app.py

import gradio as gr
import os
from transformers import AutoTokenizer, AutoModel
from sentence_transformers import SentenceTransformer
import pickle
import nltk
nltk.download('punkt') # tokenizer
nltk.download('averaged_perceptron_tagger') # postagger
import time

from input_format import *
from score import *

# load document scoring model
#torch.cuda.is_available = lambda : False  # uncomment to test with CPU only
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
#pretrained_model = 'allenai/specter'
pretrained_model = 'allenai/specter2'
tokenizer = AutoTokenizer.from_pretrained(pretrained_model)
doc_model = AutoModel.from_pretrained(pretrained_model) 
doc_model.to(device)

# load sentence model 
sent_model = doc_model # have the same model for document and sentence level

# OR specify different model for sentence level
#sent_model = SentenceTransformer('sentence-transformers/gtr-t5-base')
#sent_model.to(device)

def get_similar_paper(
    title_input,
    abstract_text_input, 
    author_id_input, 
    results={}, # this state variable will be updated and returned
): 
    progress = gr.Progress()
    num_papers_show = 10 # number of top papers to show from the reviewer
    if title_input == None:
        title_input = '' # if no title is given, just focus on abstract.
    print('retrieving similar papers...')
    start = time.time()
    input_sentences = sent_tokenize(abstract_text_input)
    
    # Get author papers from id
    #progress(0.1, desc="Retrieving reviewer papers ...")
    name, papers = get_text_from_author_id(author_id_input)
    
    # Compute Doc-level affinity scores for the Papers 
    # print('computing document scores...') 
    #progress(0.5, desc="Computing document scores...")
    # TODO detect duplicate papers?
    titles, abstracts, paper_urls, doc_scores = compute_document_score(
        doc_model, 
        tokenizer,
        title_input,
        abstract_text_input, 
        papers,
        batch=10
    )
    
    results = {
        'name': name,
        'titles': titles,
        'abstracts': abstracts,
        'urls': paper_urls,
        'doc_scores': doc_scores
    }
    
    # Select top K choices of papers to show
    titles = titles[:num_papers_show]
    abstracts = abstracts[:num_papers_show]
    doc_scores = doc_scores[:num_papers_show]
    paper_urls = paper_urls[:num_papers_show]
    
    display_title = ['[ %0.3f ] %s'%(s, t) for t, s in zip(titles, doc_scores)]
    end = time.time()
    retrieval_time = end - start
    print('paper retrieval complete in [%0.2f] seconds'%(retrieval_time))
    
    progress(0.9, desc="Obtaining relevant information from the papers...")
    print('obtaining highlights..')
    start = time.time()
    input_sentences = sent_tokenize(abstract_text_input)
    num_input_sents = len(input_sentences)
    
    for aa, (tt, ab, ds, url) in enumerate(zip(titles, abstracts, doc_scores, paper_urls)):
        # Compute sent-level and phrase-level affinity scores for each papers
        sent_ids, sent_scores, info, top_pairs_info = get_highlight_info(
            sent_model, 
            tokenizer,
            abstract_text_input, 
            ab,
            K=None, # top two sentences from the candidate 
            top_pair_num=3, # top five sentence pairs to show upfront 
        )
        num_cand_sents = sent_ids.shape[1]
        
        # get scores for each word in the format for Gradio Interpretation component
        word_scores = dict()
        for i in range(num_input_sents):
            word_scores[str(i)] = dict()
            for j in range(1, num_cand_sents+1):
                ww, ss = remove_spaces(info['all_words'], info[i][j]['scores'])
                word_scores[str(i)][str(j)] = {
                    "original": ab,
                    "interpretation": list(zip(ww, ss))
                } 

        results[display_title[aa]] = {
            'title': tt,
            'abstract': ab,
            'num_cand_sents': num_cand_sents,
            'doc_score': '%0.3f'%ds,
            'source_sentences': input_sentences,
            'highlight': word_scores,
            'top_pairs': top_pairs_info,
            'url': url
        }
       
    end = time.time()
    highlight_time = end - start
    print('done in [%0.2f] seconds'%(highlight_time)) 
    
