Soowon Lee

Graph structure of the Korea web

The study of the Web graph not only yields valuable insight into Web algorithms for crawling, searching and community discovery, and the sociological phenomena that characterize its evolution, but also helps us understand the evolution process of the Web. In this paper, we report the experiments on properties of the Korea Web graph with over 116 million pages and 2.7 billion links. This paper presents the power law distributions from the Korea Web and then compares them with other web graphs. Our analysis reveals that the Korea Web graph has different properties in comparison with the other graphs in terms of the structure of the Web.

A clustering validity assessment index

Clustering is a method for grouping objects with similar patterns and finding meaningful clusters in a data set. There exist a large number of clustering algorithms in the literature, and the results of clustering even in a particular algorithm vary according to its input parameters such as the number of clusters, field weights, similarity measures, the number of passes, etc. Thus, it is important to effectively evaluate the clustering results a priori, so that the generated clusters are more close to the real partition. In this paper, an improved clustering validity assessment index is proposed based on a new density function for intercluster similarity and a new scatter function for intra-cluster similarity. Experimental results show the effectiveness of the proposed index on the data sets under consideration regardless of the choice of a clustering algorithm.

Recommending valuable ideas in an open innovation community: A text mining approach to information overload problem

Purpose Open innovation communities are a growing trend across diverse industries because they provide opportunities of collaborating with customers and exploiting their knowledge effectively. Although open innovation communities can be strategic assets that can help firms innovate, firms nonetheless face the challenge of information overload incurred due to the characteristic of the community. The purpose of this paper is to mitigate the problem of information overload in an open innovation environment. Design/methodology/approach This study chose MyStarbucksIdea.com (MSI) as a target open innovation community in which customers share their ideas. The authors analyzed a large data set collected from MSI utilizing text mining techniques including TF-IDF and sentiment analysis, while considering both term and non-term features of the data set. Those features were used to develop classification models to calculate the adoption probability of each idea. Findings The results showed that term and non-term features play important roles in predicting the adoptability of ideas and the best classification accuracy was achieved by the hybrid classification models. In most cases, the precisions of classification models decreased as the number of recommendations increased, while the models’ recalls and F1s increased. Originality/value This research dealt with the problem of information overload in an open innovation context. A large amount of customer opinions from an innovation community were examined and a recommendation system to mitigate the problem was proposed. Using the proposed system, the firm can get recommendations for ideas that could be valuable for its business innovation in the idea generation phase, thereby resolving the information overload and enhancing the effectiveness of open innovation.

A Document Classification System Using Modified ECCD and Category Weight for each Document

Web information service needs a document classification system for efficient management and conveniently searches. Existing document classification systems have a problem of low accuracy in classification, if a few number of feature words is selected in documents or if the number of documents that belong to a specific category is excessively large. To solve this problem, we propose a document classification system using `Modified ECCD` feature selection method and `Category Weight for each Document`. Experimental results show that the `Modified ECCD` feature selection method has higher accuracy in classification than and the ECCD method. Moreover, combining the `Category Weight for each Document` feature value and `Modified ECCD` feature selection method results better accuracy in classification.

Extracting and evaluating topics by region

Analyzing streaming data that contains regional information can derive the interest trends of a region and the differences from those of other regions. The results of analyzing regional differences can be used for making important decisions in areas such as regional marketing and national policy establishment. In this paper, we propose a method to extract topics that represent regional interests from news articles collected by region. The proposed method consists of a novel word-weighting step to extract regional keywords and a word-clustering step to extract regional topics based on the associations between the extracted keywords. The validity of the extracted regional topics is evaluated through a comparison with a ground-truth topic set. Since each topic is represented by a set of words, and a regional topic set is represented by a family of sets, we propose a new clustering validity index for families of sets for a given set of regions. Using the proposed clustering validity index, the optimal parameters for the collected data are presented through experiments.

