Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Silence is also evidence: interpreting dwell time for recommendation from psychological perspective
Yin P., Luo P., Lee W., Wang M.  KDD 2013 (Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Chicago, IL, Aug 11-14, 2013)989-997.2013.Type:Proceedings
Date Reviewed: Apr 7 2014

In many social media sites, people who are given an opportunity to vote rarely do so. Yin et al. note: “We call the phenomenon [of] a user silently viewing an item without expressing [his or her] opinions (i.e. giving a vote) ‘silent viewing behavior.’”

In a commercial environment, users encounter dwell time analysis when they visit a web page and examine a specific item on the page. Some web sites will notice a dwell time that exceeds a certain threshold and may send the user an email if nothing is placed in the shopping cart. The email will recommend similar popular items that other customers have purchased.

The authors are interested in dwell time within the context of a social media environment where voting is possible on a specific item. Dwell time is defined by the authors as “the time a user spends on an item.” Dwell time is converted into a user’s “pseudo votes” on items. Pseudo votes are used to enrich the sparse user-vote matrix with the goal of improving recommendation performance.

A viewing-voting (VV) model was developed by analyzing 638,899 records from an iPhone application called JokeBox. Using these records, the authors found that almost 96 percent of people never vote. They note that the voting sparsity problem is caused by the users’ infrequent voting behavior, although they may be quite active in viewing items.

The authors assume that each user has several latent action bounds (LABs). When viewing an item, the user will randomly select a LAB, which together with the quality of the item, jointly affects the user’s voting behavior. If the LAB is greater than the quality of the item, the user is less likely to vote. Conversely, if the LAB is less than the quality of the item, the user is more likely to vote.

“As the experiment shows, the performance of traditional recommendation is greatly improved with the support of our VV model,” the authors conclude. In effect, the viewing-voting model can be used to interpret dwell time; it then uses that information to make a pseudo vote regarding a reader’s reaction to an item.

The authors should consider making the data set that they used for developing their viewing-voting model available on the Internet. Also, the URL that they provide in footnote 2 on page 989 (http://itunes.apple.com/us/app/all-in-1-joke-box-no-ads!/id363494433?mt=8) redirects to the Apple iTunes application, and may not be available to people who do not have iTunes installed.

Reviewer:  W. E. Mihalo Review #: CR142141 (1407-0573)
Bookmark and Share
  Featured Reviewer  
 
Human Factors (H.1.2 ... )
 
Would you recommend this review?
yes
no
Other reviews under "Human Factors": Date
A theory of computer semiotics
Andersen P., Cambridge University Press, New York, NY, 1990. Type: Book (9780521393362)
Aug 1 1992
An experimental comparison of tabular and graphic data presentation
Powers M., Lashley C., Sanchez P., Shneiderman B. International Journal of Man-Machine Studies 20(6): 545-566, 1984. Type: Article
May 1 1985
Organizing for human factors
Thomas J. (ed)  Human factors and interactive computer systems (, New York,461984. Type: Proceedings
May 1 1985
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy