Yang Sok Kim

Collaborative filtering for people to people recommendation in social networks

Predicting people other people may like has recently become an important task in many online social networks. Traditional collaborative filtering approaches are popular in recommender systems to effectively predict user preferences for items. However, in online social networks people have a dual role as both “users” and “items”, e.g., both initiating and receiving contacts. Here the assumption of active users and passive items in traditional collaborative filtering is inapplicable. In this paper we propose a model that fully captures the bilateral role of user interactions within a social network and formulate collaborative filtering methods to enable people to people recommendation. In this model users can be similar to other users in two ways – either having similar “taste” for the users they contact, or having similar “attractiveness” for the users who contact them. We develop SocialCollab, a novel neighbour-based collaborative filtering algorithm to predict, for a given user, other users they may like to contact, based on user similarity in terms of both attractiveness and taste. In social networks this goes beyond traditional, merely taste-based, collaborative filtering for item selection. Evaluation of the proposed recommender system on datasets from a commercial online social network show improvements over traditional collaborative filtering.

Knowledge Management and Acquisition for Smart Systems and Services

The last decade has seen an increasing interest in the use of 3D virtual environments for educational applications. However, very few studies investigated the influence of the learning context, such as class type and learning type, on learners’ academic performance. This paper studied the impact of class type (i.e. comprehensive or selective) classes, as well as learning type (i.e. guided or challenge and guided), on students’ level of usage of a Virtual Learning Environment (VLE) as well as on their academic performance. The results showed that, unlike class type, there is a significant difference between learners’ in their usage of the VLE. Moreover, the results showed that the levels of using a VLE significantly correlated with learners’ academic performance.

Adaptive Web document classification with MCRDR

With the explosive increase in Web based information, the need for an intelligent agent for automatic classification has also been increased resulting in many research discoveries in this area. Machine learning (ML) based document classification is now the prevalent approach. However, classification by ML may not keep the same performance because the knowledge generated from the training set may not be appropriate for certain types of Web information. People are often concerned more about the newly uploaded information such as Web based online news than information already available. This explains why it is not widely used in real applications. However, the manual classification method, by the domain users, cannot be a solution either until the knowledge acquisition bottleneck issue is resolved. Multiple classification ripple down rules, an incremental knowledge acquisition method, is suggested to overcome this problem with fast learning and low cost maintenance.

Interaction-based collaborative filtering methods for recommendation in online dating

We consider the problem of developing a recommender system for suggesting suitable matches in an online dating web site. The main problem to be solved is that matches must be highly personalized. Moreover, in contrast to typical product recommender systems, it is unhelpful to recommend popular items: matches must be extremely specific to the tastes and interests of the user, but it is difficult to generate such matches because of the two way nature of the interactions (user initiated contacts may be rejected by the recipient). In this paper, we show that collaborative filtering based on interactions between users is a viable approach in this domain. We propose a number of new methods and metrics to measure and predict potential improvement in user interaction success, which may lead to increased user satisfaction with the dating site. We use these metrics to rigorously evaluate the proposed methods on historical data collected from a commercial online dating web site. The evaluation showed that, had users been able to follow the top 20 recommendations of our best method, their success rate would have improved by a factor of around 2.3.

Dynamic Web content filtering based on user's knowledge

This paper focuses on the development of a maintainable information filtering system. The simple and efficient solution to this problem is to block the Web sites by URL, including IP address. However, it is not efficient for unknown Web sites and it is difficult to obtain complete block list. Content based filtering is suggested to overcome this problem as an additional strategy of URL filtering. The manual rule based method is widely applied in current content filtering systems, but they overlook the knowledge acquisition bottleneck problems. To solve this problem, we employed the multiple classification ripple-down rules (MCRDR) knowledge acquisition method, which allows the domain expert to maintain the knowledge base without the help of knowledge engineers. Throughout this study, we prove the MCRDR based information filtering system can easily prevent unknown Web information from being delivered and easily maintain the knowledge base for the filtering system.

