Journal of Distance Education Technologies, 3(2), 30-47, April- June 2005 35 inthenextsectionandprovidesthebackthisuser(http://find.slatemsn.com/coder ground for our QSIA system theFray/the Fray. asp) as does the Kazaa P2P system. Social comparison theor SOCIAL RECOMMENDING (Festinger, 1954)distinguishes between SYSTEMS: OVERCOMING physical reality and social reality; while THE“ BLACK BOX”FAD the former(physical reality) is usually based on objective scales of reality with no need of others'perspectives, the latter(social Alan Turings test for an intelligent reality)demands examination of and com- machine--requiring that the system seem parison to others. According to the theory intelligent(Turing 1950)-is often men-(Festinger, 1954), people choose similar tioned by HCI researchers(Moon Nass, ones to participate in the comparison pro- 1998: Nass Moon, 2000). Aspiring to cess as only they can satisfy the function pass Turings test may be a necessary but of itable insufficient condition for recommender sys- recommender systems will emphasize the tems Recommender systems goal is not sense of community while black-box sys process automation, but rather process sup- tems. on the other hand can reflect a feel port and augmentation Recommender sys- ing of a user facing an automatic machine tems need to allow for situatedness and whose sole function is to produce recom- peculiarities of human cognition (Lueg& mendations Landolt, 1998), sometimes with the help of User-dependent procedures refer explanations and reasoning(Herlocker et mainly to the influence a user has over the al.,2000) recommendation procedure and We argue that a large portion of the cally, over the formation of the advising shortcomings of recommender systems can group. Hints to the benefits of controlling be understood as a failure to construct so- the recommendation-providers group had cial recommender systems as opposed to been suggested(Herlocker et al., 2000), black-boxnon-social) ones. Table l out- stating that sometimes the user might wish lines some of the proposed distinctions be- to ignore some members of the neighbors tween these two spectrum-edges of sys- group. Existing approaches in tems recommender systems lack the user's con Social awareness can be encour- trol of who advises him or her; the neigh aged by several means. Some of them are bors group is automatically formed and as- group activities and cooperative learning signed by the system according to algo activities(Selman, 2003). Some systems, rithms(Terveen Hill, 2001) especially consumer-focused and e-com- Explicit ratings and explicit user merce ones, such as Amazon. com and models-Recommendation providers and Consumerreview.com,encouragetheseekersexpresstheiropinionsandprefer awareness of the presence of other users ences on items by various means, forms by focusing on the user's rating and re- and scales. These could be dichotomous- marks (Rafaeli Noy, 2002). Epinions. com likes and dislikes(Pazzani, 1999)or, on an fosters a strong sense of awareness of oth- interval-numeric scale(Herlocker et al ers by making their opinion the item under 2000 Maltz Ehrlich 1995: Shardanand concern. The FRAY community, for ex- Maes, 1995), explicit or implicit(Oard ample, enables the user to track more from Kim, 1998), allowing user comments and opyright 2005, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of ldea Group Inc. is prohibited Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
36 Journal of Distance Education Technologies, 3(2), 30-47, April- June 2005 Table 1. Social vs. black-box recommender tion patterns(e. g, browsing data, purchase systenLs history, etc. )(Breese, Heckerman et al 1998) The cognitive effort to assign ratings acts as disincentive, making it difficult to assemble large user populations and con- tribute to data sparsity (Oard Kim 1998). Implicit feedback techniques seek 9 to infer ratings that a user would assign f from observations available to the system e A positive correlation has been found be tween reading time in USENet and ex- plicit user ratings( Konstan, Miller et al 分总 1997; Morita shinoda, 1994). Construct a ing a profile with implicit preference rat a ings is neither error proof (Herlocker et al 2000)nor free of privacy-concerns (Ramakrishnan, Keller et al 2001) One should note that explicit rating an environment of social awareness, espe- cially in an anonymous community, migl nomenon al Rosenthal 1964); specific recommendations a large number of other anonymous rec 劇都事制复 Commendations, and the effort, responsibil ity, and thought put into it may be reduced by diffusion Data-driven recommender sys. a s tems are best represented by content- based systems; content-based filtering re- fers to analyzing the information stored in Lim Kim, 2001). Examples are key words-based filtering and latent semantic sa relations between the content of the items and the users stored preferences (Shardanand Maes, 1995). This approach is appropriate when rich content informa- pointers(Maltz& Ehrlich, 1995)or just tion is available in articles and Web pages, votes. Implicit voting refers to interpreting for example( Lim Kim, 2001).One ma user behavior or selections to impute a pref- jor advantage is the possibility for yet un- erence and can be based on any informa- seen items to be recommended. The dis Copyright e 2005. Idea Group Inc. Copying or distributing in print or electronic forms without written Reproduced with permission of the copyright owner. Further reproduction prohibited without permission
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission