app.py

import requests
import streamlit as st
from bs4 import BeautifulSoup
import pandas as pd
from transformers import pipeline
import plotly.express as px
import time
import io
import os  
from comet_ml import Experiment 
import zipfile
import re
from streamlit_extras.stylable_container import stylable_container
import numpy as np
from transformers import AutoTokenizer, AutoModelForTokenClassification

st.set_page_config(layout="wide", page_title="Named Entity Recognition App")

COMET_API_KEY = os.environ.get("COMET_API_KEY")
COMET_WORKSPACE = os.environ.get("COMET_WORKSPACE")
COMET_PROJECT_NAME = os.environ.get("COMET_PROJECT_NAME")

comet_initialized = False
if COMET_API_KEY and COMET_WORKSPACE and COMET_PROJECT_NAME:
    comet_initialized = True

st.subheader("Text & URL Chinese NER App", divider="rainbow")
st.link_button("by nlpblogs", "https://nlpblogs.com", type="tertiary")

expander = st.expander("**Important notes on the Text & URL Chinese NER App**")
expander.write('''
        **Named Entities:** This Text & URL Chinese NER App predicts eighteen (18) labels ("**CARDINAL**: cardinal number”, “**DATE**: date”, “**EVENT**: event name”, “**FAC**: facilities”, “**GPE**: geopolitical entity”, "**LANGUAGE**: language", "**LAW**: law", "**LOC**: location", "**MONEY**: money", "**NORP**: ethnic, religious, political groups", "**ORDINAL**: ordinal number", "**ORG**: organization", "**PERCENT**: percent value", "**PERSON**: person", "**PRODUCT**: product", "**QUANTITY**: quantity", "**TIME**: time", "**WORK_OF_ART**: work of art"). Results are presented in an easy-to-read table, visualized in an interactive tree map, pie chart, and bar chart, and are available for download along with a Glossary of tags.
        
        **How to Use:** Paste a URL, and then press Enter. If you type or paste text, just press Ctrl + Enter.
        
        **Usage Limits:** You can request results up to 10 times.

        **Language settings:** Please check and adjust the language settings in your computer, so the Chinese characters are handled properly in your downloaded file.
        
        **Customization:** To change the app's background color to white or black, click the three-dot menu on the right-hand side of your app, go to Settings and then Choose app theme, colors and fonts.
        
        **Technical issues:** If your connection times out, please refresh the page or reopen the app's URL.
        
        For any errors or inquiries, please contact us at info@nlpblogs.com
    ''')




with st.sidebar:
    container = st.container(border=True)
    container.write("**Named Entity Recognition (NER)** is the task of extracting and tagging entities in text data. Entities can be persons, organizations, locations, countries, products, events etc.")
    st.subheader("Related NLP Web Apps", divider="rainbow")
    st.link_button("English HTML Entity Finder", "https://nlpblogs.com/shop/named-entity-recognition-ner/english-html-entity-finder/", type = "primary")

    if 'source_type_attempts' not in st.session_state:
        st.session_state['source_type_attempts'] = 0
max_attempts = 10 

def clear_url_input():
    st.session_state.url = ""

def clear_text_input():
    st.session_state.my_text_area = ""

url = st.text_input("Enter URL from the internet, and then press Enter:", key="url")
st.button("Clear URL", on_click=clear_url_input)
text = st.text_area("Type or paste your text below, and then press Ctrl + Enter", key='my_text_area')
st.button("Clear Text", on_click=clear_text_input)

source_type = None
input_content = None
text_to_process = None

if url:
    source_type = 'url'
    input_content = url
elif text:
    source_type = 'text'
    input_content = text

if source_type:
    start_time = time.time() # Start timer here

    st.subheader("Results", divider = "rainbow")
    
    if st.session_state['source_type_attempts'] >= max_attempts:
        st.error(f"You have requested results {max_attempts} times. You have reached your daily request limit.")
        st.stop()
    
    st.session_state['source_type_attempts'] += 1

    


