Karmenia Ditabaya Lontoh's Portofolio

Introduction

Customer reviews are a key factor in the hospitality industry, influencing potential guests' decisions and shaping the reputation of hotels. This project aims to develop a deep learning model to classify hotel reviews based on sentiment, categorizing them into negative, neutral, or positive feedback. By analyzing these reviews, hotel managers can gain valuable insights into guest satisfaction and improve customer service.

Dataset Overview

The dataset for this project is sourced from TripAdvisor, a platform where travelers share their experiences. It consists of two main features:

Review: The text of the customer's feedback.
Rating: A numerical rating from 1 to 5 indicating customer satisfaction.

The ratings are divided into three sentiment categories:

Negative (1-2)
Neutral (3)
Positive (4-5)

Dataset URL: TripAdvisor Hotel Reviews

Methodology

Data Loading and Inspection

Loading and inspecting the dataset to ensure data quality and completeness.

Exploratory Data Analysis (EDA)

Exploring patterns and trends; distribution of ratings and sentiment breakdown.

Feature Engineering

Prepare the review text; text preprocessing including tokenization, stopword removal, and lemmatization.

Model Training

Training a sentiment analysis model using ANN and enhancing it through transfer learning and hyperparameter tuning.

Model Evaluation

Evaluating model performance using metrics such as accuracy, F1-score, and loss functions.

Model Optimization

Refining the model through adjustments in the architecture, hyperparameters, and training process.

Models Employed

Artificial Neural Networks (ANN)

ANN is employed as the primary model for sentiment classification, using layers to learn complex patterns in the review text.

Transfer Learning

Leveraging pre-trained models for better feature extraction and faster training.

Model Analysis

Strengths

The model demonstrated high accuracy (92.69%) and consistent training/validation loss reduction, suggesting a well-learned model.

Weaknesses

The model struggles with neutral reviews, reflected by lower F1-scores for neutral sentiment classification.

Improvements

Adding more training data, fine-tuning hyperparameters, and addressing class imbalance could further improve model performance.

Conclusion

The sentiment analysis model successfully categorizes hotel reviews into negative, neutral, and positive sentiments, providing actionable insights for hotel management. Despite the high accuracy, addressing challenges with neutral sentiment classification and overfitting would improve overall performance.

Model Deployment

The model is deployed on Hugging Face, allowing hotel managers to analyze customer feedback in real-time and make data-driven decisions to improve service quality.

Explore the deployed model: Hotel Review Sentiment Analysis on Hugging Face