    # debugging only
    pickle.dump(results, open('info.pkl', 'wb'))
    
    ## Set up output elements
    
    # first the list of top papers, sentences to select from, paper_title, affinity
    title = results[display_title[0]]['title'] # set default title as the top paper
    url = results[display_title[0]]['url']
    aff_score = results[display_title[0]]['doc_score']
    title_out = """<a href="%s" target="_blank"><h5>%s</h5></a>"""%(url, title)
    aff_score_out = '##### Affinity Score: %s'%aff_score
    out = [
        gr.update(choices=display_title, value=display_title[0],  interactive=True), # set of papers (radio)
        gr.update(choices=input_sentences, value=input_sentences[0], interactive=True), # submission sentences 
        gr.update(value=title_out), # paper_title
        gr.update(value=aff_score_out)  # affinity
    ]
    
    # set up elements to visualize upfront
    top_papers_show = 3 # number of top papers to show upfront
    top_num_info_show = 2 # number of sentence pairs from each paper to show upfront
    summary_out = []
    for i in range(top_papers_show):
        if i == 0:
            out_tmp = [
                gr.update(value="""<a href="%s" target="_blank"><h4>%s</h4></a>"""%(paper_urls[i], titles[i]), visible=True),
                gr.update(value="""#### Affinity Score: %0.3f 
                          <div class="help-tip">
                                <p>Measures how similar the paper's abstract is to the submission abstract.</p>
                          </div>
                          """%doc_scores[i],  
                          visible=True) # document affinity
            ]
        else:
            out_tmp = [
                gr.update(value="""<a href="%s" target="_blank"><h4>%s</h4></a>"""%(paper_urls[i], titles[i]), visible=True),
                gr.update(value='#### Affinity Score: %0.3f'%doc_scores[i], visible=True) # document affinity
            ]
        tp = results[display_title[i]]['top_pairs']
        for j in range(top_num_info_show):
            if i == 0 and j == 0:
                out_tmp += [
                    gr.update(value="""Sentence Relevance:\n%0.3f
                            <div class="help-tip">
                                <p>Measures how similar the sentence pairs are.</p>
                            </div>"""%tp[j]['score'], visible=True), # sentence relevance
                    tp[j]['query']['original'],
                    tp[j]['query'],
                    tp[j]['candidate']['original'],
                    tp[j]['candidate']
                ]
            else:
                out_tmp += [
                    gr.update(value='Sentence Relevance:\n%0.3f'%tp[j]['score'], visible=True), # sentence relevance
                    tp[j]['query']['original'],
                    tp[j]['query'],
                    tp[j]['candidate']['original'],
                    tp[j]['candidate']
                ]
        summary_out += out_tmp

    # add updates to the show more button
    out = out + summary_out + [gr.update(visible=True)] # make show more button visible
    assert(len(out) == (top_num_info_show * 5 + 2) * top_papers_show + 5)

    out += [gr.update(value="""
        <h3>Top three relevant papers by the reviewer <a href="%s" target="_blank">%s</a></h3>
        
        For each paper, two sentence pairs (one from the submission, one from the paper) with the highest relevance scores are shown.
        
