Browsing by Author "Haddela, P. S"

Now showing 1 - 18 of 18

Embargo
Ads-In Site: Location based advertising framework with social network analyzer
(IEEE, 2014-12-10) Perera, A. A. G. A. K; Jayarathne, R. P. E. T; Thilantha, B. Y; Kalupahana, S. I. G; Haddela, P. S; Kirupananda, A; Edirisinghe, E. A. T. D
Social networks contain a vast amount of data that holds very valuable information and is nowadays identified as a very effective source in marketing. The content of social networks can be used to identify the business needs and preferences of people. This research is carried out with the aim of providing suitable advertisements for people based on their preferences by analysing their social network content. Users' demographic details are also considered in providing suitable advertisements of the shops that are most conveniently located for a particular user. The primary goal of this research is to build an advertisement framework that supports targeted advertising by analysing social network content. The information extracted by analysing the content of social networks is used to predict the advertisement categories that interest a particular user. The framework applies location based services to filter advertisements based on the location of the shop.
Embargo
Context aware stopwords for Sinhala Text classification
(IEEE, 2018-10-02) Gunasekara, S. V. S; Haddela, P. S
When working with Text Classification (TC), often the term "stopword" can be heard. Words in a document that are frequently occurring, but meaningless in terms of Information Retrieval (IR) are called Stopwords. There are various stopword lists available for many languages. According to the best of knowledge, no any generic stopword list has been built for the Sinhala language. This paper demonstrates how to generate a domain-specific stopword list from a given data set of Sinhala Newspapers. Hence, the seven stopword identification methods previously applied to other languages are presented to remove stopwords. Then, a new algorithm for building a domain-specific stopword list is proposed. For this method, it is assumed that average F-measure and average accuracy for the set of different stopword lists are measured by the performance of two classifiers. Based on the given comparative study, the most effective method to classify stopwords in Sinhala corpus can be identified.
Embargo
Document Clustering with Evolved Single Word Search Queries
(IEEE, 2021-06-28) Hirsch, L; Haddela, P. S; Di Nuovo, A
We present a novel, hybrid approach for clustering text databases. We use a genetic algorithm to generate and evolve a set of single word search queries in Apache Lucene format. Clusters are formed as the set of documents matching a search query. The queries are optimized to maximize the number of documents returned and to minimize the overlap between clusters (documents returned by more than one query in a set). Optionally, the number of clusters can be specified in advance, which will normally result in an improvement in performance. Not all documents in a collection are returned by any of the search queries in a set, so once the search query evolution is completed a second stage is performed whereby a KNN algorithm is applied to assign all unassigned documents to their nearest cluster. We describe the method and compare effectiveness with other well-known existing systems on 8 different text datasets. We note that search query format has the qualitative benefits of being interpretable and providing an explanation of cluster construction.
Embargo
Dynamic stopword removal for Sinhala Language
(IEEE, 2019-10-08) Jayaweera, A. A. V. A; Senanayake, Y. N; Haddela, P. S
In the modern era of information retrieval, text summarization, text analytics, extraction of redundant (noise) words that contain a little information with low or no semantic meaning must be filtered out. Such words are known as stopwords. There are more than 40 languages which have identified their language specific stopwords. Most researchers use various techniques to identify their language specific stopword lists. But most of them try to define a magical cut-off point to the list, which they identify without any proof. In this research, the focus is to prove that the cut-off point depends on the source data and the machine learning algorithm, which will be proved by using Newton's iteration method of root finding algorithm. To achieve this, the research focuses on creating a stopword list for Sinhala language using the term frequency-based method by processing more than 90000 Sinhala documents. This paper presents the results received and new datasets prepared for text preprocessing.
