Demo.py

import streamlit as st
import sparknlp
import os
import pandas as pd

from sparknlp.base import *
from sparknlp.annotator import *
from pyspark.ml import Pipeline
from sparknlp.pretrained import PretrainedPipeline

# Page configuration
st.set_page_config(
    layout="wide", 
    page_title="Spark NLP Financial Sentiment Analysis", 
    initial_sidebar_state="expanded"
)

# CSS for styling
st.markdown("""
    <style>
        .main-title {
            font-size: 36px;
            color: #4A90E2;
            font-weight: bold;
            text-align: center;
        }
        .section p, .section ul {
            color: #666666;
        }
        .result-positive {
            color: green;
        }
        .result-negative {
            color: red;
        }
        .result-neutral {
            color: #209DDC;
        }
    </style>
""", unsafe_allow_html=True)

@st.cache_resource
def init_spark():
    return sparknlp.start()

@st.cache_resource
def create_pipeline(model):
  document = DocumentAssembler()\
    .setInputCol("text")\
    .setOutputCol("document")

  embeddings = BertSentenceEmbeddings\
      .pretrained('sent_bert_wiki_books_sst2', 'en') \
      .setInputCols(["document"])\
      .setOutputCol("sentence_embeddings")

  sentimentClassifier = ClassifierDLModel.pretrained("classifierdl_bertwiki_finance_sentiment", "en") \
    .setInputCols(["sentence_embeddings"]) \
    .setOutputCol("class_")

  financial_sentiment_pipeline = Pipeline(
      stages=[document, 
              embeddings, 
              sentimentClassifier])
  
  return financial_sentiment_pipeline

def fit_data(pipeline, data):
    empty_df = spark.createDataFrame([['']]).toDF('text')
    pipeline_model = pipeline.fit(empty_df)
    model = LightPipeline(pipeline_model)
    results = model.fullAnnotate(data)[0]

    return results['class_'][0].result

# Set up the page layout
st.markdown('<div class="main-title">Sentiment Analysis of Financial News with Spark NLP</div>', unsafe_allow_html=True)

# Sidebar content
model = st.sidebar.selectbox(
    "Choose the pretrained model",
    ["classifierdl_bertwiki_finance_sentiment"],
    help="For more info about the models visit: https://sparknlp.org/models"
)

# Reference notebook link in sidebar
colab_link = """
<a href="https://colab.research.google.com/github/JohnSnowLabs/spark-nlp-workshop/blob/master/tutorials/streamlit_notebooks/SENTIMENT_EN_FINANCE.ipynb">
    <img src="https://colab.research.google.com/assets/colab-badge.svg" style="zoom: 1.3" alt="Open In Colab"/>
</a>
"""
st.sidebar.markdown('Reference notebook:')
st.sidebar.markdown(colab_link, unsafe_allow_html=True)

# Load examples
examples = [
    "In April 2005, Neste separated from its parent company, Finnish energy company Fortum, and became listed on the Helsinki Stock Exchange.",
    "Finnish IT solutions provider Affecto Oyj HEL: AFE1V said today it slipped to a net loss of EUR 115,000 USD 152,000 in the second quarter of 2010 from a profit of EUR 845,000 in the corresponding period a year earlier.",
    "10 February 2011 - Finnish media company Sanoma Oyj HEL: SAA1V said yesterday its 2010 net profit almost tripled to EUR297.3m from EUR107.1m for 2009 and announced a proposal for a raised payout.",
    "Profit before taxes decreased by 9% to EUR 187.8 mn in the first nine months of 2008, compared to EUR 207.1 mn a year earlier.",
    "The world's second largest stainless steel maker said net profit in the three-month period until Dec. 31 surged to euro603 million US$ 781 million, or euro3.33 US$ 4.31 per share, from euro172 million, or euro0.94 per share, the previous year.",
    "TietoEnator signed an agreement to acquire Indian research and development (R&D) services provider and turnkey software solutions developer Fortuna Technologies Pvt. Ltd. for 21 mln euro ($30.3 mln) in September 2007."
]

# User input selection
selected_text = st.selectbox("Select a sample", examples)
custom_input = st.text_input("Try it for yourself!")

# Use custom input if provided
if custom_input:
    selected_text = custom_input

st.subheader('Selected Text')
st.write(selected_text)

# Initialize Spark and create pipeline
spark = init_spark()
pipeline = create_pipeline(model)
output = fit_data(pipeline, selected_text)

# Display output sentiment
if output.lower() in ['pos', 'positive']:
    st.markdown(f"""<h3>This seems like <span class="result-positive">positive</span> news. <span style="font-size:35px;">&#128515;</span></h3>""", unsafe_allow_html=True)
elif output.lower() in ['neg', 'negative']:
    st.markdown(f"""<h3>This seems like <span class="result-negative">negative</span> news. <span style="font-size:35px;">&#128544;</span></h3>""", unsafe_allow_html=True)
else:
    st.markdown(f"""<h3>This seems like <span class="result-neutral">{output}</span> news. <span style="font-size:35px;">&#128578;</span></h3>""", unsafe_allow_html=True)