Hospital Service Quality Aspect based Sentiment Analysis

1. Introduction and Background

Sentiment analysis involves the extraction of opinions, emotions, and attitudes related to specific aspects, which, in our case, is assessing the quality of service in hospitals from a patient's perspective. We aim to analyze user sentiments within specific aspects mentioned in patient reviews, providing a fine-grained and nuanced approach.

Objective:

• Assess the overall quality of a hospital from a patient's viewpoint by extracting fine-grained sentiment expressions.

2. Literature Survey

2.1 Web Scraping

• Automated extraction of data from websites using web scraping techniques.

2.2 Text Pre-processing

• Essential NLP steps like tokenization, lowercasing, and removal of special characters, enhancing sentiment analysis.

2.3 Automated Data Annotation

• Utilization of advanced pre-trained models like BERT for intelligent data annotation.

2.4 Model Training Evaluation

• Using dedicated test datasets to assess model performance, employing metrics like Precision, Recall, and F1 Score.

2.5 KeyBERT

• Utilizing BERT embeddings for keyword extraction within a sentence or document.

3. Problem Definition and Objective

• Implementing automated web scraping to gather hospital reviews for NLP analysis.
• Applying NLP techniques for text pre-processing, aspect-level data annotation, model development, and training.
• Creation of annotated training datasets using KeyBERT-based pre-trained model checkpoints.
• Comparative analysis of annotation methods for effective sentiment analysis.

4. Methodology

4.1 Data Collection and Preparation

• Employing Python-based web scraping on MouthShut.com, resulting in a dataset of 56,080 reviews for analysis.

4.2 Text Cleaning for Aspect Extraction

• A detailed process involving the removal of HTML tags, special characters, and stop words for enhancing the dataset quality.

4.3 Data Annotation for Aspect-Based Sentiment Analysis

• Automatic annotation methods leveraging advanced models like E2EABSA and KeyBERT.

4.4 Model Development for Aspect-Sentiment Analysis

• Usage of a powerful text generative base model to derive aspects and their corresponding sentiments.

4.5 Model Training

• Fine-tuned training of the model on 'english,' 'multilingual,' and 'KeyBERT' datasets.

4.6 Model Evaluation

• Comprehensive evaluation using precision, recall, and F1 scores on different datasets.

4.7 Demo Test Result

4.7 Results on Training, Validation, and Test Data

• Detailed evaluation results on training, validation, and test data.

4.8 Comparative Assessment

• Comparison of different trained models to understand their performance differences.