Embargo
EEG Based Neuromarketing Recommender System for Video Commercials
(IEEE, 2022-01-03) Bandara, S. K; Wijesinghe, U. C; Jayalath, B. P; Bandara, S. K; Haddela, P. S; Wickramasinghe, L. M
Video commercials are well known and extremely popular because of the innovative progression in technology and competition in the business fields. Therefore, breaking down the effect of these video commercials are essential before the start of marketing and promotions. Mainly this study addresses a strategy that can assess the viability of movie commercials which are basically trailers, utilizing Electroencephalogram (EEG) signals caught from a Brain-Computer Interface (BCI). This helps the business to have a thought regarding the effect of the commercial and the value. The study consists of a recommend system which suggests movie advertisements based on the preferences of the users. The particular video will be evaluated from emotion, attention and enjoyment aspect of the user. Random Forest prediction algorithm with 91.97% accuracy was used for emotion analysis, c-Support Vector Classifier (c-svc) algorithm with 91.70% accuracy was used for attention analysis while using statistical approach Central Limit theorem and Empirical rule was used for enjoyment analysis to isolate particular emotion state. From the initial results received, confirms that the proposed framework is producing promising results. Although this research is currently focused on movie and entertainment industry, and also has the potential to be developed and applied in many other industries.
Embargo
Effective Mechanism to Mitigate Injuries During NFL Plays
(IEEE, 2019-12-18) Ahamed, A. A. A; Arulanantham, A; Rajalingam, G; Ingran, K. S; Haddela, P. S
NFL (American football), which is regarded as the premier sports icon of America, has been severely accused in the recent years of being exposed to dangerous injuries that proves to be a bigger crisis as the players' lives have been increasingly at risk. Concussions, which refer to the serious brain traumas experienced during the passage of NFL play, have displayed a dramatic rise in the recent seasons concluding in an alarming rate in 2017/18. Acknowledging the potential risk, the NFL has been trying to fight via NeuroIntel AI mechanism [3] as well as modifying existing game rules and risky play practices to reduce the rate of concussions. As a remedy, we are suggesting an effective mechanism to extensively analyse the potential concussion risks by adopting predictive analysis to project injury risk percentage per each play and positional impact analysis to suggest safer team formation pairs to lessen injuries to offer a comprehensive study on NFL injury analysis. The proposed data analytical approach differentiates itself from the other similar approaches that were focused only on the descriptive analysis rather than going for a bigger context with predictive modelling and formation pairs mining that would assist in modifying existing rules to tackle injury concerns. The predictive model that works with Kafka-stream processor real-time inputs and risky formation pairs identification by designing FP-Matrix, makes this far-reaching solution to analyse injury data on various grounds wherever applicable.
Embargo
Effectiveness of rule-based classifiers in Sinhala text categorization
(IEEE, 2017-09-14) Haddela, P. S; Lakmali, K. B. N
In the recent past, the growth of Sinhala text usage on the web has been increasing rapidly due to the advancement in the field of information and communication technologies in Sri Lanka. With this change in society, automatic text categorization becomes important for many operations in computing. Therefore, the aim of this research is to assess commonly used rule based text classification algorithms against the Sinhala dataset. This study is limited to rule based classifiers as they are humanly interpretable by nature, which gives an added advantage to text classification. This paper presents a. The comparison of experiment results of rule based classifiers b. SinNG5 corpus and Sinhala stop word list named as SinSWL. The corpus and stop word list are freely available for academic researches.
Embargo
Enhanced Tokenizer for Sinhala Language
(IEEE, 2019-10-08) Senanayake, S. Y; Kariyawasam, K. T. P. M; Haddela, P. S
Tokenization process plays a prominent role in natural language processing (NLP) applications. It chops the content into the smallest meaningful units. However, there is a limited number of tokenization approaches for Sinhala language. Standard analyzer in apache software library and natural language toolkit (NLTK) are the main existing approaches to tokenize Sinhala language content. Since these are language independent, there are some limitations when it applies to Sinhala. Our proposed Sinhala tokenizer is mainly focusing on punctuation-based tokenization. It precisely tokenizes the content by identifying the use case of punctuation mark. In our research, we have proved that our punctuation-based tokenization approach outperforms the word tokenization in existing approaches.
