Sexual Prediction based on e-commerce information using counting

Sexual Prediction based on
e-commerce information
using counting with weight
method
Pham Ngoc An
FPT University – Hoa Lac Hitech Park – Ha Noi – Viet Nam
Overview
Introduction to the method
What is counting?
How is weight?
Re-using for improvement
Introduction to the method
Frequency chart of numbers product in sessions
Introduction to the method
In applying Naïve Bayes, Neuron Network,…
model will lead to the fact that the number of
the feature is overcrowded.
Because of the lost of information caused by the
falling of the data samples in extracting
features.
Introduction to the method
In fact, people will be attracted to products
which is suitable for their sexes
The information that is divided into groups
(clustered products)
Count and compare the numbers of different
types of product, predict the sex of user based
on their interest for the clustered products
What is counting?
Counting number of man/woman accesses to products
What is counting?
Counting number of man/woman accesses to catalogues
How is weight?
 Since the distribution of labels in the data is not
balanced so WeightFemale and WeightMale
 Sex-value of a Product: V(Pi) = *Number of female *
(-1)*WeightFemale + Number of male * WeightMale
 Sex-value of a Session: V(Si) =Total number of
female of all products in the session * (1)*WeightFemale + Total number of male of all
products in the session * WeightMale
How is weight?
. To simplify, The WeightMale value is set to 1.
The WeightMale will be estimates base on below
conditions:
V(Si) < 0 when the i’th session is female
V(Si) >= 0 Si when the i’th session is male
0 < WeightMale < 1
How is weight?
The Accuracy on Training Set depends on FemaleWeight
Re-using for improvement
Because of the lack of evidences to
predict
Choose the high confident results to
re-train the model
Re-using for improvement
The High confident results are stable in
ratio of number male and female on all
catalogues
Thank you for
your attention