http://ir.haramaya.edu.et//hru/handle/123456789/144 2026-04-20T12:45:41Z 2026-04-20T12:45:41Z Elias Debelo Dr. Wondwossen Mulugeta Akubazgi Gebremariam http://ir.haramaya.edu.et//hru/handle/123456789/8138 2025-01-08T06:59:50Z 2019-08-01T00:00:00Z

AUTOMATIC ANAPHORA RESOLUTION FOR AFAAN OROMOO LANGUAGE Elias Debelo; Dr. Wondwossen Mulugeta; Akubazgi Gebremariam Anaphora Resolution is a process of finding an entity introduced earlier in the discourse referred to by current entity back in discourse. The referenced entity is called antecedent, of referring entity which is called anaphor. There are number of anaphora types in a text, pronominal anaphora is the most prevalent one. Anaphora resolution is an important subtask that can be used in many Natural Language Processing applications. This study aims at developing Anaphora Resolution model based on Machine Learning approach for Afaan Oromoo language. The language is morphologically complex in that pronoun itself exist hidden inside the verbs. Input data cleaning, tokenization and part of speech tagging, noun phrase extraction and hidden pronoun extraction are useful steps toward Anaphora resolution. From every valid antecedent-anaphor pairs in training and testing sets, feature vectors will be generated. Machine Learning classifier trained using positive and negative instances generated from training set. Sklearn python package was used for as a trainer using fit function and as a predictor using predict function. Sklearn is set of packages consisting of implementations of Machine Learning algorithms. Three types of dataset, gathered from Afaan Oromoo News, Bible verses and Oromo Fictions, were used for training and testing. Five top best features were used for training and testing out of 14 features extracted from the text. Using 10-fold cross validation technique, the three datasets were divided into 10% testing and 90% training at each run. Each test sets of datasets were tested by range of 1 to 10 sentence distance between antecedent and anaphor on three Machine Learning algorithm Decision Tree (DT), Support Vector Machine (SVM) and Naïve Bayes (NB). Performance of the models on the three datasets were represented as mean average of the 10-folds on ten sentence range. Generally, average precision achieved for Bible-Fiction dataset 52.3% on DT, 51.25% using SVM and 53.57% using NB, for News dataset 57.67% using DT, 47.77% using SVM and 57.5% using NB and for compiled dataset 47.62% using DT, 46.82% using SVM and 50.15% using NB was achieved for combined independent and hidden anaphors. This result could be enhanced primarily by finding better ways of getting feature values for antecedent-anaphor pairs 96p.

2019-08-01T00:00:00Z Elias Debelo Dr. Wondwossen Mulugeta http://ir.haramaya.edu.et//hru/handle/123456789/7904 2024-11-08T11:59:56Z 2019-08-01T00:00:00Z

AUTOMATIC ANAPHORA RESOLUTION FOR AFAAN OROMOO LANGUAGE Elias Debelo; Dr. Wondwossen Mulugeta Anaphora Resolution is a process of finding an entity introduced earlier in the discourse referred to by current entity back in discourse. The referenced entity is called antecedent, of referring entity which is called anaphor. There are number of anaphora types in a text, pronominal anaphora is the most prevalent one. Anaphora resolution is an important subtask that can be used in many Natural Language Processing applications. This study aims at developing Anaphora Resolution model based on Machine Learning approach for Afaan Oromoo language. The language is morphologically complex in that pronoun itself exist hidden inside the verbs. Input data cleaning, tokenization and part of speech tagging, noun phrase extraction and hidden pronoun extraction are useful steps toward Anaphora resolution. From every valid antecedent-anaphor pairs in training and testing sets, feature vectors will be generated. Machine Learning classifier trained using positive and negative instances generated from training set. Sklearn python package was used for as a trainer using fit function and as a predictor using predict function. Sklearn is set of packages consisting of implementations of Machine Learning algorithms. Three types of dataset, gathered from Afaan Oromoo News, Bible verses and Oromo Fictions, were used for training and testing. Five top best features were used for training and testing out of 14 features extracted from the text. Using 10-fold cross validation technique, the three datasets were divided into 10% testing and 90% training at each run. Each test sets of datasets were tested by range of 1 to 10 sentence distance between antecedent and anaphor on three Machine Learning algorithm Decision Tree (DT), Support Vector Machine (SVM) and Naïve Bayes (NB). Performance of the models on the three datasets were represented as mean average of the 10-folds on ten sentence range. Generally, average precision achieved for Bible-Fiction dataset 52.3% on DT, 51.25% using SVM and 53.57% using NB, for News dataset 57.67% using DT, 47.77% using SVM and 57.5% using NB and for compiled dataset 47.62% using DT, 46.82% using SVM and 50.15% using NB was achieved for combined independent and hidden anaphors. This result could be enhanced primarily by finding better ways of getting feature values for antecedent-anaphor pairs 96p.