Embargo
Google map and camera based fuzzified adaptive networked traffic light handling model
(IEEE, 2018-12-05) Nirmani, A; Thilakarathne, L; Wickramasinghe, A; Senanayake, S; Haddela, P. S
Rising traffic congestion has turned into a certain issue as the number of vehicles on roads are increasing. This research study was conducted to develop `Google Map and Camera Based Fuzzified Adaptive Networked Traffic Light Handling Model'. The main road with six major junctions was selected as the target route for the project. During this study, we were able to plan a limit and control traffic congestion utilizing two neural networks which process together to provide an efficient, productive and optimized solution based on real-time situations. Real-time video streams and Google Map traffic layer were used as primary input sources to the system. The Main algorithm was used to reduce traffic at a specific point whereas secondary algorithm was used to produce optimum decisions for the overall network. As a further advancement, REST endpoint was implemented to get the best route considering all the accessible data. With the aid of the previously mentioned techniques, an optimal traffic management model was developed.
Embargo
Hybrid framework for master data management
(IEEE, 2017-09-06) Fernando, L, K. W; Haddela, P. S
Master data management and real-time data warehousing are gaining increased prominence within the worlds of business and technology. Past research efforts have shed light towards development of many master data management approaches. On the other hand, growth of technology has demanded real-time analytics and real-time processing of data. This trend has shed light in developing multiple real-time data warehousing approaches to perform real-time analytics based on an organization's requirements. With the evolution of real-time data warehousing, Master Data Management was an issue for large organizations' when both systems are working in the same business environment. Since both systems focus on real-time integration, similar, duplicated and parallel data extraction processes were executed by these applications. This was due to the fact that master Data Management was designed to focus on the operational aspect and real-time data warehouse was designed to focus on analysis aspect of the organization. Hence, each had its own ways of managing master data. These duplicated extractions caused data quality issues in these parallel applications. This research provides a framework that combines both master data management and real-time data warehousing and ultimately proposing to build a Hybrid Real-Time Data Warehousing Architecture in order to achieve enterprise wide master data management.
Embargo
iRetina: An Intelligent Mobile Application for the Visually Impaired in an Indoor Environment
(IEEE, 2021-12-09) Labeeshan, A; Satharasinghe, S. A. R. L. P; Dhananjana, A. M; Rupasinghe, T. D; Rajapaksha, S, K; Haddela, P. S
The Visually Impaired face a lot of problems while they carry out their day-to-day activities. Even though there are various ways that technologies have been used together to come up with a solution, there is still no proper, efficient, user-friendly, low-cost solution crafted just for them. Mobile Phones these days are beyond smartness, and they could be used to achieve so many things now compared to before. In this application, the camera of the mobile phone acts as the eyes for the Visually Impaired. The main purpose of this system is to intelligently navigate the visually impaired. This research covers some components namely, intelligent environmental identification by using the YOLOv4 object detection algorithm, calculation of distance by using the Inertial Measurement Unit (IMU) sensors, focusing, guiding and detecting the expiry date of packaged products, identifying and enhancing objects in a low-light environment. The communication between the system and the Visually Impaired will be a verbal voice communication through the microphones and the speakers of the mobile phone. The results presented show that the proposed application successfully achieves its goals by providing the required functionality.
Embargo
Moderate Automobile Accident Claim Process Automation Using Machine Learning
(IEEE, 2021-01-27) Imaam, F; Subasinghe, A; Kasthuriarachchi, H; Fernando, S; Haddela, P. S; Pemadasa, N
In modern-day, traditional automobile accident claim process struggles to keep up with the recurring automobile accidents and furthermore, the claim itself is a critical point in which the policyholder may decide to switch to a different automobile insurance provider. In this paper, the authors present a system which can be used to automate the processing of claims for automobiles which were involved in less severe accidents in a much quicker manner. The presented system comprises of four components, each with a model developed using computer vision or machine learning techniques to facilitate the automation process. The models are built and fine-tuned using transfer learning and ensemble learning techniques in order to determine the damaged component of the automobile, determine the make and model of the automobile, compute an accurate repair estimate and also compute the likeliness of the policyholder may churn, to ensure that the policyholder is satisfied with the appraised amount and will be retained by the insurance provider.
