Abstract
Internet auctions are more than just electronic commerce; they have also become crucial for actual commerce. Because such activity occurs on the Internet, we need a way to determine people's reputations for such commerce. Thus reputation mechanisms are being developed and are widely being used. Most existing reputation mechanisms are based on simple numerical evaluation mechanisms in which users provide feedback as ratings, stars, and evaluation values that usually range from 0 to 5. These scores are totaled, and this sum is assumed to represent the user's reputation. The following problems are obvious: First, the score total might lack or overlook some important information. Second, there is no score propagation of values, even if relations exist among people on the network. On the other hand, for example, reputation mechanisms for websites, including PageRank and HITS, utilize the characteristics of links (hyperlinks) to evaluate a node (a page) by the weights on it (the page). We believe trading networks have similar characteristics, enabling us to build more effective reputation mechanisms. In this paper we investigate several new reputation mechanisms called Auction Network Trusts (ANTs) for defining user reputations in trading networks. We propose five types of ANTs: ANT(HITS), tANT, rANT, mANT, and dANT. Our aim is to analyze these new mechanisms with actual online auction datasets to determine whether they can effectively utilize the trading links between sellers and buyers. We show our findings of these new reputation mechanisms from our current experimental results. These findings are well-known abstract-level properties in the field of reputation mechanisms. By using ANT mechanisms, we can find such findings in real world auction data. What we can get by the ANT mechanism is technically sound, and ANTs can be applicable to real world auction data.