app.py

import matplotlib as mpl
import pandas as pd
import plotly.express as px
import plotly.graph_objects as go
import numpy as np
import re
import streamlit as st

from io import StringIO, BytesIO
from transformers import pipeline

try:
    st.set_page_config(layout="wide")
except:
    st.beta_set_page_config(layout="wide")

st.title("Sentiment Structure Visualizer")

user_input = st.text_area(
    "Paste English text here",
    height=200,
)

st.markdown(
    """
----------------------------
"""
)

sentiment = pipeline(
    "sentiment-analysis",
    model="distilbert-base-uncased-finetuned-sst-2-english",
    tokenizer="distilbert-base-uncased-finetuned-sst-2-english",
)

def clean_text(text):
    text = text.encode("ascii", errors="ignore").decode(
        "ascii"
    )  # remove non-ascii, Chinese characters
    text = text.lower()
    text = re.sub(r"\n", " ", text)
    text = re.sub(r"\n\n", " ", text)
    text = re.sub(r"\t", " ", text)
    text = text.strip(" ")
    text = re.sub(r"[^\w\s]", "", text)  # remove punctuation and special characters
    text = re.sub(
        " +", " ", text
    ).strip()  # get rid of multiple spaces and replace with a single
    return text

if user_input != "":
    with st.spinner("..."):
        input_text = (
            pd.DataFrame(user_input.split("."))
            .stack()
            .reset_index()
            .rename(columns={0: "Paras"})
            .drop("level_0", axis=1)
            .drop("level_1", axis=1)
        )

        input_text["Clean_Text"] = input_text["Paras"].map(
            lambda text: clean_text(text)
        )

        corpus = list(input_text["Clean_Text"].values)

        input_text["Sentiment"] = sentiment(corpus)

        input_text["Sentiment_Label"] = [
            x.get("label") for x in input_text["Sentiment"]
        ]

        input_text["Sentiment_Score"] = [
            x.get("score") for x in input_text["Sentiment"]
        ]

        cols = ["Paras", "Sentiment_Label", "Sentiment_Score"]
        df = input_text[cols].copy()

        df = df[df["Paras"].str.strip().astype(bool)]

        df["Sentiment_Score"] = np.where(
            df["Sentiment_Label"] == "NEGATIVE",
            -(df["Sentiment_Score"]),
            df["Sentiment_Score"],
        )

        df["Sentiment_Score"] = df["Sentiment_Score"].round(6)

        overall_sentiment_score = df["Sentiment_Score"].sum().round(3)

        sentiment_count = df["Sentiment_Label"].value_counts().to_string()

        fig = go.Figure(
            data=go.Heatmap(
                z=df["Sentiment_Score"],
                x=df.index,
                y=df["Sentiment_Label"],
                colorscale=px.colors.sequential.RdBu,
            )
        )

        fig.update_layout(
            title=go.layout.Title(text="Sentiment Sequence, By Sentence"), autosize=True
        )

        fig.update_layout(yaxis_autorange="reversed")

        st.plotly_chart(fig, use_container_width=True)

        buffer = StringIO()
        fig.write_html(buffer, include_plotlyjs="cdn")
        html_bytes = buffer.getvalue().encode()

        st.download_button(
            label="Download Interactive Chart",
            data=html_bytes,
            file_name="chart.html",
            mime="text/html",
        )

col1, col2 = st.columns([1,3])

with col1:
    st.metric(
        "Overall Sentiment Score",
        overall_sentiment_score,
        delta=None,
        delta_color="normal",
    )

with col2:
    st.metric(
        "How Many Positive & Negative Sentences?",
        sentiment_count,
        delta=None,
        delta_color="normal",
    )


st.markdown(
    """
----------------------------
"""
)

st.subheader("Note To Users:")

st.write("1. The model under the hood is distilbert-base-uncased-finetuned-sst-2-english. Clone this app and switch to another transformer model if you have a different use case.")

st.write("2. This chart is interactive, and can be downloaded. Hover over the bars to see each sentence's sentiment score and label")

st.write(
    "3. You may or may not agree with the sentiment label generated for each sentence. Unfortunately there's no way to amend the output within the app."
)