        **<span style="color:black;background-color:#65B5E3;">Blue highlights</span>**: phrases that appear in both sentences.
        """%(author_id_input, results['name']),
        visible=True)] # result 1 description

    out += [gr.update(visible=True), gr.update(visible=True)] # demarcation line between results

    # progress status
    out += [gr.update(value='Done (in %0.1f seconds)'%(retrieval_time+highlight_time), visible=True)]
    
    # result 2 description
    desc = """
        ##### Click a paper by %s on the left (sorted by affinity scores), and a sentence from the submission on the right, to see which parts of the paper are relevant.
    """%results['name']
    out += [gr.update(value=desc)]
    
    # slider to control the number of highlights
    out += [gr.update(value=1, maximum=len(sent_tokenize(abstracts[0])))]
    
    # finally add the search results to pass on to the Gradio State varaible
    out += [results]
    
    return tuple(out)
        
def show_more(info):
    # show the interactive part of the app
    return (
        gr.update(visible=True), # description
        gr.update(visible=True), # set of papers
        gr.update(visible=True), # submission sentences
        gr.update(visible=True), # title row
        gr.update(visible=True), # affinity row
        gr.update(visible=True), # highlight legend
        gr.update(visible=True), # highlight slider
        gr.update(visible=True), # highlight abstract
    )

def show_status():
    # show search status field when search button is clicked
    return gr.update(visible=True)

def update_name(author_id_input):
    # update the name of the author based on the id input
    name, _ = get_text_from_author_id(author_id_input)
    
    return gr.update(value=name)

def change_sentence(
    selected_papers_radio, 
    source_sent_choice, 
    highlight_slider, 
    info={}
):
    # change the output highlight based on the sentence selected from the submission
    if len(info.keys()) != 0: # if the info is not empty
        source_sents = info[selected_papers_radio]['source_sentences']
        highlights = info[selected_papers_radio]['highlight']
        idx = source_sents.index(source_sent_choice)
        return highlights[str(idx)][str(highlight_slider)]
    else:
        return

def change_paper(
    selected_papers_radio, 
    source_sent_choice, 
    highlight_slider,
    info={}
): 
    if len(info.keys()) != 0: # if the info is not empty
        source_sents = info[selected_papers_radio]['source_sentences']
        title = info[selected_papers_radio]['title']
        num_sents = info[selected_papers_radio]['num_cand_sents']
        abstract = info[selected_papers_radio]['abstract']
        aff_score = info[selected_papers_radio]['doc_score']
        highlights = info[selected_papers_radio]['highlight']
        url = info[selected_papers_radio]['url']
        title_out = """<a href="%s" target="_blank"><h5>%s</h5></a>"""%(url, title)
        aff_score_out = '##### Affinity Score: %s'%aff_score
        idx = source_sents.index(source_sent_choice)
        if highlight_slider <= num_sents:
            return title_out, abstract, aff_score_out, highlights[str(idx)][str(highlight_slider)], gr.update(value=highlight_slider, maximum=num_sents)
        else: # if the slider is set to more than the current number of sentences, show the max number of highlights
            return title_out, abstract, aff_score_out, highlights[str(idx)][str(num_sents)], gr.update(value=num_sents, maximum=num_sents)
    else:
        return

def change_num_highlight(
    selected_papers_radio, 
    source_sent_choice,
    highlight_slider, 
    info={}
):
    if len(info.keys()) != 0: # if the info is not empty
        source_sents = info[selected_papers_radio]['source_sentences']
        highlights = info[selected_papers_radio]['highlight']
        idx = source_sents.index(source_sent_choice)
        return highlights[str(idx)][str(highlight_slider)] 
    else:
        return 

with gr.Blocks(css='style.css') as demo:
    info = gr.State({})  # cached search results as a State variable shared throughout
    
    # Text description about the app and disclaimer
    ### TEXT Description

    # General instruction
    general_instruction = """
# R2P2: An Assistance Tool for Reviewer-Paper Matching in Peer Review 

#### Who is it for?
It is for meta-reviewers, area chairs, program chairs, or anyone who oversees the submission-reviewer matching process in peer review for academic conferences, journals, and grants. 

<center><img src="file/tool-img.jpeg" width="70%" alt="general workflow"></center>

#### How does it help?
A typical meta-reviewer workflow lacks supportive information on **what makes the pre-selected candidate reviewers a good fit** for the submission. Only affinity scores between the reviewer and the paper are shown, without additional detail. 

