The Social Media Sentiment Analysis is a machine learning-based tool designed to classify text data into three categories: positive, neutral, and negative sentiment. This project combines Natural Language Processing (NLP) techniques with machine learning algorithms to analyze and understand the sentiments conveyed in text data. It is built using Python, spaCy, scikit-learn, and Streamlit for a highly interactive and visually appealing web application.

The primary goal of this project is to offer a scalable solution for businesses that need to analyze customer feedback, social media data, reviews, and survey responses. By understanding customer sentiments, companies can better tailor their strategies, make data-driven decisions, and improve overall customer experience.

Sentiment analysis is a powerful tool for businesses across multiple industries. This application can be particularly useful for the following purposes:

• Companies can use sentiment analysis to compare their brand's public perception with that of their competitors.
• Analyzing reviews and social media mentions of competing products can provide insights into areas of strength and weaknesses, helping companies differentiate their offerings.

• Marketing teams can assess the success of their campaigns by analyzing customer feedback and social media posts.
• This app allows companies to measure sentiment trends before and after a campaign launch to understand customer reactions in real-time.

• By analyzing reviews and customer feedback, product teams can identify areas for improvement.
• Positive sentiment can highlight features that customers love, while negative sentiment can pinpoint pain points or issues that need attention.

• Sentiment analysis can be applied to customer service interactions, emails, or survey responses to gauge customer satisfaction.
• By identifying and addressing negative sentiments early, companies can improve customer retention and brand loyalty.

• Brands can monitor social media platforms in real-time to track public sentiment regarding their products, services, or company as a whole.
• This helps in quickly responding to customer concerns or capitalizing on positive trends.

• Preprocessing: Text data is preprocessed using spaCy, where it undergoes cleaning (removing stopwords, punctuation, etc.), tokenization, and lemmatization to ensure that the text is in its simplest and most informative form.

• Feature Extraction: The TF-IDF vectorizer is employed to convert the cleaned and tokenized text into numerical values (features) that can be fed into a machine learning model. This approach emphasizes words that are more unique and relevant within the dataset.

• Model: The Logistic Regression model is used for classifying the sentiment of the text. The model is trained using a dataset of customer feedback labeled with positive, neutral, and negative sentiments.

• Evaluation Metrics:

  • Confusion Matrix: Used to visualize the true vs. predicted values, helping in understanding the performance of the model.
  • Classification Report: Includes precision, recall, F1-score, and accuracy metrics for a more detailed evaluation.

• Interactive Dashboard: The web app is built using Streamlit, allowing users to input text and receive instant sentiment predictions. It also provides real-time visualization of model performance through confusion matrices and key evaluation metrics.

• Programming Language: Python
• Libraries:
  • spaCy: For text preprocessing (tokenization, lemmatization, and stopword removal)
  • scikit-learn: For machine learning (TF-IDF vectorization, Logistic Regression model)
  • matplotlib & seaborn: For visualizing the confusion matrix
  • Streamlit: For building the interactive web application
  • pickle: For saving/loading models and vectorizers
• Data: The dataset contains 41,642 customer reviews across three sentiment categories: positive, neutral, and negative.

The dataset used for this project includes customer feedback that has been labeled with one of three possible sentiment labels:

1. Positive (Label: 2)
2. Neutral (Label: 1)
3. Negative (Label: 0)

Each data entry has the following columns:

• id: A unique identifier for each review.
• text: The actual text of the customer feedback.
• target: The numerical label for sentiment (0 for negative, 1 for neutral, 2 for positive).
• sentiment: The sentiment as a string (positive, neutral, or negative).

Sample Data:

The full dataset contains more than 40,000 rows, offering a substantial amount of data for training and testing the sentiment analysis model.