shrut27/ESG_Report_Analysis

1import streamlit as st2from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline3import spacy4from tika import parser5import requests6import pandas as pd7 8# Loading spaCy model outside the streamlit cache9nlp = spacy.load("en_core_web_sm")10 11@st.cache_resource()12def load_environmental_model():13    name_env = "ESGBERT/EnvironmentalBERT-environmental"14    tokenizer_env = AutoTokenizer.from_pretrained(name_env)15    model_env = AutoModelForSequenceClassification.from_pretrained(name_env)16    return pipeline("text-classification", model=model_env, tokenizer=tokenizer_env)17 18@st.cache_resource()19def load_social_model():20    name_soc = "ESGBERT/SocialBERT-social"21    tokenizer_soc = AutoTokenizer.from_pretrained(name_soc)22    model_soc = AutoModelForSequenceClassification.from_pretrained(name_soc)23    return pipeline("text-classification", model=model_soc, tokenizer=tokenizer_soc)24 25@st.cache_resource()26def load_governance_model():27    name_gov = "ESGBERT/GovernanceBERT-governance"28    tokenizer_gov = AutoTokenizer.from_pretrained(name_gov)29    model_gov = AutoModelForSequenceClassification.from_pretrained(name_gov)30    return pipeline("text-classification", model=model_gov, tokenizer=tokenizer_gov)31 32@st.cache_resource()33def load_sentiment_model():34    model_name = "climatebert/distilroberta-base-climate-sentiment"35    model = AutoModelForSequenceClassification.from_pretrained(model_name)36    tokenizer = AutoTokenizer.from_pretrained(model_name, max_len=512)37    return pipeline("text-classification", model=model, tokenizer=tokenizer)38 39# Streamlit App40st.title("ESG Report Classification using Natural Language Processing")41 42# Get report URL from user input43url = st.text_input("Enter the URL of the report (PDF):")44 45# Model selection dropdown46st.write("Environmental Model, Social Model, Governance Model would give the percentage denoting the parameter chosen.")47st.write("Sentiment Model shows if the company is a risk or opportunity based on all 3 parameters.")48selected_model = st.selectbox("Select Model", ["Environmental Model", "Social Model", "Governance Model", "Sentiment Model"])49 50if url:51    # Download PDF content from the URL52    response = requests.get(url, stream=True)53 54    if response.status_code == 200:55        # Parse PDF and extract text56        raw_text = parser.from_buffer(response.content)['content']57        # Extract sentences using spaCy58        doc = nlp(raw_text)59        sentences = [sent.text for sent in doc.sents]60        # Filtering and preprocessing sentences61        sequences = list(map(str, sentences))62        sentences = [x.replace("\n", "") for x in sequences]63        sentences = [x for x in sentences if x != ""]64        sentences = [x for x in sentences if x[0].isupper()]65        sub_sentences = sentences[:100]  66        # Classification using different models based on user selection67        if selected_model == "Environmental Model":68            pipe_model = load_environmental_model()69        elif selected_model == "Social Model":70            pipe_model = load_social_model()71        elif selected_model == "Governance Model":72            pipe_model = load_governance_model()73        else:74            pipe_model = load_sentiment_model()75 76        # Get predictions for the selected model77        model_results = pipe_model(sub_sentences, padding=True, truncation=True)78        model_labels = [x["label"] for x in model_results]79 80        # Display count of sentences labeled as the selected model81        st.subheader(f"{selected_model} Sentences Count")82        st.write(pd.DataFrame({"sentence": sub_sentences, selected_model: model_labels}).groupby(selected_model).count())83 84    else:85        st.error("Error fetching PDF content from the provided URL. Please check the URL and try again.")86