Oard, D., & Kim, J. (1998). Implicit feedback for recommender systems. In Proceedings of the AAAI workshop on recommender systems, Madison, WI, USA (pp. 80--82).  Home/Search   Document Details and Download   Summary   Related Articles   Check

....content based techniques, these systems depend on their users providing ratings or feedback. The scarcity of ratings and scalability are both issues in social filtering. User profiles in this case are usually sparse vectors of ratings. A partial solution to this might be to use implicit feedback [11], or methods to increase the density of the dataset. Scalability is a problem because computation grows linearly with the number of users and items. Finding the nearest neighbors to the active user in order to retrieve recommendations is a task that requires the definition of the term ....

Oard, D.W., Kim, J.: Implicit feedback for Recommender Systems. In: Proceedings of the AAAI Workshop on Recommender Systems. Madison, WI (1998) 81-83

....for users to spend a greater length of time reading those articles rated as interesting, as opposed to those rated as not interesting. This finding has been replicated by others in similar environments [4] Other behaviors that have been explored include printing, saving, scrolling and bookmarking [6]. The work reported here is an explicit test of the work of [5] in an IR context other than information filtering. Three sources of implicit feedback are of particular interest: reading time per document, scrolling and interaction. The specific hypotheses for this study are, accordingly: H1: ....

Oard, D. W., & Kim, J. (1998). Implicit feedback for recommender systems. Proceedings of the AAAI Workshop on Recommender Systems. http://www.glue.umd.edu/~oard/research.html.

....query, which in turn is used to re rank the list of top ranking sentences. Specifically, our notion of context is based on the assumption that searchers will spend a longer time reading interesting (potentially relevant) material and less time reading irrelevant material. Several studies, e.g. [7], have found a correlation between the positive relevance of a document and viewing time. These studies, focusing on static corpora and full text documents, have found that if a document is subjected to a lot of read wear [4] it is likely to be relevant. In this study, we assume that these ....

D. Oard and J. Kim. Implicit feedback for recommender systems. Proceedings of the AAAI Workshop on Recommender System. Madison, Wisconsin. 1998.

.... [Den82] has been used to describe a variety of processes involving the delivery of information to users [BC92, SM93] Information filtering is an information seeking process in which documents are selected from a stream of incoming data to satisfy a relatively stable and specific information need [OK98] The goal of an information filtering system is to sift through large volumes of dynamically generated information and to present users with information likely to satisfy their information requirement [OM96] Information filtering is related to processes such as routing (with a heritage in ....

....and unobtrusively. This approach is called implicit feedback [HHWM92, HRS94, Nic97, AZ97] When users deal with objects, e.g. when users examine, retain, or reference objects, the system may observe this user behavior and use this data to estimate how the user would rate the respective objects [OK98, p.81] see also Section 5.5.3) Related work shows, for example, that the amount of time users spend on reading a news articles can be a good predictor for the relevance of that article [MS94] Implicit feedback helps reducing the user effort and is therefore an attractive alternative to ....

[Article contains additional citation context not shown here]

D.W. Oard and J. Kim. Implicit feedback for recommender systems. In

....ratings may be used in several ways: the rst is to provide more ratings upon which to base predictions, and the second is as a check on explicit ratings to decide when to ignore them or not. We propose to provide experimental evaluation of the e ectiveness of implicit ratings. Oard and Kim [14] build upon work by Nichols [13] by categorizing implicit ratings, dividing them into Examination , where a user studies an item, Retention where a user saves an item for later use, and Reference where a user links all or part of an item into another item. They suggest two strategies for ....

....In addition to browsing the Web at large that we present here, we have considered casually reading an online newspaper, looking up a topic in an online encyclopedia, and searching for information using a search engine. There are many more implicit interest indicators present in other literature [13, 14], such as bookmarking or printing, that need to be empirically evaluated as we have begun to do for time and mouse activity. 9. ACKNOWLEDGMENTS We would like to acknowledge the help of Professor Isabel Cruz for encouraging her students to participate in our user study, as well as the students ....

D. Oard and J. Kim. Implicit Feedback for Recommender Systems. In Proceedings of the AAAI Workshop on Recommender Systems, July 1998.

....to enter explicit ratings can alter normal patterns of browsing and reading; 2) unless users perceive that there is a benefit providing the ratings, they may stop providing them. Implicit rating is much more difficult to determine. Oard and Kim divide implicit ratings into three categories [11]: rating based on examination, when a user examines an item; rating based on retention, when a user saves an item; rating based on reference, when a user links all or part of an item into an other item. 6 How can user preferences with implicit ratings be determined Some criteria were ....

Oard, D., Kim, J. Implicit Feedback for Recommender Systems. In Proceedings of AAAI Workshops on Recommender Systems. July 1998.

....can be augmented with a more fine grained measure of relevancy obtained from read time data. The time a user spends reading a job description has been shown to correlate with that user s degree of interest, Goecks and Shavlik, 1999; Konstan et al. 1997; Morita and Shinoda, 1994; Nichols, 1997; Oard and Kim, 1998; Sarwar et al. 1998] For this reason, CASPER also calculates read time information from the server logs by noting the time difference between successive requests by the same user. Again, a suitable thresholding technique is necessary to eliminate spurious read times due to a user logging off ....

D. Oard and J Kim. Implicit feedback for recommender systems. In Proceedings of AAAI Workshop on Recommender Systems, Madison, Wisconsin, USA, July 1998.