R2P2 provides more information about each reviewer. It searches for the **most relevant papers** among the reviewer's previous publications and **highlights relevant parts** within them. 
    """
    # TODO add instruction video link
    # More details (video, addendum)
    more_details_instruction = """Check out <a href="", target="_blank">this video</a> for a quick demo of what R2P2 is and how it can help. You can find more details <a href="file/details.html", target="_blank">here</a>, along with our privacy policy and disclaimer."""

    gr.Markdown(general_instruction)
    gr.HTML(more_details_instruction)
    gr.Markdown("""---""")
    
    
    ### INPUT
    with gr.Row() as input_row:
        with gr.Column(scale=3):
            with gr.Row():
                title_input = gr.Textbox(label='Submission Title', info='Paste in the title of the submission.')
            with gr.Row():
                abstract_text_input = gr.Textbox(label='Submission Abstract', info='Paste in the abstract of the submission.')
        with gr.Column(scale=2):
            with gr.Row():
                author_id_input = gr.Textbox(label='Reviewer Profile Link (Semantic Scholar)', info="Paste in the reviewer's Semantic Scholar link")
            with gr.Row():
                name = gr.Textbox(label='Confirm Reviewer Name', info='This will be automatically updated based on the reviewer profile link above', interactive=False)
                author_id_input.change(fn=update_name, inputs=author_id_input, outputs=name)
                
    # Add examples
    example_title ="The Toronto Paper Matching System: An automated paper-reviewer assignment system"
    example_submission = """One of the most important tasks of conference organizers is the assignment of papers to reviewers. Reviewers' assessments of papers is a crucial step in determining the conference program, and in a certain sense to shape the direction of a field. However this is not a simple task: large conferences typically have to assign hundreds of papers to hundreds of reviewers, and time constraints make the task impossible for one person to accomplish. Furthermore other constraints, such as reviewer load have to be taken into account, preventing the process from being completely distributed. We built the first version of a system to suggest reviewer assignments for the NIPS 2010 conference, followed, in 2012, by a release that better integrated our system with Microsoft's popular Conference Management Toolkit (CMT). Since then our system has been widely adopted by the leading conferences in both the machine learning and computer vision communities. This paper provides an overview of the system, a summary of learning models and methods of evaluation that we have been using, as well as some of the recent progress and open issues."""
    example_reviewer = "https://www.semanticscholar.org/author/Nihar-B.-Shah/1737249"
    gr.Examples(
        examples=[[example_title, example_submission, example_reviewer]],
        inputs=[title_input, abstract_text_input, author_id_input],
        cache_examples=False,
        label="Try out the following example input."
    )
    
    with gr.Row():
        compute_btn = gr.Button('What Makes This a Good Match?') 

    with gr.Row():
        search_status = gr.Textbox(label='Search Status', interactive=False, visible=False)
    
        
    ### OVERVIEW
    # Paper title, score, and top-ranking sentence pairs -- two sentence pairs per paper, three papers
    ## ONE BLOCK OF INFO FOR A SINGLE PAPER
    ## PAPER1 
    with gr.Row():
        result1_desc = gr.Markdown(value='', visible=False)
    with gr.Row():
        with gr.Column(scale=3):
            paper_title1 = gr.Markdown(value='', visible=False)
        with gr.Column(scale=1):
            affinity1 = gr.Markdown(value='', visible=False)
    with gr.Row() as rel1_1:
        with gr.Column(scale=1):
            sent_pair_score1_1 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source1_1 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source1_1_hl = gr.components.Interpretation(sent_pair_source1_1)
        with gr.Column(scale=4):
            sent_pair_candidate1_1 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate1_1_hl = gr.components.Interpretation(sent_pair_candidate1_1)
    with gr.Row() as rel1_2:
        with gr.Column(scale=1):
            sent_pair_score1_2 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source1_2 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source1_2_hl = gr.components.Interpretation(sent_pair_source1_2)
        with gr.Column(scale=4):
            sent_pair_candidate1_2 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate1_2_hl = gr.components.Interpretation(sent_pair_candidate1_2)
    
    with gr.Row(visible=False) as demarc1:
        gr.Markdown(
            """---"""
        )
            
