Sentiment-Analysis

Description

The Social Media Emotions Analysis Tool is an application that allows users to input a sentence or keyword and receive real-time analysis of sentiment and emotions associated with it across social media platforms. This tool enhances decision-making processes, improves marketing strategies, and provides actionable insights based on sentiment and emotional responses across various social media platforms. In current development there is a feature of the input being measured based on a chosen social media platform and data from last 7 days.

This project explores emotion detection along with text preprocessing and feature engineering methods using BERT model. It is end-to-end pipeline that provides insights about any topic or product through real-time tweets sentiment analysis and emotion detection. Final project for the Data Science and Machine Learning bootcamp at IronHack.

Currently pulling real-time data from Twitter, and predicting emotions to generate insights in a local file with the goal to move it to the app.

Files

The project consists of the following files:

• Tableau Dashboard: On overview of data across Facebook,Instagram and Twitter from sentimentdataset.csv

• model_code.py: Contains functions for loading and saving model, vectorizer, data, and performing predictions.

• predictions_json.py: Code runs the model, input your json file with text to make predictions.

• ML_codes_all.ipynb:Codes for different trained models, transformers, label_encoders.

• requirements.txt: Lists all the packages required for the project.

• datacleaning.sql: Data cleaning, preprocessing and saving. Very similar to datapreparation.py which I also included.

• data: A data folder containing datasets used for model training.

  • You can use val_data folder to test your code as well, the data there is not in the combined_emotions.csv.

  • combined_emotions.csv: dataset used for model training.

  • The file the combined_emotions.csv contains approx 2.1 mln online posts and assosciated emotions with them.

  • Data was downloaded from the below links, as it is too big I am not adding it to the repo, you can find the way to combine and criate the training .csv in datapreparation.py :

    • https://www.kaggle.com/datasets/kashishparmar02/social-media-sentiments-analysis-dataset
    • https://www.kaggle.com/datasets/kazanova/sentiment140
    • https://www.kaggle.com/datasets/parulpandey/emotion-dataset
    • https://www.kaggle.com/datasets/bhavikjikadara/emotions-dataset
    • https://huggingface.co/datasets/dair-ai/emotion
    • https://www.kaggle.com/datasets/chandrug/textemotiondetection?select=val.txt
    • https://github.com/SannketNikam/Emotion-Detection-in-Text/blob/main/data/emotion_dataset.csv
    • Pre-labeled sentiment dataset Sentiment140

Getting Started

To run the project, follow the steps below:

1. Clone the repository:

   git clone https://github.com/adrianna-d/Sentiment-Analysis.git

2. Optional but encougraged: Create a virtual environment for the project:

   python -m venv venv
   venv/Scripts/activate # or source venv/bin/activate

3. Install the required packages listed in requirements.txt:

   pip install -r requirements.txt

4. Optional, if using APIs save your credentials in a safe place in a .yaml or .cfg file ;)

5. Run the model_code.py to create your model, tokenizer, label_encoder.

6. Runn the predictions_json.py with your json file to precidt emotions of taxt from the file.

7. Run the app.py to with predictions locally

Project overview: Methodology and Approach

Project Pipeline

• Data Collection: Collect datasets from kaggle. Collect tweets and posts from Twitter and Reddit using respective APIs.
• Data Pre-processing: Clean and preprocess the data, tokenize using BERT tokenizer.
• Model Training: Train the BERT model on the preprocessed data for emotion detection.
• Model Evaluation: Evaluate the model performance using appropriate metrics.
• Real-time Data Fetching: Fetch real-time data from Twitter and Reddit.
• Prediction and Reporting: Predict emotions and generate reports.

Data Pre-processing and Feature Engineering

The data pre-processing step includes the following steps, data was preprocessed in SQL before.

• Data Cleaning: Remove Nan values.
• Tokenization: Convert text to token IDs using BERT tokenizer.
• Padding and Truncation: Ensure uniform input length for the BERT model.

Model Training and Evaluation

The model training and evaluation step includes the following steps:

• Train/Test Split: Split the data into training and testing sets.
• Model Training: Train the model using the training set.
• Model Evaluation: Evaluate the model using the testing set.
• Metrics: Use accuracy, precision, recall, F1-score to evaluate the model.
• Confusion Matrix: Generate a confusion matrix to evaluate the model's performance.

Real-time Data Fetching

• Twitter API: Use the Twitter API to fetch real-time tweets (paid)
• Tweepy: Use the tweepy package to fetch tweets from the Twitter API (you stil need to pay to X)
• WebScrapping: use webstrapping like Apify
• Data Pre-processing: Pre-process the fetched tweets using the same steps as well as the same vectorizer as the data pre-processing step.

Prediction

• Prediction: Predict the sentiment and emotions of the fetched comments/text using the trained BERT model.

Logging and Model Management with MLflow

• Experiment Tracking: Track experiments, parameters, and metrics using MLflow.
• Model Logging: Log the trained BERT model, tokenizer, and label encoder with MLflow.
• Artifacts: Save and log artifacts like the classification report, model files, and tokenizer.

References

• Images used for the app were created by rawpixel.com for Freepik.
• Big thanks to all Kaggle users!

License

Distributed under the MIT License. See LICENSE.txt for more information.