Dwi H. Widyantoro

Citation sentence identification and classification for related work summarization

Scientific article summarization is an important problem because it can be of helpful for researchers, particularly for those who start a new research topic. In this paper, we address the problem of related work summarization from scientific papers. The process of summarization comprises of extracting citation sentence followed by classifying the rhetorical category of citation sentence. Citation sentence extraction is performed by combining regular expression-based patterns, co-reference system, evidence-based approach and additional extraction rule. Citation sentence is represented as feature vectors containing term frequency, sentence length, thematic word and cue phrase feature groups. The learning of classification model is explored using Naïve Bayes, Complement Naïve Bayes and Decision Tree. Experiment results reveal that the approaches adopted for citation sentence extraction and rhetorical category classification is promising to provide the ground work for related work summarization.

Fuzzy traffic congestion model based on speed and density of vehicle

The purpose of this paper is to describe a new approach of detecting traffic jam in the city. Unlike prior works that demonstrate traffic jam through vehicle detection only, this paper attempts to introduce a new fuzzy-based method for modeling traffic congestion levels that consider both the road occupancy levels as well as vehicle movement. Our model has an advantage of describing the traffic condition more precisely. We have performed a simulation of the proposed model using Matlab Fuzzy Toolbox and the preliminary results reveal that the proposed model is promising.

Information extraction from scientific paper using rhetorical classifier

Time constraints often lead a reader of scientific paper to read only the title and abstract of the paper, but reading these parts is often ineffective. This study aims to extract information automatically in order to help the readers get structured information from a scientific paper. The information extraction is done by rhetorical classification of each sentence in a scientific paper. Rhetoric information is the intention to be conveyed to the reader by the author of the paper. This research used corpus-based approach to build rhetorical classifier. Since there was a lack of rethorical corpus, we constructed our own corpus, which is a collection of sentences that have been labeled with rhetorical information. Each sentence represented as a vector of content, location, citation, and meta-discourses features. This collection of feature vectors is used to build rhetorical classifiers by using machine learning techniques. Experiments were conducted to select the best learning techniques for rhetorical classifier. Training set consists of 7239 labeled sentences, and the testing set consists of 3638 labeled sentences. We used WEKA (Waikato Environment for Knowledge Analysis) and LibSVM libraries. Learning techniques being considered were Naive Bayes, C4.5, Logistic, Multi-Layer Perceptron, PART, Instance-based Learning, and Support Vector Machines (SVM). The best performers are the SVM and Logistic classifier with accuracy of 0.51. By applying one-against-all strategy, the SVM accuracy can be improved to 0.60.

Traffic lights detection and recognition based on color segmentation and circle hough transform

Automatic detection and recognition of traffic light system is a useful real world application. In this paper, we describe our effort in building such a system. The image of scene obtained from on-vehicle camera is first segmented and converted into three binary images representing images with red, green and yellow color, respectively. Each binary image is then smoothed using Gaussian filter in order to remove noise. The traffic light is detected by finding circular object on the smoothed binary images. We employ Circle Hough transform for circular object detection. The recognition of traffic light is based on the color represented by the binary image with detected circular object. Evaluation of the above method with two real traffic videos demonstrates the effectiveness of the method.

Learning to play Tic-tac-toe

This paper reports our experiment on applying Q Learning algorithm for learning to play Tic-tac-toe. The original algorithm is modified by updating the Q value only when the game terminates, propagating the update process from the final move backward to the first move, and incorporating a new update rule. We evaluate the agent performance using full-board and partial-board representations. In this evaluation, the agent plays the tic-tac-toe game against human players. The evaluation results show that the performance of modified Q Learning algorithm with partial-board representation is comparable to that of human players.

Survey on research paper's relations

This paper discusses research studies related to formation of relations among research papers. From the analysis that has been conducted to some references, it can be concluded that there are two types of research papers relation i.e. a direct relation and indirect relation. The direct relation has the good potential to be further developed than indirect relation. There are three approaches in this relation type i.e. content-based approach, citation analysis approach and citation context approach. The content-based approach cannot accommodate the special characteristics of research papers like citation and author. This approach only suitable implemented in regular documents and only able to identify similarity relations. Specific characteristics such as the citation and the authors can be handled by using citation analysis approach. Focus of the citation analysis approached is increasing the accuracy of the paper relation, where the results obtained so far is good. This approach can identify more diverse relations types such as relations based on citation or author. However, this approach failed to identify relationships between paper based on the meaning contained in text. The relations based on the meaning contained in a text can only be conducted by using citation context approach. Research related to the classification of citation sentences has the big potential to construct the relation based on meaning contained in a text. The construction process of the relation network based on citation context is also proposed in this paper. This process is important in the utilization of the relations in various fields of research. Finally, we also present research opportunity and state of the art research in three main areas, i.e. the recommender system, retrieval system and research map visualization.

