You are on page 1of 4

PERSONALIZED PHARMACEUTICAL

PRODUCT RECOMMENDATION SYSTEM


USING USER-BASED COLLABORATIVE
FILTERING WITH NEARBY HOSPITAL DATA
DAEWOONG FOUNDATION
GLOBAL AI & BIG DATA SCHOLARSHIP 2021
SYSTEM ARCHITECTURE

Customer Nearby Customer Nearby


Order Data Hospital Data Order Data Hospital Data

User Similarity User Similarity User Similarity User Similarity


Calculation Calculation Calculation Calculation

Rating Rating Combine the similarity score with


Prediction Prediction weighted schema (?)

Combine the predicted rating with Rating


weighted schema (?) Prediction

Top-N recommendation Top-N recommendation

The proposed system architecture (B)


The proposed system architecture (A)
2
CONCEPT DESIGN

1 Unique ID Medical Department Data


3 User- Nearby Department Utility Matrix
(n x k)
H_Nam D_Nam dk
H_key … Long Lat d1 d2 d3 …
e e
u1 1 1 2 2
HK1 h1 d1 Matrix
Transformation u2 2 1 1
Feedback:
HK2 h1 d2 Frequency
u3 2 3 1 3 (within a radius)
HK3 h2 d3
…. … IF radius changed, THEN
…. … the values will be changed
un 1 1 1 2 too
hp dk

(Nearest) Neighborhood Extraction ?


Normalized values?
2 Middle
Representation
Pharmacy Data / Customer Data
4 i1 i2 i3 … im
U_key P_Name … Long Lat Neighbors

u1 { HK1,HK2, …, HKt } u1 1 1 2 2
u2 2 1 1
u2 Feedback:
Transformation u3 2 3 1 3 Frequency or
u3
…. … Rating
….
un 1 1 1 2
un

User-Item Utility Matrix


Purchase Order
Each user has a set of (n x m)
History neighbors within a radius
{ HK: H_name, D_name, longitude, latitude, … }
This representation
Radius is fixed / determined by Experts ! User-based CF (kNN)
reflects on Solution A
where each pharmacy
have different # of 5 Similarity Calculation, Rating Prediction, Top-N 3
neighbors
Recommendation
THANK YOU

You might also like