(keywords, venues, min_year, max_year):
    return [["https://example.com"], ["https://anotherexample.com"]]
    
with gr.Blocks() as demo:
    with gr.Row():  # Organize inputs and outputs in a row (side by side)
        with gr.Column(scale=1):  # Input section (narrower)
            # Textbox for keywords
            textbox = gr.Textbox(
                label="Enter comma-separated keywords",
                placeholder="Enter keywords, separated by commas...",
                lines=2
            )
            # Vertical checkbox group for actions
            checkbox = gr.CheckboxGroup(
                ["NeurIPS/ICLR/ICML", "*ACL", "CVPR/ECCV/ICCV"],
                label="Choose Venues to Search",
                value=["NeurIPS/ICLR/ICML", "*ACL", "CVPR/ECCV/ICCV"],
                type="value"
            )
            # Year range slider
            min_year_slider = gr.Slider(minimum=2015, maximum=current_year, value=2020, label="Select Min Year", step=1)
            max_year_slider = gr.Slider(minimum=2015, maximum=current_year, value=current_year, label="Select Max Year", step=1)

            submit_button = gr.Button("Search")
            
        with gr.Column(scale=3):  # Output section (wider)
            # Output table
            output_table = gr.DataFrame(
                headers=["Paper Link",],# "Title", "Authors"
                label="Results"
            )

    # Link the input components to the output function
    submit_button.click(
        search,
        inputs=[textbox, checkbox, min_year_slider, max_year_slider],
        outputs=output_table
    )

if __name__ == "__main__":
    demo.launch()