Computational Advertising  Homework 3

CS 498: Computational Advertising 
Homework 3

1 General Instructions
• This assignment is due at 11:59 PM on the due date. We will be using Gradescope and Compass to
collect this assignment. The homework MUST be submitted in pdf format on gradescope.
Contact TAs if you face technical difficulties in submitting the assignment. We shall NOT accept any
late submission!
Please make sure to appropriately map/assign the pages of your submitted pdf to each sub-question
listed in the homework outline. Handwritten answers are not acceptable. Name your pdf file as
YourNetid-HW3.pdf
• For all questions, you need to explain the logic of your answer/result for every sub-part. A result/answer without any explanation will not receive any points.
• For the programming question of the assignment, you must submit your code in compass. Name your
compass submission as YourNetid-LastName.zip.
• It is OK to discuss with your classmates and TAs regarding the methods, but it is NOT OK to
work together or share code. Plagiarism is an academic violation to copy, to include text from other
sources, including online sources, without proper citation. To get a better idea of what constitutes
plagiarism, consult the CS Honor code (http://cs.illinois.edu/academics/honor-code) on academic
integrity violations, including examples, and recommended penalties. There is a zero tolerance policy
on academic integrity violations; Any student found to be violating this code will be subject to
disciplinary action.
• Please use Piazza if you have questions about the homework. Also, feel free to send TAs emails and
come to office hours.
Question 1 (3 points)
Explain the difference in auction prices between ads (assuming the ads are in the same
category) shown on the search network and display network.
Question 2 (3 points)
How is a VCG auction for three items (say paintings), different from a VCG auction for
three ad slots?
1
Question 3 (3 points)
Recent research shows that users sometimes prefer ads that did not win the auction for a
slot. Explain possible reasons for this observation.
Question 4 (3 points)
Why does google ad sense manager offer advertisers an opportunity to automatically generate
bids on phrases other than bid phrase chosen by the advertiser (advanced match)?
Question 5 (6 x 3 = 18 points)
In this question, you are required to implement the SimRank algorithm and SimRank with
two forms of evidence weights in a programming language of your choice. In each case, your
iterations must begin with user updates, alternating with ad similarity updates. We will use
this reference[https://arxiv.org/pdf/0712.0499.pdf] for the algorithm details. We provide
two test cases (a) and (b) as input. You can use the sample test case (a), which is smaller
in size, to debug your code. The test case (b) is significantly larger and hence may require
an efficient implementation. Thus, you should use partial sum sharing to speed up your
implementation.
The input consists of weighted User-Ad links. The first line specifies the total number of
links N, followed by N lines each containing three comma separated entries - U, A, S where
U and A are integers representing User and Ad Ids, 0 ≤ U ≤ 1, 000, 000 and 0 ≤ A ≤
1, 000, 000, and score S is a float score value for that link, 0.0 ≤ S ≤ 1, 000.0. The score is
based on how fast the user responded to the Ad (a higher score denotes a greater proclivity).
This is then followed by a single line with two ids, QU and QA. You need to output
the three most similar users to QU and the three most similar ads to QA with each of the
variations of simrank. In case of a tie break, report all users (ads). A sample input looks
like:
N
U1, A1, S1
U2, A2, S2
. . .
UN , AN , SN
QU , QA
2
Note U1 is a user-id, while A1 is the advertisement-id for a specific ad clicked by U1 and
S1 is the link weight. Note that a given user can click more than one ad, meaning multiple
lines could have the same user-id, but no two lines are identical. You are required to build
the bipartite weighted User-Ad graph using the above links.
1. Task One - Simple SimRank iterations: Implement conventional SimRank and
compute the similarities of users to each other and ads to other ads. You need to use
the partial sums trick described in the links provided with the assignment description
in your implementation. Initialize the algorithm with the usual procedure (similarity
of a node to itself is 1 and 0 to all others), and perform K = 10 iterations of User and
Ad-similarity updates (start with user similarity updates). The constants C1 and C2
are both set to 0.8. Let us call the similarity scores obtained after K = 10 iterations
simple simrank scores.
2. Task Two - Incorporate evidence: In section 7 of the reference, two forms are
introduced for evidence scores (geometric in equation 7.3 and exponential in equation 7.4). You will apply each of these forms to the results obtained after 10 iterations of Simple SimRank and obtain the new similarity scores for users and ads.
Let us call these two new sets of scores evidence geometric scores and evidence exponential scores.
Your output should contain 6 lines:
Line 1 - 3 most similar users to QU with simple simrank scores.
Line 2 - 3 most similar ads to QA with simple simrank scores.
Line 3 - 3 most similar users to QU with evidence geometric scores.
Line 4 - 3 most similar ads to QA with evidence geometric scores.
Line 5 - 3 most similar users to QU with evidence exponential scores.
Line 6 - 3 most similar ads to QA with evidence exponential scores.
The sample test case (a) (with input and output) is provided in sample input.txt
and sample output.txt. The large test case (b) on which your code (and results) will be
evaluated, is present in the input b.txt. You should submit your code on compass and the
code outputs in the pdf submitted to gradescope.
3