Sameerchand Pudaruth

Categorisation of Supreme Court Cases Using Multiple Horizontal Thesauri

Text classification is a branch of Artificial Intelligence which deals with the assignment of textual documents to a controlled group of classes. The aim of this paper is to assess the use of a controlled vocabulary in the categorisation of legal texts. Controlled vocabularies such as the Medical Subject Headings, Compendex and AGROVOC have been proved to be very useful in the fields of biomedical research, engineering and agriculture, respectively. In this work, a number of lexicons are created for some pre-defined areas of law through an automated approach. The lexicons are then used to categorise cases from the Supreme Court into eight distinct areas of law. We then compared the performance of these lexicons with each other. We found that lexicons which have a mixture of single words and short phrases performs slightly better than those consisting simply of single words. Weights were also assigned to the terms and this had a significant positive impact on the classification accuracy. The number of words in each thesaurus was kept constant. A hierarchical classification was also attempted whereby cases were first classified into either a civil case or a criminal case. Civil cases were then further classified into company, labour, contract and land cases while criminal cases were classified into drugs, homicide, road traffic offences and other criminal offences. Our best model achieves a global accuracy of 78.9 %. Thus, we have demonstrated that it is possible to get good classification accuracies with legal cases through the use of automatically generated thesauri. This outcome of this research can become an integral part of the eJudiciary project that has already been initiated by the government. In line with the vision of the Judiciary, we are hereby in the process of creating an intelligent legal information system which will benefit all legal actors and will have a definite positive impact on the legal landscape of the Republic of Mauritius. Lawyers, attorneys and their assistants would spend less time on legal research and hence they would have more time to prepare their arguments for their case. We are optimistic in believing that this will make the whole business of providing justice more effective and more efficient through the reduction in postponement of cases and a reduction in the average disposal time of cases.

Authorship Attribution Using Stylometry and Machine Learning Techniques

Plagiarism is considered to be a highly unethical activity in the academic world. Text-alignment is currently the preferred technique for estimating the degree of similarity with existing written works. Due to its dependency on other documents it becomes increasingly tedious and time-consuming to scale up to the growing number of online and offline documents. Thus, this paper aims at studying the use of stylometric features present in a document in order to verify its authorship. Two machine learning algorithms, namely k-NN and SMO, were used to predict the authenticity of the writings. A computer program consisting of 446 features was implemented. Ten PhD theses, split into different segments of 1000, 5000 and 10000 words, were used, totaling 520 documents as our corpus. Our results show that authorship attribution using stylometry method has generated an accuracy of above 90 %, except for 7-NN with 1000 words. We also showed how authorship attribution can be used to identify potential cases of plagiarism in formal writings.

Collaborative Driving Support System in Mobile Pervasive Environments

The Bluetooth protocol can be used for inter-vehicle communication equipped with Bluetooth devices. This work investigates the challenges and feasibility of developing intelligent driving system providing time-sensitive information about traffic conditions and roadside facilities. The architecture for collaborative vehicle communication system is presented using the concepts of wireless networks and Bluetooth protocol. We discuss how vehicles can form mobile ad-hoc networks and exchange data by the on-board Bluetooth sensors. The key design concepts of the intelligent driving service infrastructure are analyzed showing collaborative fusion of multiple positional data could give a better understanding of the surrounding traffic conditions for collaborative driving. The technical feasibility of using Bluetooth for data exchange among moving vehicles is evaluated.

Automatic Recognition of Medicinal Plants using Machine Learning Techniques

The proper identification of plant species has major benefits for a wide range of stakeholders ranging from forestry services, botanists, taxonomists, physicians, pharmaceutical laboratories, organisations fighting for endangered species, government and the public at large. Consequently, this has fueled an interest in developing automated systems for the recognition of different plant species. A fully automated method for the recognition of medicinal plants using computer vision and machine learning techniques has been presented. Leaves from 24 different medicinal plant species were collected and photographed using a smartphone in a laboratory setting. A large number of features were extracted from each leaf such as its length, width, perimeter, area, number of vertices, colour, perimeter and area of hull. Several derived features were then computed from these attributes. The best results were obtained from a random forest classifier using a 10-fold cross-validation technique. With an accuracy of 90.1%, the random forest classifier performed better than other machine learning approaches such as the k-nearest neighbour, naive Bayes, support vector machines and neural networks. These results are very encouraging and future work will be geared towards using a larger dataset and high-performance computing facilities to investigate the performance of deep learning neural networks to identify medicinal plants used in primary health care. To the best of our knowledge, this work is the first of its kind to have created a unique image dataset for medicinal plants that are available on the island of Mauritius. It is anticipated that a web-based or mobile computer system for the automatic recognition of medicinal plants will help the local population to improve their knowledge on medicinal plants, help taxonomists to develop more efficient species identification techniques and will also contribute significantly in the protection of endangered species.

