Page 143 - Special Topic Session (STS) - Volume 1
P. 143
STS420 Jianhua (Klyment) H.
Before further processing the above data, necessary computation will
be done such as normalization and noise rejection.
3.3.2 Metrics selection
For an individual credit score, different companies have different
weights on the metrics of the rating model. Typically, credit repaying
history will contribute the most, followed by transaction records in terms
of volume and frequency.
3.3.3 Univariate analysis
Through univariate analysis, we can find out the projection maps
between univariate and default rate. For example, a continuous overdue
of credit card bills may be more likely to indicate the individual is a
defaulter compared to his late confirmation of order upon received.
3.3.4 Model fitting and parameter estimation
For example, a logistic regression model using forward feature
selection. The parameter estimation of the model usually adopts the
optimization method such as the least squares method, maximum
likelihood estimation, and maximum posterior probability.
Also, for a commercial enterprise like Tencent, the social network data is
a featured source which they can apply in credit score modeling in two
ways, feature development, and credit score correction.
Feature development consists of two essential metrics, self-network
structure and friends network structure. Selfnetwork structure mainly
refers to the user’s network topology, which can be described using some
features. For example, the user’s reading behavior can reveal his latest
interest which can be taken into consideration when issuing micro-loan.
The friend’s network structure is a look-like analysis model which may
reflect a user’s potential behavior. For instance, a friend’s packet money
interaction and group package money interaction will conclude some
relative effects based on their demographics.
3.4 Applications of Personal Credit Score
Generally, there are several mainstream services which personal credit score
is applied including shopping, renting, accommodation, recycling,
transportation, telecommunication, and micro-finance. In this paper, we
mainly discuss the application of renting and micro-finance.
Since Chinese mobile phones users spend much time on the handset such
as watching videos, playing games, a typical scenario is running out of battery
when they hang out with friends. In the past, they may need to carry a power
bank along which makes it inconvenient but now thanks to companies like
JieDian; there are many power bank renting shops in shopping malls,
restaurant, and cinemas. The process of borrowing a power bank is quite
straightforward. Firstly, use an app like WeChat to scan the QR code on the
power bank machine, then agree to the terms and conditions, pay the deposits
132 | I S I W S C 2 0 1 9