Subarna Shakya

Word Sense Disambiguation using WSD specific WordNet of polysemy words

This paper presents a new model of WordNet that is used to disambiguate the correct sense of polysemy word based on the clue words. The related words for each sense of a polysemy word as well as single sense word are referred to as the clue words. The conventional WordNet organises nouns, verbs, adjectives and adverbs together into sets of synonyms called synsets each expressing a different concept. In contrast to the structure of WordNet, we developed a new model of WordNet that organizes the different senses of polysemy words as well as the single sense words based on the clue words. These clue words for each sense of a polysemy word as well as for single sense word are used to disambiguate the correct meaning of the polysemy word in the given context using knowledge-based Word Sense Disambiguation (WSD) algorithms. The clue word can be a noun, verb, adjective or adverb.

Word sense disambiguation in Nepali language

This paper presents a modified Adapted Lesk algorithm for word sense disambiguation in Nepali language. We included the synset, gloss, example and hypernym of the context words to form the final collection of context words. The context window contains all the words from the whole sentence except articles, prepositions and pronouns. The collection of words for each sense of a target word is also formed by including their synset, gloss, example and hypernym. Each word in the collection of context words is compared with every word in the collection of words for each sense of a target word to count the overlaps. Moreover, the numbers of examples for each word in our sample Nepali WordNet have been increased to four in average. The experiments performed on 348 words including 59 polysemy words shows the accuracy level of our algorithm to be 88.05%.

Autonomous use of fractal structure in low cost, multiband and compact navigational antenna

Different fractal structures and their relevance in navigational antennas have been studied. Based on multiband characteristics and pe rformance of si erpinski gask et fr actal structures, a dual band, low profile antenna is devised in thi s paper. Cla ssical fra ctal st ructures h ave be en g enerated with extensive use of MATL AB, dime nsions o f the parasitic lay ers are determined v ia sev eral optimizations in Ma thCAD and finally design analysis is carried out using Ansoft Designer. The performance of this antenna is theoretically measured in terms of its return loss, gain radiation pattern and axial ratio. The multi layered physical antenna has been fabricated using glass epoxy substrate material contributing acceptable bandwidth in both bands. The measured performance of the fabricated antenna has been analyzed and evaluated with the theoretical outcomes.

PolyWordNet: A lexical database

We developed a new lexical database named as ‘PolyWordNet’. The PolyWordNet organizes multiple senses of a polysemy word in such a way that each sense of the polysemy word is linked with its related words by dividing these related words into verbs, nouns, adverbs and adjectives. Each related word in the PolyWordNet is linked only with a single sense of a polysemy word except for the case of bridging related word. This is because such related word will lead to the multiple senses of a polysemy word during the sense disambiguation process if the related word is liked to more than one sense of the same polysemy word introducing the ambiguity in ambiguity as in the case of contextual overlap count WSD approaches that use the Princeton WordNet for sense disambiguation. The PolyWordNet resolves this problem which is produced due to the common information collected from Princeton WordNet. The results obtained from the experiments show exceptionally high accuracy (96.11%) of our Word Sense Disambiguation algorithm that uses our lexical database PolyWordNet. This accuracy is significantly higher than that of the accuracy (58.33%) of the other contextual overlap count Word Sense Disambiguation method that used the Princeton WordNet for sense disambiguation.

Human behavior prediction using facial expression analysis

Computer Vision seeks to emulate human vision by analyzing digital image inputs as human perception does. To detect an emotion will not be a difficult task to for human but for any computer, detecting an emotion will be a difficult job to perform as they are unaware of that human nature. In our paper, first of all, we converted a video into frame sequence which is used for human behavior prediction by correlating frames. To analyze face, a face is detected and the ROI for expression using Viola-Jones AdaBoost method on color filtered image algorithm to detect peoples faces, noses, eyes, mouth, or upper body, as these parts have higher entropy for emotion detection. We used a PCA algorithm for feature extraction which calculates coefficient, score, and latent; where latent is used to calculate Eigenvector. As we found that left to right and top to the bottom approach to facial movement gives facial gesture change and facial expression change respectively and so on. For human emotion prediction, we used Euclidian distance calculation. From the various sequence of emotion, appropriate behavior is predicted. Finally, we have developed an algorithm for continuous tracking and monitoring the moving suspicious faces continuously using Kalman filter estimation. The proposed method is simple and very useful for the high-security alert zones e.g. hospital, airport, bank etc.

