Building a Sentiment Analysis App with Streamlit and Hugging Face Transformers

In this tutorial, we will create a sentiment analysis app using Streamlit and Hugging Face's Transformers library. This app will classify the sentiment of a given text as positive, neutral, or negative.

Prerequisites

Before we start, make sure you have Python installed on your system. You also need to install the following libraries:

• transformers
• numpy
• scipy
• streamlit

You can install these libraries using pip:

pip install transformers numpy scipy streamlit

Step 1: Importing the Required Libraries

First, we need to import the required libraries and modules:

import streamlit as st
from transformers import AutoModelForSequenceClassification, AutoTokenizer, AutoConfig
import numpy as np
from scipy.special import softmax

Step 2: Loading the Pre-trained Model and Tokenizer

We will use a pre-trained model from the Hugging Face model hub. In this example, we are using the cardiffnlp/twitter-roberta-base-sentiment-latest model.

model_path = "cardiffnlp/twitter-roberta-base-sentiment-latest"

tokenizer = AutoTokenizer.from_pretrained(model_path)
config = AutoConfig.from_pretrained(model_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path)

Step 3: Defining Example Options

We will provide some example text options for the user to choose from:

options = {
    "Option 1 - I love this! It's amazing.": "I love this! It's amazing.",
    "Option 2 - Neutral text here.": "Neutral text here.",
    "Option 3 - I hate this, it's terrible.": "I hate this, it's terrible."
}

Step 4: Defining the Sentiment Analysis Function

Next, we define a function that takes a text input, processes it using the tokenizer and the model, and returns the sentiment scores.

def sentiment_analysis(text):
    encoded_input = tokenizer(text, return_tensors='pt')
    output = model(**encoded_input)
    scores_ = output.logits.detach().numpy()
    scores_ = softmax(scores_)

    labels = ['Negative', 'Neutral', 'Positive']
    scores = {l: float(s) for (l, s) in zip(labels, scores_[0])}
    return scores

Step 5: Creating the Streamlit Interface

Now, we create the Streamlit interface for our sentiment analysis function. This interface includes a text input, a button to trigger the analysis, and columns to display the results.

st.title('Sentiment Analysis app')
st.write('This app classifies text as having a positive, neutral, or negative sentiment.')

option = st.selectbox(
    "Choose an option to fill the text input:",
    options.keys()
)

input_text = st.text_area(
    "or write to the input text",
    value=options[option]
)

result = {
    "Negative": 0,
    "Neutral": 0,
    "Positive": 0
}

if st.button("Analyze"):
    result = sentiment_analysis(input_text)

col1, col2, col3 = st.columns(3)
col1.metric("Negative", str(round(result["Negative"] * 100, 2)) + '%', "")
col2.metric("Neutral", str(round(result["Neutral"] * 100, 2)) + '%', "")
col3.metric("Positive", str(round(result["Positive"] * 100, 2)) + '%', "")

Step 6: Using cnvrg to launch the App

1. Click Apps in your project.
2. Select Generic App.
3. For Command, type the command required to run your app. streamlit run streamlit-app.py --server.port 9200
4. For Image, select a relevant Docker image or custom image that suits your application.
5. For Compute, select the appropriate compute resources to run your application.
6. Click Publish. cnvrg will deploy your application based on the provided configuration.

Here is the complete code for the sentiment analysis app:

import streamlit as st
from transformers import AutoModelForSequenceClassification, AutoTokenizer, AutoConfig
import numpy as np
from scipy.special import softmax

model_path = "cardiffnlp/twitter-roberta-base-sentiment-latest"

tokenizer = AutoTokenizer.from_pretrained(model_path)
config = AutoConfig.from_pretrained(model_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path)

options = {
    "Option 1 - I love this! It's amazing.": "I love this! It's amazing.",
    "Option 2 - Neutral text here.": "Neutral text here.",
    "Option 3 - I hate this, it's terrible.": "I hate this, it's terrible."
}

def sentiment_analysis(text):
    encoded_input = tokenizer(text, return_tensors='pt')
    output = model(**encoded_input)
    scores_ = output.logits.detach().numpy()
    scores_ = softmax(scores_)

    labels = ['Negative', 'Neutral', 'Positive']
    scores = {l: float(s) for (l, s) in zip(labels, scores_[0])}
    return scores

st.title('Sentiment Analysis app')
st.write('This app classifies text as having a positive, neutral, or negative sentiment.')

option = st.selectbox(
    "Choose an option to fill the text input:",
    options.keys()
)

input_text = st.text_area(
    "or write to the input text",
    value=options[option]
)

result = {
    "Negative": 0,
    "Neutral": 0,
    "Positive": 0
}

if st.button("Analyze"):
    result = sentiment_analysis(input_text)

col1, col2, col3 = st.columns(3)
col1.metric("Negative", str(round(result["Negative"] * 100, 2)) + '%', "")
col2.metric("Neutral", str(round(result["Neutral"] * 100, 2)) + '%', "")
col3.metric("Positive", str(round(result["Positive"] * 100, 2)) + '%', "")