    ## PAPER 2
    with gr.Row():
        with gr.Column(scale=3):
            paper_title2 = gr.Markdown(value='', visible=False)
        with gr.Column(scale=1):
            affinity2 = gr.Markdown(value='', visible=False)
    with gr.Row() as rel2_1:
        with gr.Column(scale=1):
            sent_pair_score2_1 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source2_1 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source2_1_hl = gr.components.Interpretation(sent_pair_source2_1)
        with gr.Column(scale=4):
            sent_pair_candidate2_1 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate2_1_hl = gr.components.Interpretation(sent_pair_candidate2_1)
    with gr.Row() as rel2_2:
        with gr.Column(scale=1):
            sent_pair_score2_2 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source2_2 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source2_2_hl = gr.components.Interpretation(sent_pair_source2_2)
        with gr.Column(scale=4):
            sent_pair_candidate2_2 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate2_2_hl = gr.components.Interpretation(sent_pair_candidate2_2)
            
    with gr.Row(visible=False) as demarc2:
        gr.Markdown(
            """---"""
        )
    
    ## PAPER 3 
    with gr.Row():
        with gr.Column(scale=3):
            paper_title3 = gr.Markdown(value='', visible=False)
        with gr.Column(scale=1):
            affinity3 = gr.Markdown(value='', visible=False)
    with gr.Row() as rel3_1:
        with gr.Column(scale=1):
            sent_pair_score3_1 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source3_1 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source3_1_hl = gr.components.Interpretation(sent_pair_source3_1)
        with gr.Column(scale=4):
            sent_pair_candidate3_1 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate3_1_hl = gr.components.Interpretation(sent_pair_candidate3_1)
    with gr.Row() as rel3_2:
        with gr.Column(scale=1):
            sent_pair_score3_2 = gr.Markdown(interactive=False, value='', visible=False)
        with gr.Column(scale=4):
            sent_pair_source3_2 = gr.Textbox(label='Sentence from Submission', visible=False)
            sent_pair_source3_2_hl = gr.components.Interpretation(sent_pair_source3_2)
        with gr.Column(scale=4):
            sent_pair_candidate3_2 = gr.Textbox(label="Sentence from Reviewer's Paper", visible=False)
            sent_pair_candidate3_2_hl = gr.components.Interpretation(sent_pair_candidate3_2)

    ## Show more button
    with gr.Row():
        see_more_rel_btn = gr.Button('Explore more', visible=False)
    
    ### PAPER INFORMATION
    
    # Description for Explore More Section
    with gr.Row():
        result2_desc = gr.Markdown(value='', visible=False)
        
    # Highlight description 
    hl_desc = """
    <font size="2">**<span style="color:black;background-color:#DB7262;">Red</span>**: sentences simiar to the selected sentence from submission. Darker = more similar.</font>
    
    <font size="2">**<span style="color:black;background-color:#65B5E3;">Blue</span>**: phrases that appear in both sentences.</font>
    """ 
    #---"""
    # show multiple papers in radio check box to select from
    paper_abstract = gr.Textbox(label='Abstract', interactive=False, visible=False)
    with gr.Row():
        with gr.Column(scale=1):
            selected_papers_radio = gr.Radio(
                choices=[], # will be udpated with the button click
                visible=False, # also will be updated with the button click
                label='Top Relevant Papers from the Reviewer'
            )
        with gr.Column(scale=2):
            # sentences from submission
            source_sentences = gr.Radio(
                choices=[], 
                visible=False, 
                label='Sentences from Submission Abstract',
            )
        with gr.Column(scale=3):
            # selected paper and highlight
            with gr.Row():
                # slider for highlight amount
                highlight_slider = gr.Slider(
                    label='Number of Highlighted Sentences', 
                    minimum=1, 
                    maximum=15, 
                    step=1, 
                    value=2, 
                    visible=False
                )
            with gr.Row():
                # highlight legend
                highlight_legend = gr.Markdown(value=hl_desc, visible=False)
            with gr.Row(visible=False) as title_row:
                # selected paper title
                paper_title = gr.Markdown(value='')
            with gr.Row(visible=False) as aff_row:
                # selected paper's affinity score
                affinity = gr.Markdown(value='')
            with gr.Row(visible=False) as hl_row: 
                # highlighted text from paper
                highlight = gr.components.Interpretation(paper_abstract) 
                

