Alexa Sentiment-Analysis

---

Overview:

---

The objective of this project is to build sentiment analyzer to predict weather the review falls under positive sentiment or negative.

DataSet:

---

• Source: https://www.kaggle.com/sid321axn/amazon-alexa-reviews/data
• Observations: 3150
• Features: 6

Data Visualization:

---

• Its seems that positive class is more than the negative class, which indicate imbalance of class.

• Moreover there are more people has provided rating of 5 star to Alexa product which indicate that customer is satosfied with the product.

• Wordcloud: All words, Positive and Negative

Model Building:

---

Machine Learning Model:

Model	Accuracy Score	AUC ROC Score
Naive Bayes + BOW	93 %	60 %
Naive Bayes + TFIDF	91 %	50 %
SVM + BOW	93 %	75 %
SVM + TFIDF	93 %	61 %

Deep Learning Model:

---

Model	Accuracy Score	AUC ROC Score
BiLSTM	94 %	89 %
Multi Conv Net Architecture	95 %	94 %

Sequential Model

Functional API

Streamlit Deployment:

---

• Model Deplayment was done using streamlit.

• If you've entred nothing.
• Positive Sentiment.
• Negative Sentiment

---

Thank You!

Follow @ShehzadaAlam