Embargo
Product Recommendation System for Supermarket
(IEEE, 2020-12-14) Satheesan, P; Haddela, P. S; Alosius, J
Customers who seek the services at supermarkets are subjected to inconsistencies & ambiguities over choosing their desired products from a wide range of products with the closest quality. Meanwhile, supermarkets find it very difficult to satiate the customers' demand. Therefore, proposing a method to analyze the customers' need plays an important role in attracting new and regular customers. The purpose of this study is to formulate a product recommendation system which analyze customers' needs and thus recommend the best products. This system recommends products to the regular customers and to the new customers as well. New customers mean obviously the customers with no purchasing history at the supermarket in question. The system referred to recommends the products to the new customers using up two methods. One method recommends the most popular products while the other method solely focuses on the product description for recommendation. The system recommends the products to the regular customers using up user-based collaborative filtering, item based collaborative filtering and association rule mining. It recommends products to regular customers based on purchasing history and priority ratings given by other users who bought the products. Initially, the recommendation algorithm finds a set of customers who purchased and rated the products that overlap with the user who purchased and rated the products. The algorithm aggregates products from the customers with similar preference and eliminates the products the user has already purchased or rated. The proposed methodology improves the shopping experience of customers by recommending accurately and efficiently the products that are personalized to the need of the customers.
Embargo
Real time deception detection for criminal investigation
(IEEE, 2019-10-08) Lakshan, I; Wickramasinghe, L; Disala, S; Chandrasegar, S; Haddela, P. S
Deception Detection System (PREDICTOR) is a solution to support the criminal investigation process by providing a technological analysis in justifying the guilt of an accused criminal in the investigation process. This study gives guidelines to substantiate decision making in the interrogation. In judicature, the importance of a platform that is capable of analyzing the genuineness and the (a) reliability of a lie and a truth, (b) emotion of the suspect and the (c) attentiveness has been recognized for a long period. The feasibility of using Machine Learning (ML) techniques to build such platforms has been explored before. However, no known platform could identify the suspect's authenticity, emotion, and attentiveness. The goal is to analyze the brain waves and build a real-time deception detection application to analyze lie/truth, emotion and the attentiveness, which will support the investigation process in a wide range of angles to decision making. Electroencephalogram (EEG) based real-time lie detection, emotion detection, and attention detection will be implemented using ML tools and techniques along with the help of special hardware equipment called MUSE 2 headband. Especially this equipment is required for the data acquisition as well as the creation of the final application. The outcome of this system is a solution to be used during the criminal investigation process as a deception detection system for lie, emotion and attentiveness of the suspect. This is more effective in the questioning process to get an idea of the suspect. This system will have a major impact on the Police Department, Criminal Investigation Department, and Judicial System to ensure the real criminal and reduce the workload of Criminal Investigation officers.
Embargo
A rule based stemmer for Sinhala language
(IEEE, 2020-04-13) Kariyawasam, K. T. P. M; Wickramasinghe, S. Y; Haddela, P. S
Stemming, as its word implies it converts the original word into its root/base format which is called as stem. Stemming process plays a prominent role in natural language processing (NLP) because it makes applications more efficient and effective. Though stemming is such an important task, it is hard to find a stemming method for Sinhalese language which is official language of Sri Lanka. There are common language analyzers which cannot be use for stemming since they are highly language dependent. In this paper, we present a rule-based stemming method by using suffix and prefix rules in Sinhalese language.
