Prolog Sentiment Analysis

A sentiment analysis system implemented in Prolog, combining Natural Language Processing techniques with the SentiWordNet lexicon to analyze text sentiment.

Authors: Federico Chiaradia (f.chiaradia5@studenti.uniba.it), Yonas Babulet Abetew (y.babuletabetew@studenti.uniba.it)

Institution: Department of Computer Science, University of Bari

---

📋 Table of Contents

• Introduction
• Features
• System Architecture
• Prerequisites
• Installation
• Usage
  • GUI Version
  • Console Version
• How It Works
  • Processing Pipeline
  • Text Preprocessing
  • Lexical Analysis
  • Score Calculation
• Project Structure
• Examples
• Limitations and Future Improvements
• License

---

🎯 Introduction

Sentiment analysis, also known as opinion mining, is a computational technique that determines the sentiment or emotional tone expressed in text. This project leverages Prolog, a logic programming language, to implement a rule-based sentiment analysis system.

The system classifies input text into three categories:

• Positive (score > 0.3)
• Negative (score < -0.3)
• Neutral (otherwise)

The implementation combines NLP techniques with data from the SentiWordNet lexicon to compute sentiment scores based on user input sentences.

---

✨ Features

• Dual Interface: Access via GUI (JavaFX) or console (SWI-Prolog)
• Rule-Based System: Uses declarative Prolog rules for linguistic pattern matching
• SentiWordNet Integration: Leverages the SentiWordNet v3.0.0 lexicon for sentiment scoring
• Advanced NLP Pipeline: Includes stopword removal, intensifier detection, negation handling, and neighbor influence calculation
• Transparent Processing: Shows intermediate steps for educational purposes
• Error Handling: Provides helpful feedback for invalid inputs

---

🏗️ System Architecture

The system consists of three main modules:

1. GUI (Main.java): JavaFX-based graphical interface for user interaction

2. Prolog Bridge (Prolog.java): Java-Prolog interface using JPL library

3. Sentiment Analyzer (Sent_Analysis.pl)

   : Core Prolog logic with two sub-modules:

   • sent_modifiers.pl: Dictionary of indicators, intensifiers, and negations
   • sent_wordnet.pl: SentiWordNet lexicon database

---

📦 Prerequisites

• SWI-Prolog: Required for running the Prolog engine
• Java: Required for the GUI version (JDK 8 or higher recommended)
• JavaFX SDK: Included with the launch files for the GUI version

---

🚀 Installation

1. Clone this repository:

git clone https://github.com/yourusername/prolog-sentiment-analysis.git
cd prolog-sentiment-analysis

1. Ensure SWI-Prolog is installed on your system
2. For the GUI version, verify that Java is installed:

java -version

---

💻 Usage

GUI Version

1. Navigate to the GUI folder
2. Run the batch file:

PrologSentimentAnalysis.bat

1. The JavaFX interface will open

1. Enter a sentence in the input field and click Submit
2. View the analysis results in the modal window

Accessing Documentation: Click on the "Help" menu in the top-left corner to view usage instructions.

Error Handling: If invalid input is entered, an error modal will appear with helpful tips.

---

Console Version

1. Navigate to the console folder
2. Open SWI-Prolog and consult the main file:

?- consult('Sent_Analysis.pl').

Alternatively, directly open the file with SWI-Prolog.

⚠️ Important: Before consulting, modify line 40 in Sent_Analysis.pl to update the path to cons_doc.txt:

1. The main menu will launch automatically. To run manually:

?- main.

1. Choose an option:
   • 1: Analyze a sentence
   • 2: View documentation
   • 3: Exit

Example Usage:

---

🔧 How It Works

Processing Pipeline

The system processes input text through the following stages:

1. Input Sentence
2. Text Preprocessing
   • Splitting
   • Stopword removal
3. Lexical Analysis
   • Apply intensifiers
   • Calculate neighbor influence
   • Retrieve scores from SentiWordNet
4. Calculate Scores
   • Filter scores
   • Add weights
   • Apply negations
5. Final Score

---

Text Preprocessing

1. Splitting Words

The input sentence is tokenized into individual words and stored in a list.

