Download presentation
Presentation is loading. Please wait.
Published byRoderick Warren Bond Modified over 8 years ago
1
ClusCite:Effective Citation Recommendation by Information Network-Based Clustering Date: 2014/10/16 Author: Xiang Ren, Jialu Liu,Xiao Yu, Urvashi Khandelwal, Quanquan Gu, Lidan Wang, Jiawei Han Source: KDD’14 Advisor: Jia-ling, Koh Speaker: Sheng-Chih, Chu
2
Introduction Model Overview Model Learning Experiment Conclusion Outline 2
3
Introduction 3 Based on content + KDD? +citation behavior? +Social network? +Dr.Koh? A small set of paper Google Scholar
4
Introduction 4 What’s heterogeneous network? What’s Meta-path?
5
Introduction Citation Recommendation 5 Heterogenous network Learning model
6
Introduction Model Overview Model Learning Experiment Conclusion Outline 6
7
Model overview 7 Define Score function s(q,p) : q = query manuscript, p = target paper : how likely query is to belong to the k group. : the relatedness between q, p according to k-th group. : the relative importance of p within the k-th group.
8
Feature Weight : weight change from importance : different meta path-based feature 8
9
Paper relevance Ex : k=1~4, 9 q p Interest Group 1 Interest Group 2 Interest Group 3 Interest Group 4 0.2 0.6 0.5 0.2 0.8
10
Relative Authority Affect Relative Authority Score: 1.Published in highly reputed venues. 2.Written by Authority authors. 3.Related to high quality paper. G() function is propagation function. 10
11
Paper-Speif c 11 q A V T Interest Group 1 Interest Group 2 Interest Group 3 Interest Group 4 0.8 0.3 0.4 0.2 0.7 0.6 query’s group menbership
12
Introduction Model Overview Model Learning Experiment Conclusion Outline 12
13
Loss function Actual value – prediction = loss 13
14
Graph regularization Normalization on authority vector 14 degree of P i in R A i-th column vector of F p j-th column vector of F A degree of A j in R A R K*n R K*|A| R n*n R |A|*|A| R |V|*|V| R K*|V|
15
Joint optimization problem Joint optimization problem : 15 Tikhonov rehularizes
16
ClusCite Algorithm 16
17
Introduction Model Overview Model Learning Experiment Conclusion Outline 17
18
Experiment Data Set :DBLP dataset2 and PubMed dataset3 18
19
Process work 19
20
20
21
21
22
22
23
Conclusion Propose novel cluster-based framework to satisfy a user’s diverse citation intents. Develop efficient algorithm and better performance this paper is good. 23
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.