Extraction of Roads from Remotely Sensed Images using a Multi-Angled Template Matching Technique

The extraction of road networks from satellite or aerial images has profound applications in the fields of urban planning, setting up of transportation networks, disaster management, cartography and in Geographical Information Systems. In this paper, we have developed a multi-shaped and multi-angled template matching algorithm in order to extract the road network from medium and high-resolution satellite images. We used a quadruple orthogonal line filter to extract lines from four different directions. Small isolated points and edges are removed using appropriately designed clearing filters. Gaps in the road network are bridged using our edge linking algorithm, which is based primarily on the spectral property of the original image pixels. The four images are cleared again using directional clearing filters to remove broken edges that cannot be linked to the road network. Finally, the output from these four separate images are fused into a single image in order to get the final output image which represents the road network. The results obtained demonstrate the practicability of our proposed method in rural and semi-urban regions.

An intelligent question answering system for ICT

Demand for Question Answering systems is increasing day by day since they deliver short, precise and question specific answers. Huge amount of redundant data from irrelevant documents through the World Wide Web has given rise to an Information Technology (IT) Question Answering system to be used by secondary students. The main aim of developing such a system is to allow students obtain concise and relevant information about the IT subject in a short time period without having to search thousands of documents or websites. Hence, an interactive platform has been developed which allows the user to pose questions and the system responds by displaying the most relevant answer. Text mining techniques have been used to select documents across the web which contains answers to IT questions and use these documents to form a structured knowledge base. Algorithms have been derived to extract basic components from questions and methods to match and compare these with the knowledge base and to finally display the most relevant answer.

Gender Profiling from PhD Theses Using k-Nearest Neighbour and Sequential Minimal Optimisation

Author profiling is a subfield of text categorisation in which the aim is to predict some characteristics of a writer. In this paper, our objective is to determine the gender of an author based on their writings. Our corpus consists of 10 PhD theses which was split into equal sized segments of 1000, 5000 and 10000 words. From this corpus, a total of 446 features were extracted. Some new features like combined-words, new words endings and new POS tags were used in this study. The features were not separated into categories. Two machine learning classifiers, namely the k-nearest neighbour and a support vector machines classifier were used to assess the practicability and utility of our study. We were able to achieve 100% accuracy using the sequential minimal optimisation (SMO) algorithm with 40 document parts. Surprisingly, the simple and lazy k-nearest neighbour (kNN) classifier which is often discarded in gender profiling studies achieved a 98% accuracy with the same group of documents. Furthermore, 5-NN and 7-NN even outperformed SMO when using 400 document parts of 1000 words each. These values are much higher than those obtained in previous studies. However, we have used a new dataset and the results are therefore not directly comparable. Thus, our experiments provide further evidence that it is possible to infer the gender of an author using a computational linguistic approach.

Automated generation of song lyrics using CFGs

This paper gives an overview of how challenging song writing is and gives an insight on how we developed a semi-automatic lyric generator for English songs. Writing lyrics has always been a challenging task as it involves not only creativity but also inspiration. Prior to implementation of the lyrics generator, much analysis were carried out so as to get in-depth information about the requirements of good lyrics. Research has been done in various fields such as artificial intelligence and natural language processing to be able to master the various techniques for text processing and to be able to use them in our own way. And finally, we carried out our evaluation by making a survey about the generated and existing lyrics and it proved to be very satisfactory. Many people rated the generated lyrics as being an existing lyric.

Using Artificial Intelligence Approaches to Categorise Lecture Notes

Lecture materials cover a broad variety of documents ranging from e-books, lecture notes, handouts, research papers and lab reports amongst others. Downloaded from the Internet, these documents generally go in the Downloads folder or other folders specified by the students. Over a certain period of time, the folders become so messy that it becomes quite difficult to find our way through them. Sometimes files downloaded from the Internet are saved without the certainty that they will be used or revert to in the future. Documents are scattered all over the computer system, making it very troublesome and time consuming for the user to search for a particular file. Another issue that adds up to the difficulty is the improper naming conventions. Certain files bear names that are totally irrelevant to their contents. Therefore, the user has to open these documents one by one and go through them to know what the files are about. One solution to this problem is a file classifier. In this paper, a file classifier will be used to organise the lecture materials into eight different categories, thus easing the tasks of the students and helping them to organise the files and folders on their workstations. Modules each containing about 25 files were used in this study. Two machine learning techniques were used, namely, decision trees and support vector machines. For most categories, it was found that decision trees outperformed SVM.

Emergency preparedness through game playing

ABSTRACT The lack of emergency preparedness in Mauritius has been the cause of many tragedies. Our approach to tackle this problem was by developing an emergency preparedness game layered and fused with a disaster warning and guidance system that emanates clarity to the unfathomable bearings of emergencies and natural disasters. The emergency preparedness game is based on a selection of diverse real life-threatening difficulties that entail different strategies aimed at bettering the survival instincts of users. It uses story-telling scenarios along with in-game footnotes that yield directives on how to brave fierce and unpredictable calamities. The game reinforces a sense of self-composedness and suppressing untimely fears of users in horrendous circumstances. With regard to the warning system, it unremittingly feeds users with notifications during emergencies, that encases shortest escape routes to lead them to safe locations via a fully functional GPS map. This application brings some novelties that are virtually non-existent in related applications. For instance, this application includes a warning and guidance system, a 3D scenario game to prepare its users for disasters, an interactive survival toolkit selection, an SMS rescue feature and a mass notification system via the web.