Keunho Choi

A hybrid online-product recommendation system: Combining implicit rating-based collaborative filtering and sequential pattern analysis

Many online shopping malls in which explicit rating information is not available still have difficulty in providing recommendation services using collaborative filtering (CF) techniques for their users. Applying temporal purchase patterns derived from sequential pattern analysis (SPA) for recommendation services also often makes users unhappy with the inaccurate and biased results obtained by not considering individual preferences. The objective of this research is twofold. One is to derive implicit ratings so that CF can be applied to online transaction data even when no explicit rating information is available, and the other is to integrate CF and SPA for improving recommendation quality. Based on the results of several experiments that we conducted to compare the performance between ours and others, we contend that implicit rating can successfully replace explicit rating in CF and that the hybrid approach of CF and SPA is better than the individual ones.

A new similarity function for selecting neighbors for each target item in collaborative filtering

As one of the collaborative filtering (CF) techniques, memory-based CF technique which recommends items to users based on rating information of like-minded users (called neighbors) has been widely used and has also proven to be useful in many practices in the age of information overload. However, there is still considerable room for improving the quality of recommendation. Shortly, similarity functions in traditional CF compute a similarity between a target user and the other user without considering a target item. More specifically, they give an equal weight to each of the co-rated items rated by both users. Neighbors of a target user, therefore, are identical for all target items. However, a reasonable assumption is that the similarity between a target item and each of the co-rated items should be considered when finding neighbors of a target user. Additionally, a different set of neighbors should be selected for each different target item. Thus, the objective of this paper is to propose a new similarity function in order to select different neighbors for each different target item. In the new similarity function, the rating of a user on an item is weighted by the item similarity between the item and the target item. Experimental results from MovieLens dataset and Netflix dataset provide evidence that our recommender model considerably outperforms the traditional CF-based recommender model.

CRM strategies for a small-sized online shopping mall based on association rules and sequential patterns

As dot-com bubble burst in 2002, an uncountable number of small-sized online shopping malls have emerged every day due to many good characteristics of online marketplace, including significantly reduced search costs and menu cost for products or services and easily accessing products or services in the world. However, all the online shopping malls have not continuously flourished. Many of them even vanished because of the lack of customer relationship management (CRM) strategies that fit them. The objective of this paper is to propose CRM strategies for small-sized online shopping mall based on association rules and sequential patterns obtained by analyzing the transaction data of the shop. We first defined the VIP customers in terms of recency, frequency and monetary (RFM) value. Then, we developed a model which classifies customers into VIP or non-VIP, using various data mining techniques such as decision tree, artificial neural network, logistic regression and bagging with each of these as a base classifier. Last, we identified association rules and sequential patterns from the transactions of VIPs, and then these rules and patterns were utilized to propose CRM strategies for the online shopping mall.

Classification cost: An empirical comparison among traditional classifier, Cost-Sensitive Classifier, and MetaCost

Loan fraud is a critical factor in the insolvency of financial institutions, so companies make an effort to reduce the loss from fraud by building a model for proactive fraud prediction. However, there are still two critical problems to be resolved for the fraud detection: (1) the lack of cost sensitivity between type I error and type II error in most prediction models, and (2) highly skewed distribution of class in the dataset used for fraud detection because of sparse fraud-related data. The objective of this paper is to examine whether classification cost is affected both by the cost-sensitive approach and by skewed distribution of class. To that end, we compare the classification cost incurred by a traditional cost-insensitive classification approach and two cost-sensitive classification approaches, Cost-Sensitive Classifier (CSC) and MetaCost. Experiments were conducted with a credit loan dataset from a major financial institution in Korea, while varying the distribution of class in the dataset and the number of input variables. The experiments showed that the lowest classification cost was incurred when the MetaCost approach was used and when non-fraud data and fraud data were balanced. In addition, the dataset that includes all delinquency variables was shown to be most effective on reducing the classification cost.

Building and evaluating a collaboratively built structured folksonomy

Flat folksonomy uses simple tags and has emerged as a powerful instrument for classifying and sharing a huge amount of knowledge on Web 2.0. However, it has semantic problems, such as ambiguous and misunderstood tags. To alleviate such problems, researchers have built structured folksonomies with a hierarchical structure or relationships among tags. Structured folksonomies, however, also have some fundamental problems, such as limited tagging of pre-defined vocabulary and time-consuming manual effort required to select tags. To resolve these problems, we suggested a new method of attaching a tag with its category, which we call a categorized tag (CT), to web content. CTs entered by users are automatically and immediately integrated into a collaboratively built structured folksonomy (CSF), reflecting the tag-and-category relationships supported by the majority of users. Then, we developed a CT-based knowledge organization system (CTKOS), which builds upon the CSF to classify organizational knowledge and enables us to locate appropriate knowledge. In addition, the results of the evaluation, which we conducted to compare our proposed system with the flat folksonomy system, indicate that users perceive CTKOS to be more useful than the flat folksonomy system in terms of knowledge sharing (i.e. the tagging mechanism) and retrieval (i.e. the searching mechanism).