2. Remove Stopwords

Common stopwords (defined in the knowledge base) are removed:

stopwords('so').
stopwords('the').

3. Assign Labels

Each word is labeled based on its linguistic function:

• Indicator (pos/neg): Words indicating positive or negative sentiment
• Intensifier (in-[score]): Words that amplify sentiment (e.g., "very", "extremely")
• Negation (n): Words that negate sentiment (e.g., "not", "never")
• Neutral (neut): Default label if no match is found

Labels are defined in sent_modifiers.pl.

---

Lexical Analysis

1. Apply Intensifiers

Intensifier scores are merged with adjacent indicators. Non-adjacent intensifiers receive neutral weights.

2. Remove Intensifiers

Intensifier words are removed from the word list while preserving order.

3. Calculate Neighbor Influence

For each word, the system computes influence from neighboring words using their labels. This produces a triple of scores (-1, 0, 1) representing negative, neutral, and positive influence.

4. Retrieve Scores from SentiWordNet

Sentiment scores are retrieved from the SentiWordNet v3.0.0 lexicon, represented as:

synset(a, 0.125, 0.375, ['fearful']).
synset(n, 0.875, 0.0, ['honorableness', 'honorableness']).

Where:

• First parameter: Part of speech (a=adjective, n=noun, etc.)
• Second parameter: Positive score
• Third parameter: Negative score
• Fourth parameter: List of synonyms

---

Score Calculation

1. Calculate Scores

Based on neighbor influence:

• If influence is positive/negative (1 or -1): Use maximum positive/negative score
• If influence is neutral: Use average positive score minus average negative score

2. Add Weights

Multiply each score by its intensifier weight.

3. Apply Negations

Words preceded by negations have their scores multiplied by -1.

4. Compute Final Score

Sum all word scores to get the final sentiment score:

negative if score < -0.3
positive if score > 0.3
neutral otherwise

---

📁 Project Structure

prolog-sentiment-analysis/
├── GUI/
│   ├── PrologSentimentAnalysis.bat
│   ├── Main.java
│   ├── Prolog.java
│   ├── jpl.jar
│   └── JavaFX SDK files
├── console/
│   ├── Sent_Analysis.pl
│   ├── sent_modifiers.pl
│   ├── sent_wordnet.pl
│   └── cons_doc.txt
├── Documentation.pdf
└── README.md

---

📝 Examples

Good Input Examples:

• "I like to eat apple"
• "he is a very bad person"
• "the reason of his good behaviour is that he is a good person"

Bad Input Examples:

• "i like to play abcdef" (contains unknown words)
• "we are not happy today" (contains unaccepted words)
• "me and you love to watch a movie" (grammatical issues)

---

⚠️ Limitations and Future Improvements

Current Limitations:

• String Matching: Substring matching can cause false positives
• Case Sensitivity: All text is converted to lowercase for matching
• Limited Grammar Rules: Uses general linguistic patterns rather than comprehensive English grammar
• Non-exhaustive Dictionary: Limited coverage in sent_modifiers.pl

Potential Improvements:

• Stemming/Lemmatization: Reduce words to root forms for better matching
• Advanced Intensifier/Negation Handling: Apply these operations multiple times or to wider contexts
• Extended Neighbor Analysis: Consider more neighboring words with additional filters
• Configurable Thresholds: Allow users to adjust positivity/negativity score ranges
• Plural Handling: Better processing of plural forms
• Enhanced Grammar Rules: Incorporate more sophisticated linguistic patterns

---

📄 License

This project was developed as part of coursework at the University of Bari, Department of Computer Science.

---

🤝 Contributing

This is an academic project. If you have suggestions or improvements, feel free to open an issue or submit a pull request.

---

📧 Contact

For questions or feedback:

• Federico Chiaradia: f.chiaradia5@studenti.uniba.it
• Yonas Babulet Abetew: y.babuletabetew@studenti.uniba.it

---

Department of Computer Science University of Bari