A decision support system for discrepancy reduction in Saudi courts' rulings

Dissertant

Madyuni, Malikah

Thesis advisor

Asim, Nasir
Rashidi, Taj al-Din

University

Al Akhawayn University

Faculty

School of Science and Engineering

Department

Software Engineering

University Country

Morocco

Degree

Master

Degree Date

2017

English Abstract

This research study aims at finding the optimal combination set of methods (text preprocessing algorithms and clustering models) to serve in a decision support system in Saudi Arabia‟s court rulings.

This is part of an overall effort to help reduce discrepancies in rulings by offering judges the possibility to retrieve similar past cases and the corresponding “Sharia” texts upon which the judgments were based.

Specifically, in this work we investigate the impact of Arabic text preprocessing tools (such as normalization, stemming, etc..) on clustering models‟ accuracy in the context of court rulings.

We also come up with the best clustering model to properly group similar rulings for the purpose of computing similarity between the description of a new case and past rulings.

Therefore, this study compares clustering algorithms (namely, K-Medoid, and Hierarchical Clustering with its diverse linkage methods: Single-Link, Complete-Link, Average-Link, Centroid Similarity and Ward‟s Method) both in terms of purity measures, and pairwise-accuracy measures (using Jaccard Coefficient and Rand Statistic).

Using Java and R, we implemented generic APIs for clustering and retrieving similar rulings, using either of the methods discussed earlier.

These APIs were then leveraged in a simple test-bed system, which served for experimentation and analysis of the different combinations of preprocessing and clustering methods using different parts of the ruling (“Hukm”).

The evaluation of the various combinations of methods was performed on a set of 105 rulings, belonging to 7 different manually labeled categories, and the experimental results show that (i) The use of the entire “Hukm” text leads to better pairwise similarity measures, compared to the use of summaries, (ii) Applying a Light Stemmer does not considerably reduce the data dimensionality, nevertheless it delivers the best clustering accuracy results both in terms of pairwise-similarity and matching-based measures, and finally (iii) given the two cluster validation methods used (Matching-Based, and Pairwise Measures), we come to the conclusion that Ward‟s method used in Hierarchical Agglomerative Clustering leads to higher clusters‟ purities, and provides the largest coefficients using both Jaccard and Rand Statistics.

Main Subjects

Information Technology and Computer Science

Topics

• Algorithms
• Decision support systems

No. of Pages

99

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-775468