    
    @st.cache_resource
    def load_ner_model():
        return pipeline("token-classification", model= "ckiplab/bert-base-chinese-ner", aggregation_strategy="max")
    
    model = load_ner_model()
    experiment = None 
    
    try:
        if source_type == 'url':
            if not url.startswith(("http://", "https://")):
                st.error("Please enter a valid URL starting with 'http://' or 'https://'.")
            else:
                with st.spinner(f"Fetching and parsing content from **{url}**...", show_time=True):
                    f = requests.get(url, timeout=10)
                    f.raise_for_status()  # Raise an HTTPError for bad responses (4xx or 5xx)
                    soup = BeautifulSoup(f.text, 'html.parser')
                    text_to_process = soup.get_text(separator=' ', strip=True)
                    st.divider()
                    st.write("**Input text content**")
                    st.write(text_to_process[:500] + "..." if len(text_to_process) > 500 else text_to_process)
                            
        elif source_type == 'text':
            text_to_process = text
            st.divider()
            st.write("**Input text content**")
            st.write(text_to_process[:500] + "..." if len(text_to_process) > 500 else text_to_process)

        if text_to_process and len(text_to_process.strip()) > 0: 
            with st.spinner("Analyzing text...", show_time=True):
                entities = model(text_to_process)
                data = []
                for entity in entities:
                    data.append({
                        'word': entity['word'],
                        'entity_group': entity['entity_group'],
                        'score': entity['score'],
                        'start': entity['start'], # Include start and end for download
                        'end': entity['end']
                    })
                df = pd.DataFrame(data)
                                
                pattern = r'[^\w\s]'
                df['word'] = df['word'].replace(pattern, '', regex=True)
                                
                df = df.replace('', 'Unknown')
                st.dataframe(df)
                                
                if comet_initialized:
                    experiment = Experiment(
                        api_key=COMET_API_KEY,
                        workspace=COMET_WORKSPACE,
                        project_name=COMET_PROJECT_NAME,
                    )
                    experiment.log_parameter("input_source_type", source_type)
                    experiment.log_parameter("input_content_length", len(input_content))
                    experiment.log_table("predicted_entities", df)

                with st.expander("See Glossary of tags"):
                    st.write('''
                    '**word**': ['entity extracted from your text data']
            
                    '**score**': ['accuracy score; how accurately a tag has been assigned to a given entity']
            
                    '**entity_group**': ['label (tag) assigned to a given extracted entity']
            
                    '**start**': ['index of the start of the corresponding entity']
            
                    '**end**': ['index of the end of the corresponding entity']
                    ''')

                
                entity_groups = {"CARDINAL": "cardinal number",                         
                                 "DATE": "date",                         
                                 "EVENT": "event name",                         
                                 "FAC": "facilities",                         
                                 "GPE": "geopolitical entity",                         
                                 "LANGUAGE": "language",                          
                                 "LAW": "law",                         
                                 "LOC": "location",                         
                                 "MONEY": "money",                         
                                 "NORP": "ethnic, religious, political groups",                         
                                 "ORDINAL": "ordinal number",                         
                                 "ORG": "organization",                         
                                 "PERCENT": "percent value",                         
                                 "PERSON": "person",                         
                                 "PRODUCT": "product",                         
                                 "QUANTITY": "quantity",                         
                                 "TIME": "time",                         
                                 "WORK_OF_ART": "work of art",                                                                                               
                                }
                
                st.subheader("Grouped entities", divider = "rainbow")
                                   
                # Convert entity_groups dictionary to a list of (key, title) tuples
                entity_items = list(entity_groups.items())
                    
