An Architecture for Behavior-Based Library Recommender Systems

摘要: Library systems are a very promising application area for behavior-based recommender services. By utilizing lending and searching log files from online public access catalogs through data mining, customer-oriented service portals in the style of Amazon.com could easily be developed. Reductions search evaluation costs documents readers, as well an improvement customer support collection management librarians, some possible benefits. In this article, architecture distributed services based on stochastic purchase incidence model is presented. Experiences with that has been operational within scientific library system Universitat Karlsruhe since June 2002 described. ********** Almost all libraries feature electronic systems. With their (OPACs), they possess requirements almost same manner digital value-added A add-on traditional systems, necessity which arises need scientists students efficient literature research, shown by survey Klatt et al. (1) Due to--among other things--information overload difficult quality assessment, information seekers more incapable compiling relevant conventional database-oriented catalog time-efficient manner. Therefore, reveals, rely heavily peers recommendations. Considering tight schedule many students, university teachers, researchers, it worth effort to free up valuable time consumed steering each standard fields, done expert advice Moreover, scenario, can also profit combined knowledge users contrast restricted personal networks. Consumer acceptance convenience huge success broad variety different offered at commercial bookstore sites (such Amazon.com). People getting used these appreciate them. So question ask is: Why not broader scale libraries? Discussing librarians computer scientists, following reasons were discovered: * Privacy. Librarians considerate privacy patrons. Transaction-level reading histories must protected. Budget restrictions. Public general run under budget New millions might require prohibitively high additional technology (IT)-investments. Data size. The number contained or academic least one order magnitude higher than most organizations. This implies transaction-level scattered documents. While would expect better chance finding meaningful patterns, becomes increasingly detect patterns due sparsity, because computational complexity counting such association rules exponential objects. Standard association-rule algorithms reduce deleting objects do receive sufficient support. context, sparsity data, unfortunately, makes approach feasible. Increasing threshold will lead pruning but weak may below threshold, still statistically significant. article presents strategy overcome obstacles recommendations efficiently generated anonymous session off-the-shelf PC …