隐式反馈下的协同过滤
2016-01-30 15:33:59 5 举报
AI智能生成
论文笔记,Collaborative Filtering for Implicit Data
作者其他创作
大纲/内容
previous work
neighborhood models
common approach
user-oriented
early
item-oriented
better
amenable
no distinction
preferences
confidence
latent factor models
SVD
adequate regularized model
superior to neighborhood model
modification
model formulation
optimization technique
our model
borrow from LFM
model formulation modification
optimization technique
confidence levels
no positive action
other reasons
not liking
positive action
other reasons
liking
notion
r
confidence
raw observations
p
binary
derived by r
user preference
p = x * y
c
confidence of p
alpha
x
user factors
y
item factors
m
number of users
n
number of items
matrix factorization
similar to explicit
distinctions
varying confidence levels
account for all
cost functions
huge numbers
optimization techniques
ALS
main properties
transfer r to p and c
efficient algorithm
explaining recommendations
reduce
linear model
weighted similarity
pastactions
break
significance to user
similarity to target
different similarity for user
experimental study
data description
television
300,000 users(set-top boxes)
17000 programs
32 million data(rui)
filter test data
large range
log scaling scheme
momentum effect
down-weight subsequent
evaluation methodology
no negative feedback
precision based matrix
not appropriate
recall-oriented measures
ours
percentile-ranking
evaluation results
baseline
popular item
neighborhood based
ours best
lowest percentile-ranking
better in top
better than middle modle
importance of confidence levels
analyze performance
better for popular shows
heavy watchers not better
recommendation explaination
just similar shows
intro
content based
need external info
CF
better
cold start
explicit feedback
vast literature
implicit feedback
characteristics
no negative
inherently noisy
numerical value
confidence
evaluation
preliminaries
zero is ok
discussion
0 条评论
下一页