app.py

# Created by Hansi at 30/08/2023
import os

import nltk
nltk.download('punkt_tab')
nltk.download('averaged_perceptron_tagger_eng')

import streamlit as st
from PIL import Image
from accord_nlp.information_extraction.convertor import entity_pairing, graph_building
from accord_nlp.information_extraction.ie_pipeline import InformationExtractor

from trubrics.integrations.streamlit import FeedbackCollector

ner_args = {
    "labels_list": ["O", "B-quality", "B-property", "I-property", "I-quality", "B-object", "I-object", "B-value", "I-value"],
    "use_multiprocessing": False,
    "process_count": 1
}

re_args = {
    "labels_list": ["selection", "necessity", "none", "greater", "part-of", "equal", "greater-equal", "less-equal", "not-part-of", "less"],
    "special_tags": ["<e1>", "<e2>"],  # Should be either begin_tag or end_tag
    "use_multiprocessing": False,
    "process_count": 1
}

@st.cache_resource
def init():
    return InformationExtractor(
        ner_model_info=('roberta', 'ACCORD-NLP/ner-roberta-large', ner_args),
        re_model_info=('roberta', 'ACCORD-NLP/re-roberta-large', re_args))


st.set_page_config(
    page_title='ACCORD NLP Demo',
    initial_sidebar_state='expanded',
    layout='wide',
)

with st.spinner(text="Initialising..."):
    ie = init()


collector = FeedbackCollector(
    # component_name="default",
    email=st.secrets["TRUBRICS_EMAIL"],
    password=st.secrets["TRUBRICS_PASSWORD"],
    project="accord-nlp-ie"
)


def main():
    image = Image.open(os.path.join(os.path.dirname(__file__), 'accord_logo.png'))
    st.sidebar.image(image)

    # st.sidebar.markdown("[![image](upload://accordproject.eu/wp-content/uploads/2022/08/accord_logo-e1662800862179.png)](https://accordproject.eu/)")
    # st.sidebar.markdown(
    #     "[![image](os.path.join(os.path.dirname(__file__), 'accord_logo.png'))](https://accordproject.eu/)")


    # st.sidebar.title("ACCORD-NLP")
    st.sidebar.header("Information Extractor")
    st.sidebar.markdown("Extract entities and their relations from textual data")
    st.sidebar.markdown(
        "[codebase](https://github.com/Accord-Project/NLP-Framework)"
    )
    st.sidebar.markdown(
        "[models](https://huggingface.co/ACCORD-NLP)"
    )


    if 'text' not in st.session_state:
        st.session_state['text'] = ''
    if 'graph' not in st.session_state:
        st.session_state['graph'] = None

    st.header("Input a sentence")

    txt = st.text_area('Sentence')

    if txt:
        if txt == st.session_state['text']:
            st.header('Entity-Relation Representation')
            st.graphviz_chart(st.session_state['graph'], use_container_width=True)

            st.session_state['text'] = txt

        else:
            st.session_state['text'] = txt
            st.session_state['graph'] = None

            # preprocess
            sentence = ie.preprocess(txt)

            # NER
            with st.spinner(text="Recognising entities..."):
                ner_predictions, ner_raw_outputs = ie.ner_model.predict([sentence])

            with st.spinner(text="Extracting relations..."):
                # pair entities to predict their relations
                entity_pair_df = entity_pairing(sentence, ner_predictions[0])

                # relation extraction
                re_predictions, re_raw_outputs = ie.re_model.predict(entity_pair_df['output'].tolist())
                entity_pair_df['prediction'] = re_predictions

            with st.spinner(text="Building graph..."):
                # build graph
                graph = graph_building(entity_pair_df, view=False)

            st.header('Entity-Relation Representation')
            # st.graphviz_chart(graph)
            st.graphviz_chart(graph, use_container_width=True)

            st.session_state['graph'] = graph

    # if st.session_state['graph'] is not None:
        # st.divider()
        # st.write("Does this prediction look correct?")
        # collector.st_feedback(
        #     component="default",
        #     feedback_type="thumbs",
        #     model="v1-test",
        #     align="flex-start",
        #     metadata={
        #         "sentence": txt
        #     },
        #     open_feedback_label="[Optional] Provide additional feedback",
        #     # single_submit=False
        # )

        # st.session_state['text'] = ''
        # st.session_state['graph'] = None


if __name__ == '__main__':
    main()