Embargo
Student Teaching and Learning System for Academic Institutions
(IEEE, 2022-07-18) Kumarasiri, A. D. S. S; Delwita, C. E. M. S. M; Haddela, P. S; Samarasinghe, R. P; Udishan, R. P. I; Wickramasinghe, L
In today's global online environment, automation is essential for establishing a competitive advantage. Conversational Artificial Intelligence Systems are an example of an automation technology that has been embraced by some of world's most famous companies. In the field, higher education, these handy gadgets come in handy for administrators.Students' responses and performances are highly important and desired areas to improve the teaching-learning environment in the education organism. Evaluate the feedback to aid in the identification of flaws and actions. Institutes and universities in the field of education collect both quantitative and qualitative responses in order to improve the teaching-learning environment. However, most educational institutions and universities lack a sufficient student feedback evaluation system for determining students' feelings. The grading technique is currently being utilized to collect input. However, the grading process does not reveal the students' genuine feelings about the educational system, whereas written feedback allows students to describe specific areas and situations. Sentiment analysis is a type of qualitative feedback analysis that is based on records. The Nave Bayes (NB) classifier, Support Vector Machine (SVM), and Random Forest were employed in the majority of recent machine learning-based Sentiment Analysis prototypes. When it comes to measuring performance, most people utilize the kids' average grades, however this is not an accurate way. Because students can benefit from the experience indefinitely.This proposal offered a method for analyzing the sentiment of students' responses by combining Natural Language Processing (NLP), Machine Learning (ML), and Artificial Intelligence algorithms (AI). The goal is to combine machine learning algorithms and natural language processing approaches to achieve high accuracy. In this proposal, I suggested a collective model that connects machine learning algorithms to improve accuracy and performance. At the end of the presentation, AI planned to collect student responses in real time. The proposed method analyzes student comments from course reviews to determine mood, emotions, and satisfaction vs. displeasure. The approach categorizes sentimentalities into two categories: positive and negative, and senders' feelings into eight categories (8 emotion groups). The Fuzzy logic technique will be used to assess student performance. We intend to make our program available for broad use in the education industry so that organizations can improve the teaching-learning environment's quality.
Embargo
Surveillance based Child Kidnap Detection and Prevention Assistance
(IEEE, 2022-07-18) Kodikara, K. A. O. V; Hettiarachchi, P; Prathapa, D. M. J; Jayakody, J. M. A. M. S; Haddela, P. S
This research paper is based on child kidnap detection and prevention to identify susceptible child kidnap by unauthorized persons. The intelligent surveillance system proposed for this is known as "AICare". The purpose behind developing a proper kidnap detection methodology is to enhance and strengthen the existing child security systems. The key is to identify the main characteristics of a kidnapper in real-time, which follows face recognition, speed detection and object detection theories. Face recognition is used to identify whether the outlined individual has covered his body, especially his face, to hide the true identity, or else the person's face is directly processed for authentication. Speed detection is helpful in calculating movement speed and capturing whether the targeted individual is moving in a hurry. Finally, the stranger is subjected to object detection in order to classify whether he/she is handling a sharp object or not. The captured outcomes are subjected to a decision tree to resolve the person as a kidnapper suspect. The system results in an overall accuracy that is above 90%. As this solution is child-sensitive and responsive, it provides a long term platform that can real-time monitor for potential kidnap based on the kidnapper characteristics to support the working from home parents to take care of their children during crucial times.
Embargo
Suspicious activity detection in surveillance footage
(IEEE, 2019-11-19) Loganathan, S; Kariyawasam, G; Haddela, P. S
Suspicious activities are of a problem when it comes to the potential risk it brings to humans. With the increase in criminal activities in urban and suburban areas, it is necessary to detect them to be able to minimize such events. Early days surveillance was done manually by humans and were a tiring task as suspicious activities were uncommon compared to the usual activities. With the arrival of intelligent surveillance systems, various approaches were introduced in surveillance. We focus on analyzing two cases, those if ignored could lead to high risk of human lives, which are detecting potential gun-based crimes and detecting abandoned luggage on frames of surveillance footage. We present a deep neural network model that can detect handguns in images and a machine learning and computer vision pipeline that detects abandoned luggage so that we could identify potential gun-based crime and abandoned luggage situations in surveillance footage.