Sample project · CSE
Review sentiment classifier
Type a product or movie review and the model predicts positive or negative, and shows why.
Try it live
A sample our team built to show the standard of a delivered project, not client work. The demo runs the actual trained model in your browser. It was trained on the UCI Sentiment Labelled Sentences dataset (3,000 reviews from Amazon, IMDb and Yelp); the scores shown are its real results on a held-out test set.
Loading the model…
What it solves
Businesses get thousands of reviews and cannot read them all. The project trains a text classifier that labels each review positive or negative and highlights the words behind the decision.
Features
- Text cleaning and tokenisation
- TF-IDF features with word pairs
- Logistic regression classifier
- Prediction with confidence
- Explanation: the words that pushed the score
- Evaluation: accuracy, precision, recall, confusion matrix
How it works
- Labelled reviews
- Clean and tokenise
- TF-IDF vectors
- Train logistic regression
- Evaluate on held-out test set
- Predict and explain
Tech stack
- Python
- scikit-learn
- TF-IDF
- Logistic regression
- Streamlit
Machine learning (NLP) · CSE
Report, slides and viva prep
Every project ships with a full report, a presentation deck and viva preparation. Here is a look at this one.
Contents
Review sentiment classifier
- Abstract
- 1. Introduction
- 2. Literature survey
- 3. Dataset and preprocessing
- 4. Feature extraction (TF-IDF)
- 5. Model and training
- 6. Results and evaluation
- 7. Conclusion and future scope
- References
Review sentiment classifier
Why logistic regression
- Fast to train on a small dataset
- Every word gets a readable weight
- Strong baseline before deep models
Final year project review · Project Squad sample
Viva questions we prepare you for
- What does TF-IDF capture that raw word counts miss?
- Why keep a held-out test set, and how did you split it?
- Where does the model fail, and how would you improve it?
Ready to stop stressing? Start with a free call.
Free 30-minute consult, no commitment. We’ll tell you honestly whether your topic will pass review and what it’ll take to deliver.