2006

Proceedings of the 3rd Annual Session 15th September 2006 Moratuwa Sri Lanka 

First Sinhala Chatbot in action
B. Hettige , Asoka. S. Karunananda 

Abstract-Chatbots are becoming popular as a means for interactive communication between human and machines. Due to their interactivity, chatbots are much better than standard machine translation systems, which may provide unrealistic solutions when the system cannot perform without user intervention. This paper reports on the design and implementation of the Sinhala Chatbot System that can communicate between computer and user, through Sinhala language. This is the first ever Sinhala Chatbot. The current Chatbot has been designed to work on Linux and Windows Operating systems. As such the current chatbot can be queried on operating system related concepts such as date, time, and also identify individuals and greet accordingly. This system has been developed as an application of a Sinhala parser that comes under a major component of our project in English to Sinhala machine translation system. Nevertheless, our chatbot is more than a mere application of the said Sinhala paper, but an extension to capture verbal syntax and semantics of Sinhala language into a machine translation. The entire system has been developed using JAVA and SWI-PROLOG that runs on both Linux and Windows. The current chatbot can be used as ‘shell’ for developing chatbots for any domain.

PDF(121KB)

 

A Review paper – Recent developments in Bayesian approach in filtering
D. Ihalagedera & U. Ratnayake

Abstract-Junk mail is one of the main problems in Internet. There are several methods for the automated construction of filters to eliminate such unwanted messages from user’s mail system. This paper is mainly concerned about the Bayesian filtering method and its different types of applications in junk e-mail filtering. Bayesian technique is trained automatically to detect spam messages. Several implementations that use Bayesian techniques are available as software. Any user can apply this software in different layers of client side or server side. But spammers are now trying to defeat Bayesian filters by including random dictionary words and/or short stories in their messages. The Bayesian filter can be moderated to block the new spammer techniques. The efficiency of the Baysian filter is greater than the other e-mail filters. If any one wants to filter spam out of email, it is strongly recommended not to automatically delete messages. The same is true for your real email; instead of deleting it, move it to another folder. That way, you’ll build a collection of spam and non-spam messages, which will come in handy for training filters.

PDF(106KB)

 

Distributed Artificial Neural Network Training Using an Intelligent Agent
G.R.C De Silva & Asoka S. Karunananda

Abstract-Obtaining the best configuration for an Artificial Neural Network (ANN) has always been problematic due to the non availability of a uniform method of finding the best configuration. The trial-and-error approach is used even today for this purpose regardless of its limitations. This project exploits intelligent Agent technology and distributed computing techniques to automate and streamline the task of ANN training. A static, centralized Agent autonomously generates and trains multiple ANN’s in varying configurations for a given dataset. By analyzing the progress of each training session the Agent discovers which ANN configuration is best suited for the training data. The distributed system on which the Agent operates allows multiple ANN’s to train concurrently which helps identify the best configuration in a much shorter timeframe. Knowledge on past ANN trainings is gathered by the Agent from which it learns to identify most suitable ANN configurations for subsequent training sessions. The end result is a fully automated system that can identify the most suitable ANN configuration for a given dataset quickly, efficiently and autonomously.

PDF(106KB)

 

Virtual Tour Agent
Rupesh Selvaraj, Asoka S. Karunananda

Abstract-Tourists normally have to rely on common tour packages that are designed and prepared by travel agents or they have to prepare their own tour itinerary which may or may not fulfill their expectations. The authors propose that Software Agents technology and other Artificial Intelligence Techniques like Expert System could be utilized to prepare a custom based solution to cater to personalized requirements. The Software Agent Technology is used to filter data from the net; Expert System is used to generate the travel itinerary; and Nearest Neighbor Algorithm is used to find the optimal pathway among locations. It also provides facilities to regenerate a travel itinerary in different circumstances. Apart from that, the travelers are also provided with travel related information plus travel support functions to provide a compressive solution.

PDF(66KB)

 

Artificial Intelligence Approach to Effective Career Guidance
Chathra Hendahewa, Maheshika Dissanayake, Savindhi Samaraweera, Narmada Wijayawickrama, Anusha Ruwanpathirana and Asoka S. Karunananda

Abstract-With the vast opportunities available for IT education, career guidance has become a crucial theme for deciding the appropriate career path of students. The lack of such expert career guidance systems in this domain inspired us to come up with the initial idea. This paper reports on a Career Advisory Expert System, named iAdvice to guide students engaged in their higher education to determine their career paths and to select their course subjects to be inline with their career goals. Expert System features such as reasoning ability, providing explanations providing alternative solutions, providing uncertainty and probability measures, questioning ability are found in iAdvice. The design of Career Advisory Expert System takes into consideration factors such as past examination performance, student preferences and skills, industry alignment with subjects, which are the main factors also considered by a human expert in providing career guidance. According to the evaluation carried out it was found out that the model has a capability of about 70% accuracy in predicting the performance, about 85% relevance of the advice provided, the self-explanatory nature of the system and about 87% provision of informative and useful advice.

PDF(245KB)

 

Widening Scope of Principal Component Analysis: A Fuzzy Approach
D.S. Kalana Mendis, Asoka S. Karunananda and U. Samarathunga

Abstract-Knowledge modeling is concerned with abstract model mapping using real world domains. Further all knowledge is tacit or rooted in tacit domains. Abstracting is mainly concerned with classification of such knowledge. In this issue statistical techniques can be issued and mainly concerned with multivariate statistical techniques. However using principal component analysis (PCA) as a multivariate technique makes problematic situation due to inability of classifying knowledge. Analysis using PCA is limited up to principal components extracted by PCA. Therefore existing algorithm of PCA to be addressed for modeling knowledge should be modified. This paper presents a novel mechanism for modifying PCA algorithm to address the problem in concerned using Fuzzy Principal component Analysis (FPCA). Here principal components have been used to define intervals for membership function. By doing so, knowledge classification is done effectively by constructing fuzzy memberships functions integrated with PCA. The experimental results using Ayurvedic medicine show that our approach is very promising.

PDF(145KB)

 

Qualitative Realization of the Visual World
D. D. M. Ranasinghe and Asoka. S. Karunananda

Abstract-Visual capability is one of the strongest senses of perception of human beings. Humans tend to represent the visually perceived world mainly in a qualitative manner and thereby attain considerably high accuracy in reasoning and prediction. Even though this is an innate ability of humans, embedding this feature in the development of cognitive vision systems has been a research challenge. A research has been carried out to develop a system that is capable of learning qualitative rules that underlies in the arrangement of the observed visual scene. A set of symbolic data generated from a dynamic visual scene that comprise object movement is considered as the input to the system and by analyzing objectobject qualitative spatial and temporal representation and reasoning mechanisms the system generates the underlying set of rules of the scene. As an application the system produces sketch images of the observed scene. This work has great potential in developing agents that can be used for autonomous learning from visual scenes in a manner closer to human learning from visual scenes.

PDF(113KB)