2019-08-01T00:00:00Z Abdisa Abdella Seid Rashid, (PhD) Faizur Hussein, (MSc) Muluken http://ir.haramaya.edu.et//hru/handle/123456789/7878 2024-11-07T06:31:31Z 2024-04-01T00:00:00Z

HYBRID IMAGE COMPRESSION BASED ON FEATURES IMPORTANCE USING GAN Abdisa Abdella Seid; Rashid, (PhD) Faizur; Hussein, (MSc) Muluken Hybrid image compression techniques are the one to be used to compress and decompress the sensitive information without losing data, non-sensitive information with losing some information and completely removing the irrelevant information. Even though different works have been done to handle these complex compression tasks, there are still problems. Problem of field specific hybridization which means the information which is sensitive in health may not be sensitive in other areas while the information which is irrelevant in health can be sensitive or relevant information in other areas. Lack of the way to control percentage of information to be lost for a given image feature and the problem of using local features of an image rather than using global image features. To save the problem by using we proposed the hybrid model that contains 4 modules namely feature extraction module, lossless module, lossy module and combiner module. The feature extraction module takes an image and name of the feature to be extracted to extract the required feature and return it. The lossy module contains two submodules called lossy 75 and lossy 25 submodules. Lossy75 uses 0.5 bits to encode each pixel of a features. Lossy25 uses 0.25bits to encode each pixel of a feature. Only the features specified are going to be extracted and compressed from an image so that irrelevant features could be removed. The lossy submodules can take more than one feature at a time. The size of the data from both lossy submodules is dynamic. Two different features could be compressed using [0.50,0.25] bpp for each and average 0.373 bpp or [0.25,0.25] bpp and averge 0.25 bpp or [0.5,0.5] bpp or average 0.5 bpp. Generally, by this work we able to make hybrid image compression model flexible by which decision on feature importance can be specified by user. Also, we are the first to make lossy compression model of dynamic output and that does not depend on context of the feature. In the future we will work on the feature extraction module to add other features not included for this thesis to improve the model 116p.

2024-04-01T00:00:00Z Abrahim Abdo (Ph.D.) Faizur Rashid http://ir.haramaya.edu.et//hru/handle/123456789/7199 2024-01-02T06:45:00Z 2023-08-01T00:00:00Z

A FRAMEWORK FOR HEALTH INFORMATION SHARING AND PRIVACY PRESERVATION USING INFORMATION HIDING AND DIFFERENTIAL PRIVACY. Abrahim Abdo; (Ph.D.) Faizur Rashid Health information exchanges (HIEs) are an important part of the healthcare system. The World Health Organization emphasizes the importance of security standards in managing personal health data, as privacy concerns among healthcare consumers are increasing. HIPAA prioritizes unauthorized disclosures, improper disposal, access, and other IT security and privacy breaches of protected health information (PHI). The study demonstrates secure health information sharing methods, ensuring trust and patient privacy through AES-128-CBC and VSS-XOR, and integrating authentication, authorization, and access control with attributes for user tracking. We perform further comparative analyses based on the visual cryptography in LSB and DWT algorithms. PSNR results describe the similarity between the recovered secret image and the original image. The novel health data sharing method using Named Entity Recognition (NER), NER BiLSTM, provides 94% more accurate PII prediction than current evidence on medical text prediction. The differential privacy application evaluates noise levels to protect patients' privacy by removing or replacing patients and was compared using word2vec and GloVe 300-dimensional embedding approaches. Word2vec performed better semantically, while GloVe outperformed due to its support for a wide range of epsilon values, ensuring patient privacy. The effectiveness and viability of the suggested scheme are demonstrated by experiments and comparisons of the metrics. It is demonstrated that our suggested scheme can withstand statistical analysis and counterattacks. By combining legal agreements and technology to ensure meaningful information exchange, our developed protocol complied with information security and HIPPA privacy basic rules 111p.

2023-08-01T00:00:00Z