Perform binary sentiment analysis (positive / negative) with tweets.
We benchmark 3 approaches:
- Turn-Key Solutions with Microsoft Cognitive Services (Sentiment Analysis);
- Low-code Solutions with Microsoft Azure Machine Learning Studio (Designer);
- Advanced and Custom Solutions with Keras / Tensorflow (RNN/LSTM and Word Embeddings)
Sentiment140 dataset with 1.6 million tweets
- ✔️ Pre-process Tweets;
- ✔️ Use Azure Text Analytics - Sentiment Analysis;
- ✔️ Use AMLS Designer, Logistic Regression model, and 2 text vectorizations;
- ✔️ Use Tensorflow / Keras with RNN / LTSM and word embeddings (from scratch and pre-trained);
- ✔️ Compare all approaches and evaluate performance (AUC, Accuracy);
- ✔️ Deploy the best model for real-time inferencing and publish endpoint;
- ✔️ Write a blog article.
NLTK, Spacy, WordCloud, Azure ML/AI SDK, scikit-learn, Azure Portal, Azure Machine Learning Studio, Tensorflow/Keras, Google Colab (with GPU), Tensorboard, pretrained word embeddings (Word2Vec, GloVE, USE)
- Text analytics Overview;
- Azure Machine Learning Studio, with dedicated Portal and its low-code Designer;
- Tensorflow/Keras : Recurrent Neural Network and Word embeddings; Text Classification with RNN;
- Pre-traind word embeddings: Word2Vec, GloVe - Global Vectors for Word Representation, Universal Sentence Encoder.