    ### EVENT LISTENERS
    
    compute_btn.click(
        fn=show_status,
        inputs=[],
        outputs=search_status
    )
    
    # retrieve similar papers and show top results
    compute_btn.click(
        fn=get_similar_paper,
        inputs=[
            title_input,
            abstract_text_input, 
            author_id_input,
            info
        ],
        outputs=[
            selected_papers_radio, # list of papers for show more section
            source_sentences, # list of sentences for show more section
            paper_title, # paper title for show more section
            affinity, # paper affinity for show more section
            paper_title1, # paper info
            affinity1,
            sent_pair_score1_1,
            sent_pair_source1_1,
            sent_pair_source1_1_hl,
            sent_pair_candidate1_1,
            sent_pair_candidate1_1_hl,
            sent_pair_score1_2,
            sent_pair_source1_2,
            sent_pair_source1_2_hl,
            sent_pair_candidate1_2,
            sent_pair_candidate1_2_hl,
            paper_title2,
            affinity2,
            sent_pair_score2_1,
            sent_pair_source2_1,
            sent_pair_source2_1_hl,
            sent_pair_candidate2_1,
            sent_pair_candidate2_1_hl,
            sent_pair_score2_2,
            sent_pair_source2_2,
            sent_pair_source2_2_hl,
            sent_pair_candidate2_2,
            sent_pair_candidate2_2_hl,
            paper_title3,
            affinity3, 
            sent_pair_score3_1, 
            sent_pair_source3_1,
            sent_pair_source3_1_hl,
            sent_pair_candidate3_1,
            sent_pair_candidate3_1_hl,
            sent_pair_score3_2,
            sent_pair_source3_2,
            sent_pair_source3_2_hl,
            sent_pair_candidate3_2,
            sent_pair_candidate3_2_hl,
            see_more_rel_btn,
            result1_desc,
            demarc1,
            demarc2,
            search_status,
            result2_desc,
            highlight_slider,
            info,
        ],
        show_progress=True,
        scroll_to_output=True
    )      
    
    # Get more info (move to more interactive portion)
    see_more_rel_btn.click(
        fn=show_more,
        inputs=info,
        outputs=[
            result2_desc,
            selected_papers_radio,
            source_sentences,
            title_row,
            aff_row,
            highlight_legend,
            highlight_slider,
            hl_row,
        ]
    )
    
    # change highlight based on selected sentences from submission
    source_sentences.change(
        fn=change_sentence,
        inputs=[
            selected_papers_radio,
            source_sentences,
            highlight_slider,
            info
        ],
        outputs=highlight
    )
    
    # change paper to show based on selected papers
    selected_papers_radio.change(
        fn=change_paper,
        inputs=[
            selected_papers_radio,
            source_sentences,
            highlight_slider,
            info,
        ],
        outputs= [
            paper_title,
            paper_abstract,
            affinity,
            highlight, 
            highlight_slider
        ]
    )
    
    highlight_slider.change(
        fn=change_num_highlight,
        inputs=[
            selected_papers_radio,
            source_sentences,
            highlight_slider,
            info
        ],
        outputs=[
            highlight
        ]
    )
    
if __name__ == "__main__":
    demo.queue().launch()  # add ?__theme=light to force light mode