An automatic classification system for consumer regulatory focus by analyzing web shopping logs

According to regulatory focus theory, a representative theory on consumer behavior, human personality can be divided into two types: promotion and prevention. These two personality types have much influence on the consumers decision in many diverse areas, such as information exploration, information processing, and the evaluation of alternatives. In this research, we try to classify the consumers regulatory focus using web shopping logs as the groundwork for adapting it to personalized recommendation. For this purpose, we define the consumers behavior variables, utilitarian preference index, and information exploration activity index by analyzing the web shopping logs. We then use these variables as inputs to learn a classifier for predicting the consumers regulatory focus. This research shows the possibility of systematization of the consumer behavior theory as an interdisciplinary research of social science and information technology. Based on this attempt, research can be extended to IT services adapting social science theories to a variety of areas, apart from the consumer behavior area.

Predicting personality traits related to consumer behavior using SNS analysis

ABSTRACT Modeling a user profile is one of the important factors for devising a personalized recommendation. The traditional approach for modeling a user profile in computer science is to collect and generalize the users buying behavior or preference history, generated from the users interactions with recommender systems. According to consumer behavior research, however, internal factors such as personality traits influence a consumers buying behavior. Existing studies have tried to adapt the Big 5 personality traits to personalized recommendations. However, although studies have shown that these traits can be useful to some extent for personalized recommendation, the causal relationship between the Big 5 personality traits and the buying behaviors of actual consumers has not been validated. In this paper, we propose a novel method for predicting the four personality traits—Extroversion, Public Self-consciousness, Desire for Uniqueness, and Self-esteem—that correlate with buying behaviors. The proposed method automatically constructs a user-personality-traits prediction model for each user by analyzing the user behavior on a social networking service. The experimental results from an analysis of the collected Facebook data show that the proposed method can predict user-personality traits with greater precision than methods that use the variables proposed in previous studies.

Multi-model or Single Model? A Study of Movie Box-Office Revenue Prediction

Although many studies tried to predict movie revenues in the last decade, the performance and conclusions are conflictive because different data is used. Some studies report that using social data like reviews can obtain the better prediction than using only metadata of movies, but we demonstrate metadata can beat social data in some cases. In this paper, we utilize EM (Expectation Maximization) algorithm to divide movies into several groups, and then for each group we learn one model to predict movie box-office revenue separately. Experimental results show that using multiple models (Multi-model) can obtain more accurate prediction than using a single model (Single-model).

Mobile App Recommendation using User's Spatio-Temporal Context

With the development of smartphones, the number of applications for smartphone increases sharply. As a result, users need to try several times to find their favorite apps. In order to solve this problem, we propose a recommendation system to provide an appropriate app list based on the users log information including time stamp, location, application list, and so on. The proposed approach learns three recommendation models including Naive-Bayesian model, SVM model, and Most-Frequent Usage model using temporal and spatial attributes. In order to figure out the best model, we compared the performance of these models with variant features, and suggest an hybrid method to improve the performance of single models.

Regulatory Focus Classification for Web Shopping Consumers According to Product Type

소비자 행동이론에 따르면 사람의 성향은 향상초점과 예방초점이라는 두 가지 조절초점 유형으로 나누어지며, 이 두 가지 성향은 다양한 영 역에 있어서 소비자의 의사결정에 많은 영향을 미치는 것으로 알려져 있다. 본 연구에서는 개인화 추천에서 Cold Start 문제의 최소화 및 추천 알고리즘 성능 개선을 위하여 조절초점이론을 적용한다. 이를 위하여, 웹쇼핑 로그로부터 소비자 별 행동변수, 정보탐색활동성 지수를 추출하고 이를 활용한 소비자 조절초점성향 분류 방법을 제안한다. 본 연구는 사회과학/IT 융합 연구로서 소비자행동 이론의 시스템화 가능성을 입증하 였다는 점에 있어서 의의를 지니며, 향후 다양한 분야의 이론들을 적용한 IT 서비스에 대한 연구로 확장하고자 한다.