Gökhan Dalkiliç

A simple yet effective spam blocking method

Content-based filtering techniques for SPAM recognition require considerable amount of computational resources. Moreover, because the spammers are able to change the content as they wish, content-based filtering is not very successful. For this reason network-based filtering is gaining importance, and Sender Policy Framework (SPF) is a protocol that can provide an efficient filtering technique. In this paper, the advantage of the presence of SPF records and an SPF prediction method developed for domains without SPF records are explained. The SPF prediction method has been applied on the e-mails of thousands of volunteering users who are in our university domain deu.edu.tr, for SPAM filtering. First results of the proposed SPF prediction method are more promising than the present popular filtering methods.

Turkish word n-gram analyzing algorithms for a large scale Turkish corpus - TurCo

To calculate some statistical properties of a language, first you need to take some samples of that language. That sample is called a corpus. An unbalanced large scale Turkish text corpus (TurCo) having /spl sim/362 MB capacity and more than 50 million words was prepared by using 12 different resources including Web sites and novels in Turkish language. Different algorithms were tested to obtain the n-gram (1/spl les/n/spl les/5) values. Efficiencies of different algorithms have been examined by applying them onto the each piece of the corpus one by one. Only detailed results of the two algorithms created without using database tables are given, because all the other algorithms need to run more than one day which makes those tests inefficient.

Zipf's law and mandelbrot's constants for turkish language using turkish corpus (turco)

Zipfs Law is a common law applied for different kinds of observations. Many investigations were carried out to find the correspondences between Zipfs Law and different languages. This study deals with the correspondence of Turkish with Zipfs Law and finding Mandelbrot constants (c and B ) by using a large scale Turkish corpus (TurCo). In order to determine these constants, coefficient of determination was used, and different c and B values were examined. As both languages show agglutinative characteristics, the most suitable B value was found smaller than 1 for Turkish like Korean, and c value was found as 0.27.

Cryptographically Supported NFC Tags in Medication for Better Inpatient Safety

Reliable sources report that errors in drug administration are increasing the number of harmed or killed inpatients, during healthcare. This development is in contradiction to patient safety norms. A correctly designed hospital-wide ubiquitous system, using advanced inpatient identification and matching techniques, should provide correct medicine and dosage at the right time. Researchers are still making grouping proof protocol proposals based on the EPC Global Class 1 Generation 2 ver. 1.2 standard tags, for drug administration. Analyses show that such protocols make medication unsecure and hence fail to guarantee inpatient safety. Thus, the original goal of patient safety still remains. In this paper, a very recent proposal (EKATE) upgraded by a cryptographic function is shown to fall short of expectations. Then, an alternative proposal IMS-NFC which uses a more suitable and newer technology; namely Near Field Communication (NFC), is described. The proposed protocol has the additional support of stronger security primitives and it is compliant to ISO communication and security standards. Unlike previous works, the proposal is a complete ubiquitous system that guarantees full patient safety; and it is based on off-the-shelf, new technology products available in every corner of the world. To prove the claims the performance, cost, security and scope of IMS-NFC are compared with previous proposals. Evaluation shows that the proposed system has stronger security, increased patient safety and equal efficiency, at little extra cost.

Analysis of two protocols using EPC Gen-2 tags for safe inpatient medication

The number of people suffering due to wrong medication is increasing. A hospital-wide ubiquitous system can prevent inpatient medication errors. A correctly designed system guarantees the correct medicine and dosage administration, at the correct time. Such a system also helps tracing the location, time of errors and generates fast alarms to prevent harm to patients. Two proposals using UHF passive radio frequency identification tags have been made. But the proposed schemes give away critical patient information, raising privacy and intentional harm concerns. This work studies the proposals and demonstrates a new full-disclosure attack, on both. In addition, the weaknesses of protocols using the PRNG and CRC functions as encryption algorithms are studied. A conclusion is drawn about the suitability of grouping proofs and other protocols using the EPC Global Class-1 Generation-2 tags, in patient medication.

Word sense disambiguation for Turkish

Word Sense Disambiguation (WSD) is the core and one of the hardest problems of many Natural Language Processing tasks. WSD is considered as an AI-complete problem. Although there are many approaches trying to solve this problem, many of them are not adequate to solve WSD problem for Turkish. Dealing with sense ambiguity for Turkish also requires dealing with stemming ambiguity as well as polysemy, homonymy and categorical ambiguity. In this study, largely known Lesk and Simplified Lesk methods are modified and adapted to Turkish. The main aim of this project is to minimize the word sense ambiguity for Turkish and this is performed by eliminating the incorrect senses as much as possible by applying proposed methods.

A 300 MB Turkish Corpus and Word Analysis

In order to determine some properties of a language, a corpus of that language should be created. To analyze Turkish language, at first, a Turkish corpus having ∼300 MB capacity and more than 44 million words was prepared by using 10 different web sites having Turkish content. Most frequently used word statistics of Turkish were calculated by using this corpus. Frequencies of most frequently used first 7 words were compared with their equivalent in English, and it was found out that most frequently used words are not nouns in natural languages Most frequently used words having 1 to 5 letters were determined and they were applied onto a randomly selected text in order to test the validity of the process.

Is NFC a Better Option Instead of EPC Gen-2 in Safe Medication of Inpatients

Wrong medication is an important problem of the citizens of many countries. Using contemporary technologies like UHF Gen-2 RFID tags helps decreasing the medication errors, experienced by many. As UHF Gen-2 tags have limited capacity, cryptographic algorithms cannot be accommodated. The only available functions PRNG and CRC cannot be used instead of cryptographic algorithms. To overcome the known weaknesses, various grouping protocols have been proposed. But, each protocol has some deficiencies. Two of those protocols are covered in this study. Some of their common deficiencies are studied and solutions are suggested.

Turkish spelling error detection and correction by using word n-grams

N-grams can be used for spelling check and correction processes. The first step to use n-grams is to find the language specific n-grams by using a corpus. But a corpus cannot be big enough to contain all the possible word n-grams. Back-off smoothing technique is one of the techniques to estimate the frequency of the unknown n-grams in a corpus. By using Back-off technique and the Minimum Edit Distance (MED) algorithm, a program was developed to check spelling errors and suggest corrections in a sentence typed in Turkish. The results were compared with the results of Microsoft Word 2003 proofing tools, and found to be much better.

Improving communication among nurses and patients

Patients use nurse call systems to signal nurses for medical help. Traditional push button-flashing lamp call systems are not integrated with other hospital automation systems. Therefore, nurse response time becomes a matter of personal discretion. The improvement obtained by integrating a pager system into the nurse call systems does not increase care efficiency, because unnecessary visits are still not eliminated. To obtain an immediate response and a purposeful visit by a nurse; regardless of the location of nurse in hospital, traditional systems have to be improved by intelligent telephone system integration. The results of the developed Nurse Call System Software (NCSS), the Wireless Phone System Software (WPSS), the Location System Software (LSS) and the communication protocol are provided, together with detailed XML message structures. The benefits of the proposed system are also discussed and the direction of future work is presented.