....ratings may be used in several ways: the rst is to provide more ratings upon which to base predictions, and the second is as a check on explicit ratings to decide when to ignore them or not. Our research provides experimental evaluation of the e ectiveness of implicit ratings. Oard and Kim [18] build upon work by Nichols [17] by categorizing implicit ratings, dividing them into Examination , where a user studies an item, Retention where a user saves an item for later use, and Reference where a user links all or part of an item into another item. They suggest two strategies for ....

....pro le is built from implicit or explicit interest indicators. Not for distribution or attribution: for review purposes only. 21 8 Future Work In this work, we have considered only implicit interest indicators alone. There are many more implicit interest indicators present in other literature [17, 18], such as bookmarking or printing, that need to be empirically evaluated. Combinations of interest detectors, such as time spent on a Web page and the amount of scrolling, may prove to be more accurate than any indicator alone. Implicit interest indication may be combined with more explicit ....

D. Oard and J. Kim. Implicit Feedback for Recommender Systems. In Proceedings of the AAAI Workshop on Recommender Systems, July 1998.

....in those new concepts. This step can be done either periodically or when user profiles are updated for other reasons. 5.2. User Profile We now describe the data structure and the update algorithm for the user profile. Building a user profile lies in the area of user modeling (see for instance, [23, 26, 28]) 5.2.1. Contents of User Profile A user profile includes the following data: User unit: department to which the user belongs. Personal information: name and e mail address of user. Interests information: disciplines (U Disc) and topics (U Topics) of interest to the user. User ....

....of a pilot system. Our study is only a first step towards the development of more dynamic communication mechanisms for organizational information distribution. In future research, we foresee several possible directions as follows: Exploring ways of receiving implicit feedback from users [23] and incorporating it into the user profile on a continuous basis. Applying the concepts and framework in other, non academic environments. It is likely that the specifics of the information management process in non academic institutions may be significantly different from academia and ....

Oard, D.W. and Kim, J.,"Implicit Feedback for Recommender Systems," Proceedings of the AAAI Workshop on Recommender Systems, Madison, WI, July, 1998.

....revisit and thus both clicks are counted. 3.2 Read Time Data Coarse grained revisit data can be augmented with a more ne grained measure of relevancy obtained from readtime data. The time a user spends reading a job description has been shown to correlate with that user s degree of interest, [2 6, 9]) For this reason, CASPER also calculates readtime information from the serverlogs by noting the time di erence between successive requests by the same user. Again, a suitable thresholding technique is necessary to eliminate spurious readtimes due to a user logging o or leaving her terminal. ....

Oard, D., and Kim, J. Implicit feedback for recommender systems. In Proceedings of AAAI Workshop on Recommender Systems (Madison, Wisconsin, USA, July 1998).

....project. This user judgments can be given explicit by or implicit. Explicit means the user markes a document relevant or not relevant. Implicit means the user is examinated, his her behavior is observed within the EG system, for example if the user reads a document or markes part of a text. In [Douglas W. Oard and Jinmook Kim 98] are di erent sources of implicit feedback surveyed. In [Rocchio 71] is a description based on the vector space model given. By reformulation of the query via relevance feedback a near to optimal query corresponding to a particular set of documents can be produced. So given the original ....

Douglas W. Oard and Jinmook Kim. (1998). Implicit Feedback for Recommender Systems. In: Proceedings of the AAAI Workshop on Recommender Systems, Madison, WI, July, 1998.

....ratings may be used in several ways: the first is to provide more ratings upon which to base predictions, and the second is as a check on explicit ratings to decide when to ignore them or not. We propose to provide experimental evaluation of the effectiveness of implicit ratings. Oard and Kim [OK98] build upon work by Nichols [Nic97] by categorizing implicit ratings, dividing them into Examination , where a user studies an item, Retention where a user saves an item for later use, and Reference where a user links all or part of an item into another item. They suggest two strategies ....

....to browsing the Web at large that we present here, we have considered casually reading an online newspaper, looking up a topic in an online encyclopedia, and searching for information using a search engine. There are many more implicit interest indicators present in other literature [Nic97, OK98] such as bookmarking or printing, that need to be empirically evaluated as we have begun to do for time and mouse activity. NOTES We would like to acknowledge the help of Professor Isabel Cruz for encouraging her students to participate in our user study, as well as the students who ....

D. Oard and J. Kim. Implicit Feedback for Recommender Systems. In Proceedings of the AAAI Workshop on Recommender Systems, July 1998.

....(1997) sought to construct a comprehensive view of implicit feedback, with a focus on its use in information filtering systems. He presented a list of potentially observable behaviors; adding purchase, assess, repeated use, refer, mark, glimpse, associate, and query to those mentioned above. Oard and Kim (1998) extended that work, organizing observable behaviors into three broad categories: examination, retention, and reference. In the next section, we present a further refinement of that framework, adding an additional category (annotation) that results in unification of implicit and explicit feedback ....

Oard, D.W., and Kim, J. (1998) Implicit Feedback for Recommender Systems. In AAAI Workshop on Recommender Systems, Madison, WI: 81-83.

No context found.

Oard, D., & Kim, J. (1998). Implicit feedback for recommender systems. In Proceedings of the AAAI workshop on recommender systems, Madison, WI, USA (pp. 80--82).

No context found.

Oard, D., Kim., J., Implicit Feedback for Recommender Systems. In Kautz, H., (ed.), Papers from the 1998 AAAI Workshop on Recommender Systems, Technical Report WS-98-08, The AAAI Press, 1998, 80-82.

No context found.

Oard, D. and Kim, J. Implicit Feedback for Recommender Systems. Proceedings of the AAAI Workshop on Recommender Systems (AAAI `98) Madison, Wisconsin, 26-30 July (1998)

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC