Egyptian Tweet Sentiment Analysis , Forecasting and Topic Modeling

The aim of the Bachelor project is to innovate a new way for Arabic (Egyptian-Dialect) Sentiment Analysis, Forecasting, and Topic Modeling using Machine Learning, Deep Learning, and Transformers!

📄View the Thesis »
· Presentation · Demo Video · Report Bug · Be a Contributer

In recent years, social media platforms have become increasingly popular for individuals to express their thoughts and opinions on various topics and situations. Monitoring sentiment and understanding the evolution of topics is crucial for governments to identify negative sentiments and respond promptly. In this study, we developed a sentiment analysis ensemble model architecture consisting of 4 different Transformers namely: MARBERT,CaMel-bert-DA, CaMel-bert-Mix, and Alanzi. This ensemble architecture was followed by a final classification layer, consisting of a three-layer feed-forward neural network. The output of each transformer applied on its logits a formula, and the resulting logits were then summed before applying the softmax activation function. Evaluating the model’s performance on the sentiment analysis test dataset, an impressive test accuracy of 83%. To analyze the temporal trends of sentiment, we applied the LSTM model using a sliding window to time-stamped tweets related to English News, which generated significant discussions on social media. That is then translated to Arabic by a translation model. we plotted the sentiment arc and observed our results with the original results. We explored the effectiveness of BERTopic, a topic modeling technique, in comparison to LDA and NMF techniques. By employing various pre-trained Arabic language models as embeddings, we conducted topic modeling and aspect-based analysis. The results highlight that BERTopic and NMF achieved comparable and competitive outcomes, demonstrating their capability to capture meaningful topics effectively. However, LDA exhibited poor performance in generating coherent and informative topics. These findings emphasize the superiority of BERTopic and NMF over LDA in topic modeling tasks.

• Tweet Collecting & Merging & Pre-processing & Analysis & Correctness Investigation

• Machine Learning Sentiment Analysis

• Deep Learning Sentiment Analysis

• Pure Transformers Sentiment Analysis

• Customized Transformers & Ensemble Model Sentiment Analysis & Website

• Sentiment Forecasting

• Aspect-Based & Topic Modeling

• Zero Shot Classification

• Python
• Data Analysis and Visualization
  • Pandas
  • Matplotlip
  • Seaborn
  • Word Cloud Fa
• Machine Learning
  • Sckitlearn
• Deep Learning
  • Tensorflow
  • Pytorch
• Transformers
  • Hugging Face
• Topic Modeling
  • Gensim
  • Bertopic

(back to top)

Users who will Use this Data should only use it for Practice and not for Commercial Purposes !

(back to top)

(back to top)

Connect with me

Made with ❤️ by Omar Sherif Ali - OSA.

© OSA - 2022

(back to top)

Daily Progress

First Week

Date	Day	Progress	Resources
2023-03-01	Saturday	reading in OReily Sckitlearn book ,revisied Python and searching for extra resources	book
2023-03-01	Sunday	reading in OReily Sckitlearn book and searching for extra resources	book
2023-03-01	Monday	Learned NLP,Tokenization,stemmation,lemmetization,count vectorizer	Udemy course on Python
2023-03-01	Tuesday	Discovered Pandas and done 5 notesbooks for practice on kaggle and finished Data cleaning course in Kaggle	Kaggle
2023-03-01	Wednesday	learned TF-IDF+ notebooks excercise and finished Machine Learning Beginner Kaggle Course
2023-03-02	Thursday	Digging deep in models and their hyperparameters and participating in kaggle competition	Kaggle competition on house prices
2023-03-03	Friday	understand feature engineering and importance, imputers ,Worked on project proposal	Medium article on best practices for ML models

Second Week

Date	Day	Progress	Resources
2023-03-01	Saturday	Understand more models SVMs,KNN and ensemble models XGBoost	Coursera course on Python for Data Science
2023-03-01	Sunday	Finished Kaggle ML interediate course,categorial encoding,and search for more resources	Kaggle
2023-03-01	Monday	ROC,AUC,Conersion matrix,Covariance matrix
2023-03-01	Tuesday	Learned about standardization and done a project on all previously learned models	Coursera course on Python for Data Science
2023-03-01	Wednesday	Activation functions and Sentiment analysis
2023-03-02	Thursday	Model Interpretation(Model-agnostics),text summarization,random walk
2023-03-03	Friday