                # Define how many tabs per row                   
                tabs_per_row = 5
                for i in range(0, len(entity_items), tabs_per_row):
                    current_row_entities = entity_items[i : i + tabs_per_row]
                    tab_titles = [item[1] for item in current_row_entities]
                    tabs = st.tabs(tab_titles)
                    for j, (entity_group_key, tab_title) in enumerate(current_row_entities):
                        with tabs[j]:
                            if entity_group_key in df["entity_group"].unique():
                                df_filtered = df[df["entity_group"] == entity_group_key]
                                st.dataframe(df_filtered, use_container_width=True)
                            else:
                                st.info(f"No '{tab_title}' entities found in the text.")
                                st.dataframe(pd.DataFrame({
                                        'entity_group': [entity_group_key],
                                        'score': [np.nan],
                                        'word': [np.nan],
                                        'start': [np.nan],
                                        'end': [np.nan]
                                    }), hide_index=True)
                                                    
                st.divider()
                                                
                if not df.empty:
                    st.markdown("---")
                    st.subheader("Treemap", divider="rainbow")
                    fig = px.treemap(df, path=[px.Constant("all"), 'entity_group', 'word'],
                                     values='score', color='entity_group',
                                     )
                    fig.update_layout(margin=dict(t=50, l=25, r=25, b=25))
                    st.plotly_chart(fig, use_container_width=True)
                    if comet_initialized and experiment: 
                        experiment.log_figure(figure=fig, figure_name="entity_treemap")
                                                            
                    value_counts = df['entity_group'].value_counts().reset_index()
                    value_counts.columns = ['entity_group', 'count']
                    col1, col2 = st.columns(2)
                    with col1:
                        st.subheader("Pie Chart", divider="rainbow")
                        fig1 = px.pie(value_counts, values='count', names='entity_group',
                                      hover_data=['count'], labels={'count': 'count'},
                                      title='Percentage of Predicted Labels')
                        fig1.update_traces(textposition='inside', textinfo='percent+label')
                        st.plotly_chart(fig1, use_container_width=True)
                        if comet_initialized and experiment: 
                            experiment.log_figure(figure=fig1, figure_name="label_pie_chart")
                    with col2:
                        st.subheader("Bar Chart", divider="rainbow")
                        fig2 = px.bar(value_counts, x="count", y="entity_group", color="entity_group",
                                      text_auto=True, title='Occurrences of Predicted Labels')
                        st.plotly_chart(fig2, use_container_width=True)
                        if comet_initialized and experiment: 
                            experiment.log_figure(figure=fig2, figure_name="label_bar_chart")
                else:
                    st.warning("No entities were extracted from the provided text.")


                dfa = pd.DataFrame(
                data={
                    'Column Name': ['word', 'entity_group','score', 'start', 'end'],
                    'Description': [
                        'entity extracted from your text data',
                        'label (tag) assigned to a given extracted entity',
                        'accuracy score; how accurately a tag has been assigned to a given entity',
                        'index of the start of the corresponding entity',
                        'index of the end of the corresponding entity',]}
                )

                                                

               
                buf = io.BytesIO()
                with zipfile.ZipFile(buf, "w") as myzip:
                    if not df.empty:
                        myzip.writestr("Summary_of_results.csv", df.to_csv(index=False))
                    myzip.writestr("Glossary_of_tags.csv", dfa.to_csv(index=False))
                with stylable_container( 
                     key="download_button", 
                     css_styles="""button { background-color: yellow; border: 1px solid black; padding: 5px; color: black; }""", 
                 ):
                    st.download_button( 
                         label="Download zip file", 
                         data=buf.getvalue(), 
                         file_name="nlpblogs_ner_results.zip", 
                         mime="application/zip",)
                                    
                st.divider()        
        else:
            st.warning("No meaningful text found to process. Please enter a URL or text.")
        
    except Exception as e:
        st.error(f"An unexpected error occurred: {e}")
    finally:
        if comet_initialized and experiment:
            experiment.end()
    
    end_time = time.time() # End timer here
    elapsed_time = end_time - start_time
    st.info(f"Results processed in **{elapsed_time:.2f} seconds**.") # Display elapsed time

st.write(f"Number of times you requested results: **{st.session_state['source_type_attempts']}/{max_attempts}**")