The brief
Atithi is a full-stack machine-learning platform designed to analyse hotel reviews and turn unstructured guest feedback into actionable customer-experience insights.
Hotel reviews contain far more information than a simple rating or positive/negative label. A guest may praise the staff while criticising the room, appreciate the location while being disappointed with breakfast, or describe a generally positive stay with one significant complaint.
Atithi was designed to capture these nuances.
The platform combines natural-language processing, ensemble machine learning, aspect-based sentiment analysis, and explainable AI to understand both how guests feel and what parts of their experience drive that sentiment.
From reviews to intelligence
Traditional sentiment analysis often reduces a review to a single classification such as positive, negative, or neutral.
That can hide the information a hotel actually needs.
Consider a review such as:
“The staff were extremely helpful and the location was excellent, but the room was small and breakfast was disappointing.”
A single sentiment label loses the distinction between these experiences.
Atithi analyses the review at a more granular level, allowing different aspects of the same review to carry different sentiment.
For example:
Staff → Positive
Location → Positive
Room → Negative
Breakfast → Negative
This creates a richer representation of the guest experience.
NLP preprocessing
Hotel reviews are unstructured natural-language data.
Guests use different vocabulary, sentence structures, informal language, and mixed sentiments. A useful classification pipeline therefore needs to prepare the text before it reaches the machine-learning models.
Atithi uses spaCy-based NLP preprocessing to transform raw review text into representations suitable for downstream classification and analysis.
This creates a consistent language-processing layer between the original review and the machine-learning pipeline.
Ensemble sentiment classification
At the core of Atithi is a Voting Classifier ensemble.
Instead of depending on a single classifier, multiple machine-learning models contribute to the final sentiment prediction.
The ensemble combines the outputs of these models into a final classification, creating a more robust prediction layer for the varied language found in real-world hotel reviews.
The workflow becomes:
Review Text → NLP Processing → Multiple Classifiers → Voting Ensemble → Sentiment
Aspect-Based Sentiment Analysis
The most useful layer of Atithi is its aspect-level analysis.
Instead of asking only:
“Was this review positive?”
the system can investigate:
“Which parts of the hotel experience were positive or negative?”
Relevant aspects can include rooms, staff, cleanliness, location, food, service, amenities, and other elements mentioned in a review.
The result is a more actionable view of customer sentiment.
Across a large collection of reviews, these aspect-level signals can help identify recurring strengths and weaknesses in the guest experience.
Explainable AI
A machine-learning prediction is more useful when users can understand why it was made.
Atithi incorporates LIME-based explainability to provide an interpretation of the model’s sentiment predictions.
Instead of presenting sentiment as an unexplained black-box output, the system can expose which parts of the review contributed to the classification.
This makes the analysis easier to inspect and provides users with more context around the model’s decisions.
Real-time analysis
Atithi exposes the machine-learning pipeline through a Flask backend, allowing review analysis to be performed through an application interface.
A review can move through the complete pipeline:
Input → NLP → Classification → Aspect Analysis → Explanation → Result
The React frontend provides the user-facing layer for interacting with the analysis system.
This turns the underlying ML models into a usable application rather than leaving them as isolated experimentation code.
Understanding the guest experience
The system can help hotel teams investigate questions such as:
- What do guests like most about the hotel?
- Which aspects consistently receive negative feedback?
- What is driving positive or negative sentiment?
- Are complaints concentrated around a particular part of the guest experience?
- Which areas appear strongest across reviews?
- Which areas may require operational attention?
Instead of manually reading every review to discover these patterns, teams can use automated analysis to surface and investigate them.
From individual reviews to broader patterns
An individual review provides one perspective.
A collection of reviews provides a much broader view of customer experience.
Atithi’s aspect-based approach makes it possible to move from individual review classification toward aggregated customer intelligence.
For example, repeated negative sentiment around rooms may indicate a broader issue, while consistently positive sentiment around staff can reveal a clear strength in the guest experience.
This makes the platform useful not only for analysing individual reviews, but also for identifying recurring patterns across customer feedback.
A complete ML product
Atithi combines several layers into one end-to-end application:
Natural Language
↓
NLP Preprocessing
↓
Ensemble Classification
↓
Aspect-Based Analysis
↓
Explainability
↓
Business Insight
The machine-learning components are served through a backend API and presented through a dedicated frontend, creating a complete path from raw customer feedback to an interpretable result.
More than sentiment classification
The central idea behind Atithi is that customer feedback should not be reduced to a single number.
A review is a description of an experience.
By combining sentiment classification with aspect-level analysis and explainability, Atithi provides a richer understanding of what customers are saying and why the system interprets their feedback in a particular way.
The idea behind Atithi
Hotel reviews are a continuous stream of customer feedback.
The challenge is turning that feedback into something teams can understand and act on.
Atithi transforms unstructured guest experiences into structured, interpretable intelligence.
Understand the sentiment. Find the source. Improve the experience.
Project highlights
- Real-time hotel review sentiment analysis
- spaCy-based NLP preprocessing
- Voting Classifier ensemble
- Aspect-Based Sentiment Analysis
- LIME-based explainable AI
- Guest experience analysis
- Multi-aspect review interpretation
- Flask-based ML backend
- React-based frontend
- End-to-end ML application architecture
- Individual review analysis
- Foundation for aggregated customer-experience intelligence