Siripun Sanguansintukul

A Framework for Tag-Based Research Paper Recommender System: An IR Approach

The Internet and the World Wide Web provide a way to store and share information, especially in academic fields. Community-based research paper sharing systems, such as CiteULike, have become popular among researchers. This paper proposes a framework for a tag-based research paper recommender system. The proposed approach exploits the use of sets of tags for recommending research papers to each user. The preliminary evaluation shows that user self-defined tags could be used as a profile for each individual user. This recommender system demonstrated an encouraging preliminary result with the overall accuracy percentage up to 91.66%.

Selecting Relevant EEG Signal Locations for Personal Identification Problem Using ICA and Neural Network

The problem of identifying a person using biometric data may be of interest. In this paper,EEG signals are used to identify a person as different persons have different EEG patterns. EEG signals can be measured from different locations. Too many signals can degrade the recognition speed and accuracy. A practical technique combining independent component analysis (ICA) for signal cleaning and a supervised neural network for classifying signals is proposed. From 16 EEG different signal locations, three truly relevant locations FP1, P3, and C4 were selected. This selection is based on signals obtained from the subjects at the Comprehensive Epilepsy Program of Chulalongkorn University Hospital, Bangkok, Thailand.

A comparison of search engine using “tag title and abstract” with CiteULike — An initial evaluation

At present, the performance and capabilities of web search engine are vital. Most researchers use search engines for research papers related to their topics of interest. This paper demonstrates the initial comparison search results between a search engine using the indexing method of “tag title and abstract” and the native CiteUlike engine. The retrieval performance of these two search engines are evaluated using mean values of Normalized Discount Cumulative Gain (NDCG). The results illustrate that the search engine with indexer using “tag, title, and abstract” provides better search results as compared to CiteUlike, particularly for those documents in the first two rank which would be considered as the most relevant documents. This primary evaluation in the experiments implies that the chosen heuristic indexer may improve the efficiency of web resource searching on social bookmarking websites.

Clustering analysis of water quality for canals in bangkok, thailand

Two clustering techniques of water quality for canals in Bangkok were compared: K-means and Fuzzy c-means. The result illustrated that K-means has a better performance. As a result, K-means cluster was used to classify 24 canals of 344 records of surface water quality within Bangkok; the capital city of Thailand. The data was obtained from the Department of Drainage and Sewerage Bangkok Metropolitan Administration during 2005-2008. Water samples were collected and analyzed on 13 different parameters: temperature, pH value (pH), hydrogen sulfide (H2S), dissolved oxygen (DO), biochemical oxygen demand (BOD), chemical oxygen demand (COD), substance solid (SS), total kjeldahl nitrogen (TKN), ammonia nitrogen (NH3N), nitrite nitrogen (NO2N), nitrate nitrogen (NO3N), total phosphorous (T-P) and total coliform. The data were analyzed and clustered. The results of cluster analysis divided the canals into five clusters. The information from clustering could enhance the understanding of surface water usage in the area. Additionally, it can provide the useful information for better planning and watershed management of canals in Bangkok.

Improving research paper searching with social tagging — A preliminary investigation

The WWW provides an efficient way to store and share information. Search engines and social bookmarking systems are important tools for web resource discovery. This study investigated three different indexing approaches applied to CiteULike — a social bookmarking system for tagging academic research papers. The indexing approaches here are known as: Tag only; Title with Abstract; and Tag, Title with Abstract. These three indexing approaches were evaluated using mean values of Normalized Discount Cumulative Gain (NDCG). The preliminary results illustrated that indexing using “Tag, Title, with Abstract” performed the best. The initial evaluation on our implementation implied that these designs might improve the accuracy and efficiency of web resource searching on social bookmarking system, not only in academics but also in other domains.

Personal identification by EEG using ICA and neural network

The problem of identifying a person using biometric data is interesting. In this paper, the uniqueness of EEG signals of individuals is used to determine personal identity. EEG signals can be measured from different locations, but too many signals can degrade the recognition speed and accuracy. A practical technique combining Independent Component Analysis (ICA) for signal cleaning and a supervised neural network for classifying signals is proposed. From 16 EEG different signal locations, four truly relevant locations F7, C3, P3, and O1 were selected. This selection can identify a group of 20 persons with high accuracy.

CiteRank: combination similarity and static ranking with research paper searching

Search engines and social bookmarking systems are important tools for web resource discovery. The performance and capabilities of web search engines are vital. This paper proposes CiteRank, a combination of a similarity ranking with a static ranking. Similarity ranking measures the match between a query and a research paper index; while a static ranking, or a query independent ranking, measures the quality of a research paper. For this particular study, a group of factors containing: number of groups contained the posted paper, year of publication, research paper posted time, and priority of a research paper was used to determine a static ranking score. The NDCG was used as an evaluation metric. CiteRank was compared with SimRank and StaticRank. The results of the experiment showed that CiteRank produces a better ranking than the other methods. This implies that CiteRank can improve the effectiveness of research paper searching on social bookmarking websites.

A comparison between the multiple linear regression model and neural networks for biochemical oxygen demand estimations

The most common test for determining the strength of organic content in wastewaters is the biochemical oxygen demand (BOD). The variables of water quality are temperature, pH value (pH), dissolved oxygen (DO), substance solid (SS), total Kjeldahl nitrogen (TKN), ammonia nitrogen (NH3N), nitrate (NO3), total phosphorous (T-P), and total coliform bacteria (T-coliform). These water quality indices affect biochemical oxygen demand. The main objective of this study was to compare between the predictive ability of the neural network (NN) models and the multiple linear regression (MLR) models to estimate the biochemical oxygen demand on data from 288 canals in Bangkok, Thailand. The data were obtained from the Department of Drainage and Sewerage, Bangkok Metropolitan Administration, during 2002–2008. The results showed that the neural network models gave a higher correlation coefficient (R=0.76) and a lower mean square error (MSE=0.0016) than the corresponding multiple linear regression models.

Server virtualization by user behaviour model using a data mining technique — A preliminary study

Server virtualization is the masking of server resources, including the number and identity of individual physical servers, processors, and operating systems, from server users. However, the problem of tuning dynamic resource allocation is a novelty. Managing heterogeneous workloads running within virtual machines is an interesting and challenging topic of server virtualization. This research applied association rule discovery, which is one of the data mining techniques to predict level of user access. The results illustrate that performance of the predictive model for a proxy server is 86.86%. The performance of the predictive model for a web server is 87.18%. Additionally, user behaviors for proxy and web servers are visualized. The results suggest that user behaviors are different in term of workload, day and time usage. This preliminary study may be an approach to improve management of data centers running heterogeneous workloads using server virtualization.

The customer lifetime value prediction in mobile telecommunications

How to treat the customer relationship is a crucial problem in the telecommunications industry. Therefore, how to measure and manage customer lifetime value (CLV) for determining the likely future profit from the customer is very important because the customer is always looking for better and cheaper products and services. The CLV value not only combines with the churn management but also considers the cross-selling and up-selling to allure customer. Earning, not just buying, customerspsila loyalty is now mandatory. Analysis and prediction of customer lifetime value (CLV) methods by using Artificial neural network (ANN) is proposed here. In this paper Multi-Layer Perceptron (MLP) network with Levenberg-Marquardt algorithm is used to predict the CLV, the strategic and operational decisions to retain a customer Lifetime Value in the Mobile Telecommunications industry.