Reciprocal and heterogeneous link prediction in social networks

Link prediction is a key technique in many applications in social networks, where potential links between entities need to be predicted. Conventional link prediction techniques deal with either homogeneous entities, e.g., people to people, item to item links, or non-reciprocal relationships, e.g., people to item links. However, a challenging problem in link prediction is that of heterogeneous and reciprocal link prediction, such as accurate prediction of matches on an online dating site, jobs or workers on employment websites, where the links are reciprocally determined by both entities that heterogeneously belong to disjoint groups. The nature and causes of interactions in these domains makes heterogeneous and reciprocal link prediction significantly different from the conventional version of the problem. In this work, we address these issues by proposing a novel learnable framework called ReHeLP , which learns heterogeneous and reciprocal knowledge from collaborative information and demonstrate its impact on link prediction. Evaluation on a large commercial online dating dataset shows the success of the proposed method and its promise for link prediction.

Experience with long-term knowledge acquisition

Evaluation has remained a major challenge for knowledge acquisition and little data is available on how experts actually use knowledge acquisition technology. A number of companies offer Ripple-Down Rules to enable on-going knowledge acquisition and maintenance while a system is in use. One of these companies, Pacific Knowledge Systems , has logged user activity over a number of years. Data from these logs demonstrate that domain experts continue to add knowledge to a knowledge base over years. The logs also demonstrate that new knowledge can be added very rapidly regardless of knowledge base size or age. We assume that the on-going knowledge acquisition observed was driven by the need to make changes and encouraged and allowed by the ease of the knowledge acquisition technology used. The question arises of whether experts in other domains would also chose to continue to add knowledge to their knowledge bases if this was supported.

Incremental Knowledge Management of Web Community Groups on Web Portals

The concept of the web portal was introduced in around 1998 when the web became a standard medium for accessing information. While HTML-based static web pages were also popular, people used the search engine websites, or specific web pages, such as the personal web page or the web browser company default page, as their web portals. Since their inception, providing information for users has been the most important function of web portals, and many of them try to provide adapted information to different users. Offering this level of service is difficult because of the quantity of information and the various types of information classification for different user groups involved. In most web portals, the collection and classification of the information is still carried out manually. Automation of this task requires domain-specific classification knowledge, which is not easy to acquire. Automated web information management and publication system has been developed using the Multiple Classification Ripple Down Rules (MCRDR) knowledge acquisition engine. Various prototype web portals are being developed and the evaluation study proves the potential of the out-of-box style web portal generation tool for the adapted service.

RDR-based open IE for the web document

The Web contains a massive amount of information embedded in text and obtaining information from Web text is a major research challenge. One research focus is Open Information Extraction aimed at developing relation-independent information extraction. Open Information Extraction (OIE) systems seek to extract all potential relations from the text rather than extracting a few pre-defined relations. Existing OIE systems such as TEXTRUNNER usually take a machine learning based approach which requires large volumes of training data. This paper presents a Ripple-Down Rules Open Information Extraction system based on processing example cases and manually adding rules when needed. The key advantages of this approach are that it can handle the freer writing style that occurs in Web documents and can correct errors introduced by natural language pre-processing tools, whereas systems like TEXTRUNNER depend on the quality of the entity-tagging preprocessing in the training data. We evaluated the Ripple-Down Rules approach against the OIE systems, TEXTRUNNER and StatSnowball. In these studies the Ripple-Down Rules approach, with minimal low-cost rule addition achieves much higher precision and somewhat improved recall compared to these other Open Information Extraction systems.

An Automated WSDL Generation and Enhanced SOAP Message Processing System for Mobile Web Services

Web services are key applications in business-to-business, business-to-customer, and enterprise applications integration solutions. As the mobile Internet becomes one of the main methods for information delivery, mobile Web services are regarded as a critical aspect of e-business architecture. In this paper, we proposed a mobile Web services middleware that converts conventional Internet services into mobile Web services. We implemented a WSDL (Web Service Description Language) builder that converts HTML/XML into WSDL and a SOAP (simple object access protocol) message processor. The former minimizes the overhead cost of rebuilding mobile Web services and enables seamless services between wired and wireless Internet services. The latter enhances SOAP processing performance by eliminating the Servlet container (Tomcat), a required component of typical Web services implementation. Our system can completely support standard Web services protocol, minimizing communication overhead, message processing time, and server overload. Finally we compare our empirical results with those of typical Web services