A Cascade Model for Externalities in Sponsored Search
In class, we have learned that the value of an advertisement on a web page is affected by the location it is placed in. For example, an advertisement with a higher value tends to be placed at the most prominent and noticeable slot of a webpage, whereas an advertisement with a lower value is probably placed at the corner of a webpage. However, David Kempe and Mohammad Mahdian from the University of Southern California, Los Angeles figured that the value of an advertisement is also decided by the set of other ads displayed. For example, higher quality ads can distract users from looking at other ads and lower quality ads may cause the users to use interest and close the whole web page. Kempe and Mahdian, therefore, made a model, which records the probability of a user’s decision on whether to continue scanning the webpage after seeing an advertisement (with the assumption that all users scan from top to the bottom, of a web page).
While reading through Kempe and Mahdian’s paper, I realized that my knowledge of information cascade is not sufficient enough to understand to entire progress of how they came up with the algorithm. Overall, they develop their algorithm using the structure of the basic cascade model and added position-dependent multipliers (which is something not mentioned in class).
Source: https://david-kempe.com/publications/cascade.pdf