Optical Character Recognition for Nepali, English Character and Simple Sketch Using Neural Network

Optical Character Recognition (OCR) is the process of text extraction from of images of typewritten or handwritten text. It deals with the recognition of optically processed characters, with the advent of digital optical scanners a lot of paper based books, textbooks, magazines, articles and documents are being transformed into an electronic version that can be manipulated by a computer. Unlike English character recognition, Nepali languages are complicated in terms of structure and computations. Nepali language are derived from Devanagari Script; written from left to right fashion having common features of containing straight line on top ‘Shiro Rekha’. The OCR systems developed for the Nepali language carry a very poor recognition rate due to error in character segmentation, ambiguity with similar character, unique character representation style. In this paper we proposed an OCR for Nepali text in Devanagari script, using multi-layer feed forward back propagation Artificial Neural Network (ANN), which improved its efficiency and accuracy. Adaptive learning rate with Gradient descent algorithm is implemented in Neural net with 2 hidden layers used with input and output and MMSE is the performance criteria. Various classifiers for training characters are created and stored. De-noised test sheet is carefully segmented and inputted in trained neural net resulted higher accuracy. Also we have included recognizing simple sketch like as tree, home, and ball.

ICT Implementation and Infrastructure Deployment Approach for Rural Nepal

Information and Communication Technology (ICT) is the basic tools of our daily lives to optimize the resources, improve work performance and efficient service delivery. Due to diverse demographic and geographic situation of Nepal which has remote areas, government itself is not able to properly provide services to the citizens at Rural. Empowering rural people by means of ICT services in service delivery is becoming quite challenging. Nepal, after adopting the liberalization policy to encourage involvement of private sectors (Government of Nepal, http://nta.gov.np/en/legislation/policies, [1]), there have been optimum competitions in ICT sectors which help to reduce the digital divide and considerably increased the ICT penetration rate (Nepal Telecommunications Authority, http://nta.gov.np/en/mis-reports-en, [2]). But the services are more squeezed to city/urban areas. Varieties of challenges at rural Nepal raised the operability and sustainability issues of ICT centers. We have analyzed the survey data carried on the rural ICT centers to visualize its status and proposed the sustainable ICT implementation approach for rural Nepal.

E-government security readiness assessment for developing countries: Case study: Nepal Govt. organizations

Since the beginning of research on E-government security problems, many facts have proved that the construction of e-government security assurance system can guarantee the running of the whole e-government system in a safe, reliable and effective mode. In Nepal, government organizations are already using few of these technologies to increase efficiency and effectiveness in service delivery and provide easier interaction between the citizens and the government. However, these benefits do not come without risks for information being misused, service disrupted or any other attacks interrupting the normal operation of computer based information systems. Problems with existing E-government systems security can be identified and taken care of by conducting regular audits. The goal of the paper is to investigate and understand the current E-government security audit readiness situation of Nepalese government organizations and suggest an approach to solve the existing situation. To meet this goal, government organizations that are intended to offer e-service has been purposely selected and assessed using questionnaire as a method of data collection. Security management domains in ISO audit checklist and ICT security readiness checklist are used to design the questionnaire to assess the security audit readiness of Nepal government organizations and suggest a framework for improvement.

ICT, Local Governance and Branchless Banking: Triangular Efforts to Make Social Cash Transfer Easy and Transparent

The proper use of Information and Communication Technology (ICT) has made human daily activities easy and fast. The service delivery of local government has been fast, efficient and reliable. The government provides social protection to its citizens so that poor citizens maintain minimum life standard. The government has been putting more efforts on making social cash transfer easy and transparent. The needy and targeted poor people have been missed in social protection programme. Those missing people can be easily captured by use of ICT and branchless banking. Branchless banking is an innovative idea to provide financial facilities to poor people of rural areas. The research explores that the local government with the proper use of ICT and branchless banking can proof social cash transfer easy and transparent.

Comparative analysis of backpropagation algorithm variants for network intrusion detection

The system security has turned into an extremely critical worry as system assaults have been extending with the ascent of hacking devices, inconvenience of systems and interruptions in number and brutality. This paper is centered around interruption identification by utilizing Multilayer Perceptron (MLP) with various calculation of backpropagation neural network. In this paper, performance of various backpropagation algorithms has been evaluated using KDDCup99 dataset. The dataset has been preprocessed to be made suitable for neural network input and the input set and target set are separated. The modified dataset has been used to evaluate the performance of BFGS Quasi-Newton, Levenberg-Marquardt, and Gradient Descent with Adaptive backpropagation algorithm. Different performance parameters such as mean square error, attack detection rate, recall rate, precision rate, epochs has been used for the algorithm comparison. Based on the evaluation results, the research purposes Levenberg-Marquardt backpropagation algorithm to be the best performing and efficient algorithm for the network intrusion detection for KDD Cup dataset. Different classes of attacks have been also determined comparing the output values obtained with the target set.