Commenders: A recommendation procedure for online book communities Hyea Kyeong Kim, Hee Young Oh, Ja Chul Gu, Jae Kyeong Kim ⇑ School of Management, Kyung Hee University, 1 Hoegi-dong, Dongdaemun-gu, Seoul, Republic of Korea article info Article history: Received 16 May 2010 Received in revised form 15 March 2011 Accepted 15 March 2011 Available online 21 March 2011 Keywords: Design science research Group recommendation system Online community Term frequency-inverse document frequency weights abstract We propose a recommendation procedure for online book communities called ‘‘Commenders.’’ Its pur- pose is to enhance the effectiveness of community recommendation and also the satisfaction of individ- ual members. The basic idea of our proposed approach is collaborative filtering (CF). It adapts a content- based (CB) filtering algorithm by representing items with keyword features. The proposed recommenda- tion procedure consists of two steps. During the first step, Commenders finds neighbors using community preferences for books and their feature information, and then it generates a CF-based recommendation list. The second step removes irrelevant books from the CF-based list using the keyword preferences of individual members. Commenders is designed to reduce individual member dissatisfaction with the pro- cess of finding desired books within an online community. To evaluate the procedure, we built a proto- type system and performed experiments. Our experimental results show that the proposed system offers higher quality recommendations than the traditional collaborative filtering system. The proposed system has consistently higher precision, and individual members are more satisfied using this system. Ó 2011 Elsevier B.V. All rights reserved. 1. Introduction Online communities are virtual spaces where groups of people with similar interests or purposes can share information and inter- act with each other via the Internet. The majority of Internet users have participated in one or more online communities either directly as interaction members or indirectly as customers of vendors utilizing customer communities (Albors et al. 2008). Communities have an important bearing on the sales of online stores. During an individual consumer’s purchasing decision, the influence from others, especially those with similar experiences or interest, is one major factor (Krelle 1973). Thus, the sharing of opinions and interests of members within an online community are often cited as main forces behind purchasing decisions (Gaertner 1974). Anyone may become a member of an online community, with the most popular online communities competing to attract more members. A success factor for an online community is a high standard of satisfaction of members with information-seeking intentions, which can be achieved by sharing high-quality informa- tion (Arguello et al. 2006). In particular, if there is a system that assists members in easily finding needed information, it can reduce search effort. In other words, it can improve the ‘‘shareability’’ of information in the community. It is a function of the community management system to filter valuable information within a com- munity and provide personalized information to each community member. Since the mid-1990s, recommendation systems have succeeded in supporting Internet users as they search vast collections of infor- mation, products, or services. In particular, collaborative filtering (CF) is recognized as the most successful recommendation tech- nique (Adomavicius and Tuzhilin 2005). It makes recommenda- tions to a user based on other user ratings on items, placing greater weight on ratings from similar users who have similar per- sonal attributes or product preferences. Amazon.com recommends products customers may like by aggregating and analyzing product ratings, clicked products, and purchased products. Other recom- mendation systems predict ratings a user will assign to a particular item. For example, Rate Your Music (rateyourmusic.com) provides a predicted rating for an album along a five-star-scale rating based on previous rating and click information. Although many success- ful systems have been used for recommending books, people, and multimedia content, current systems still require further improve- ments if they are to integrate recommendation capabilities into a community context. This is because interest in recommendation systems has focused mainly on recommending items to individuals rather on recommending activities or items to groups of people intending to consume as a group, such as movies, trip, book clubs, and restaurants (O’Connor et al. 2001). Though there have been some attempts to establish recommendations in a group context, they focus on offline environments. Currently, many group activi- ties and interactions are conducted within online communities, and the process of resolving conflicting group opinions and activi- ties is different from that in offline environments. 1567-4223/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.elerap.2011.03.002 ⇑ Corresponding author. Tel.: +82 2 961 9355; fax: +82 2 967 0788. E-mail address: jaek@khu.ac.kr (J.K. Kim). Electronic Commerce Research and Applications 10 (2011) 501–509 Contents lists available at ScienceDirect Electronic Commerce Research and Applications journal homepage: www.elsevier.com/locate/ecra