Our first presentation.
We talk about the motivation behind our project, our objective, what we have already completed, our work plan for the rest of the term, of whom our project will be of use to, and some additional related readings.
2. Background
The "birth" of a bill:
-> Proposed
-> Debate
-> House/Senate vote
-> Amendment
-> Vote again
-> (infinite repetition)
-> President signature
3. Motivation
● Process of passing a bill is quite difficult.
○ Majority of proposed bills are rejected.
○ 112th congress passed just 80 bills, lowest since
1947.
○ A typical congressman may sponsor 100 bills,
passing 0.
4. Motivation (continued)
● It would be great if we could tell if a bill
would be approved before it was formally
proposed.
○ This would save a lot of time for people trying to
put forth bills.
○ Potential "profit" from knowing the outcome
beforehand.
5. Objectives
● Want to determine if a proposed bill has no
chance of passing. In this case, avoid
proposing the bill altogether, save time.
● If a proposed bill will be definitely
approved...
● For those bills whose outcome is uncertain
...
○ (next slide)
6. Objectives (continued)
● For a bill which will not obviously fail but is
not projected to be approved, determine if
there are any "influential congressmen"
whose change in vote will change the
outcome of the bill!
○ This "swing" congressman changing his vote will
likely result in a cluster of other congressmen
changing their votes.
○ If a particular congressman is "important", we can
attempt to "lobby" (or bribe) him to change his
vote.
7. Previous Work
● Applying GMDH Algorithm to Extract Rules
from Examples (Fujimoto and Nakabayashi)
○ Machine Learning algorithm
○ Has been applied to congressmen's voting records in
the past to determine whether they were
republican or democrat
○ Can be generalized to other dividing issues
○ Difficult to apply to details in bills
● We are doing a different approach - more
graph-oriented
8. Previous Work
● The Link-Prediction Problem for Social
Networks (Nowell and Kleinberg)
○ Described methodology and equations used to
predict future social interactions based on a picture
of a social network.
○ Mainly used closeness centrality to do this but
focused on other properties as well.
● There are quite a lot of other interesting
papers which can give us ideas!
9. Steps to meet our objective
(continued)
● Acquire data for congressman
○ Given a congressman, need list of bills "aye'd",
"nay'ed" and "abstained".
○ Edge set: given congressman A, how many times did
A and B vote the same. What fraction did they vote
the same?
● Information about the proposed bill
○ Is the bill regarding healthcare, transportation,
infrastructure, etc. ?
○ Congressman's votes will depend on the type of bill.
10. Steps to meet our objective
● To graphically model the mutual influences
between congressmen and predict the
outcome.
○ Probability of voting "yes" given other's probability
distribution over votes -> outcome of the bill
○ Exactly what Probabilistic Graphical Models do!
○ Edges between congressmen are determined by the
percentage of time they agree.
○ Possible asymmetric influence as well, could result
in directed edges.
■ i.e. one congressman has influence over another
but not vice versa.
11. Steps to meet our objective
(continued)
● Predict the chance of passing the bill.
○ Need to use our constructed graph.
○ And use Probabilistic Graphical Models
● Test our prediction on data set and
modify/tune our model accordingly
12. Our Users
● Politicians
○ Best-case scenario our project will revolutionize
the way in which bills are proposed.
● Our network model can be generalized to
all types of decision making by voting!
○ Predict bills in state or regional governments.
○ Predict who will win elections with similar network
structure as our model.
● Any binary type decision making
13. Work Plan (Subject to change)
● Week 1: Gather data
○ Have the data easily accessible, i.e. given an ID
we can easily find the edges of a congressman,
the bills he approved of, his party affiliation,
etc.
(already completed!)
● Weeks 2-3: Create a graphical model
○ Create criterion for the edge set
○ Implement a basic prediction algorithm
14. Work Plan (continued)
● Weeks 4-6: Modify/tune the model
○ Experiment with different asymmetric edges
○ Adding the influence of the type of the bill
● Weeks 7-9: Implement an algorithm to
identify important "swing" voters
○ Based on importance: how their change in votes
will affect the rest of the congress
○ Difficulty to swing: their probability of voting
"yes/no", their past campaign donation amount
● Weeks 9-10: Wrap up and final report
○ Test our algorithm on undecided bills in reality!
○ Write a final paper based on our findings
15. What have we done?
● Acquired Data
○ List of bills from 2010-2012, and votes ("aye", "nay" or
"abstain") for both the Senate and the House.
○ For each Congressman A, we have gathered the number
of times Congressman A agrees with Congressman B and
the total number of bills A and B both voted on.
■ Surprisingly good data, not completely partisan as
we first expected.
■ Also gathered other information about a
congressman, such as party affiliation, number of
terms, etc.
● Have a potential edge set
○ The percentage of times congressman agree.
16. What we need to do?
● Need to complete weeks 2 - 10 :)