System development for research map visualisation

Finding a research gap is an important and difficult step in the research world. To find the research problem, one needs to read and analyze a lot of scientific articles. Although many systems and methods have been develop to assist the above process such as bibliographic analysis & visualization as well as scientific articles recommendation systems, almost all of them are limited to find relevant articles. A tool that helps researchers to get a bigger picture of research in a topic of interests is very desirable. This paper reports our effort in the development of research map visualizers, which we believe will be a useful tool for researchers. Assuming that the detail information about articles is available, the system can visualize a research map according to the user information needs. The user can specify the visualization parameters to explore the various ways of presenting a research map. The visualization parameters include selecting the information along ordinate and abscissa as well as the type of relation among articles. The system has been designed by considering usability and user experience goals. The usability testing for the system that has been conducted reveals that it can attain the goals of effectiveness, learnability, satisfying, being helpful and fun.

Inflammatory Cell Extraction and Nuclei Detection in Pap Smear Images

The automated diagnosis of cervical cancer in Pap smear images is a difficult though extremely important procedure. In order to obtain reliable diagnostic information, the nuclei and their characteristics must be correctly identified and evaluated. However, the presence of inflammatory and overlapping cells in these images complicates the detection process. In this work, a segmentation algorithm is developed to extract the inflammatory cells and enable accurate nuclei detection. The proposed algorithm is based on the combination of gray level thresholding and the definition of a distance rule, which entails in the identification of inflammatory cells. The results indicate that our method significantly simplifies the nuclei detection process, as it reduces the number of inflammatory cells that may interfere.

Video-based system development for automatic traffic monitoring

Traffic jam poses a serious problem because it creates distress and causes inefficiency of fuel consumption. Part of the solution to this problem is a system that can automatically detect the traffic congestion level in a road segment. This paper reports our work on developing such a system. It assumes a fixed camera is attached on high location, enabling it to monitor traffic flow in a specified road segment. The frame of video stream is analyzed by estimating the traffic density and traffic speed from the moving vehicles. The combination of traffic density and traffic speed is used for classifying the traffic congestion level, either belonging to free flow, slow moving or congestion category. Our evaluation using two traffic videos reveal a high accuracy of traffic congestion level prediction.

EXPLOITING UNLABELED DATA IN CONCEPT DRIFT LEARNING

The issues of security in mobile phone in recent days become crucial. Many privacy or secretly data is stored using unsecured protocol or sometimes without the security procedures at all. This will lead to great awareness about security in mobile phone. The effective ways to secure data are steganography and cryptography. The first one concentrate to data hiding in a certain media. In this paper, we present Chaotic Least Significant Bit Encoding (CLSBE) as a steganography method in our system design. The experiment results show that hidden messages in PNG form can be retrieved correctly. The implementation of system in emulator works well but depends on mobile phone features and environment.In computer graphics applications, to produce realistic images, a method that is often used is ray tracing. Ray tracing does not only model local illumination but also global illumination. Local illumination count ambient, diffuse and specular effects only, but global illumination also count mirroring and transparency. Local illumination count effects from the lamp(s) but global illumination count effects from other object(s) too. Objects that are usually modeled are primitive objects and mesh objects. The advantage of mesh modeling is various, interesting and real-like shape. Mesh contains many primitive objects like triangle or square (rare). A problem in mesh object modeling is long rendering time. It is because every ray must be checked with a lot of triangle of the mesh. Added by ray from other objects checking, the number of ray that traced will increase. It causes the increasing of rendering time. To solve this problem, in this research, new methods are developed to make the rendering process of mesh object faster. The new methods are angle comparison and distance comparison. These methods are used to reduce the number of ray checking. The rays predicted will not intersect with the mesh, are not checked weather the ray intersects the mesh. With angle comparison, if using small angle to compare, the rendering process will be fast. This method has disadvantage, if the shape of each triangle is big, some triangles will be corrupted. If the angle to compare is bigger, mesh corruption can be avoided but the rendering time will be longer than without comparison. With distance comparison, the rendering time is less than without comparison, and no triangle will be corrupted.Web index page is well known as page that arranges information by giving the title and short explanation about the information, where the complete information will be presented in other page. However since the amount of information become accumulate, the existence of a lot of index page exactly cause difficulty on getting information because it is possible to direct users into a mount of irrelevant information. Without a system which can help user navigation, the process of seeking the expected information is equal to a trial and error processing. In this paper, web index recommendation system is investigated which involved the activity of user on accessing the index page. This system will arrange the frequent term in index page and then implement Multi Instance Learning to give recommendation of the new index page automatically. The algorithm is citation kNN that will be adapted into fretCit kNN by implementing the minimal Hausdorff distance in measuring the distance. The experiments show that from the several test of users, the system give performance in average recommendation until 82,41% accuracy with 66,71% recall.