Classification and Sequential Pattern Analysis for Improving Managerial Efficiency and Providing Better Medical Service in Public Healthcare Centers

Objectives This study sought to find answers to the following questions: 1) Can we predict whether a patient will revisit a healthcare center? 2) Can we anticipate diseases of patients who revisit the center? Methods For the first question, we applied 5 classification algorithms (decision tree, artificial neural network, logistic regression, Bayesian networks, and Naïve Bayes) and the stacking-bagging method for building classification models. To solve the second question, we performed sequential pattern analysis. Results We determined: 1) In general, the most influential variables which impact whether a patient of a public healthcare center will revisit it or not are personal burden, insurance bill, period of prescription, age, systolic pressure, name of disease, and postal code. 2) The best plain classification model is dependent on the dataset. 3) Based on average of classification accuracy, the proposed stacking-bagging method outperformed all traditional classification models and our sequential pattern analysis revealed 16 sequential patterns. Conclusions Classification models and sequential patterns can help public healthcare centers plan and implement healthcare service programs and businesses that are more appropriate to local residents, encouraging them to revisit public health centers.

A personalized trustworthy seller recommendation in an open market

Although more and more customers are buying products on online stores, they have a difficulty in selecting a both trustworthy and suitable seller who sells a product they want to buy since there is a plenty number of sellers who sell the same product with different options. Therefore, the objective of this research is to propose a personalized trustworthy seller recommendation system for the customers of an open market in Korea. To that end, we first developed a module which classifies sellers into trustworthy one or not using a classification technique such as decision tree, and then developed another module which makes use of the content-based filtering method to find best-matching top k sellers among the selected trustworthy sellers. Experimental results show that our approach is worthwhile to take. This study makes a contribution at least in that to our knowledge it is the first attempt to recommend sellers, not products as done in most other studies, to customers.

Classification model for detecting and managing credit loan fraud based on individual-level utility concept

As credit loan products significantly increase in most financial institutions, the number of fraudulent transactions is also growing rapidly. Therefore, to manage the financial risks successfully, the financial institutions should reinforce the qualifications for a loan and augment the ability to detect and manage a credit loan fraud proactively. In the process of building a classification model to detect credit loan frauds, utility from classification results (i.e., benefits from correct prediction and costs from incorrect prediction) is more important than the accuracy rate of classification. The objective of this paper is two-fold: (1) to propose a new approach to building a classification model for detecting credit loan fraud based on an individual-level utility, and (2) to suggest customized interest rate for each customer - from both opportunity utility and cash flow perspectives. Experimental results show that our proposed model comes up with higher utility than the fraud detection models which do not take into account the individual-level utility concept. Also, it is shown that the individual-level utility from our model is more accurate than the mean-level utility used in previous researches, from both opportunity utility and cash flow perspectives. Implications of the experimental results from both perspectives are provided.

Development of Prediction Model for Prevalence of Metabolic Syndrome Using Data Mining: Korea National Health and Nutrition Examination Study

The purpose of this study is to investigate the attributes influencing the prevalence of metabolic syndrome and develop the prediction model for metabolic syndrome over 40-aged people from Korea Health and Nutrition Examination Study 2012. The researcher chose the attributes for prediction model through literature review. Also, we used the decision tree, logistic regression, artificial neural network of data mining algorithm through Weka 3.6. As results, social economic status factors of input attributes were ranked higher than health-related factors. Additionally, prediction model using decision tree algorithm showed finally the highest accuracy. This study suggests that, first of all, prevention and management of metabolic syndrome will be approached by aspect of social economic status and health-related factors. Also, decision tree algorithms known from other research are useful in the field of public health due to their usefulness of interpretation.

Assignment of collaborators to multiple business problems using genetic algorithm

As firms encounter new problems in the fast-changing business environment, they have to find collaborators with problem-solving expertise. Since this optimization problem takes place in a firm as the business environment changes, genetic algorithm (GA), which has shown outstanding performance in obtaining a sub-optimal solution relatively quickly, seems to be the right solution, one that is superior to goal-programming, multi-attribute decision making, and branch and bound. We therefore propose a GA-based approach to solving the problem of assigning collaborators to multiple business problems. Our solution worked well in several experiments.