Building Bibliographic Collaborating Networks in an Academic
Network Observation and Analysis
Circles and Preference
Network Observation and Analysis
Nodes sorted by reputation in descending order and placed on the circle
Degree: Low -------> High
Weight: Low -------> High
Color: Red - Yellow - Blue
Clustering and Community
They prefer to work with people with higher reputation
Academic Rankings
Degree: Low -------> High
Weight: Low -------> High
Color: Red - Yellow - Blue
Sparse structured
Many small communities
Giant Community
Important person in the giant community
First author for 2 papers, second for 1.
Henson Professor and Acting Department Chair at UNL, distinguished speaker for many comferences
Experimental Results
Experiment Design
Objective
Experimental Set
Compared with Newman's, the proposed network has:
1. Exactly same edges were kept after deconvolution
2. All edges weight heavier
Years: 2007 - 2010
Papers: 2135
Authors: 3669
Class A Conferences: 38
Network Observation and Analysis
Future Work
Experimental Process
Partitions and Institutions
Dynamic Networks
Extract paper information
Construct nodes and edges
Deconvolute networks
Compare recovery rate
Time Scale
Dynamic Changes
Scalability
Accuracy
Technical merits
Deconvolution Method
Normalization
Database
Matrix
Technical Merit: Network Visualization
Gephi Layouts
Gephi Functionality
Network Layout
Network Clustering
Shortest path Calculation
Partitions and Filters
Statistics
Procedures
Technical Merits: Network Construction
During 2006 - 2010: 7105 Papers, 11573 authors, 30385 connections
Relationship established on coauthorships
Scale up and deconvolute weighted matrices
Rankings
Data Source: Computer Science Academic Ranking Systems
Network: Weighted Coauthorship Networks
Cleanup: Network Deconvolution Method
Analysis: Network Visualization and Observation
Construct Nodes
Construct Edges
Quantify authors' contribution
Enhance relationship weighting
Experiment
Technical Merits: Network Deconvolution
Conclusions
Assumption on Relationship Matrix
Assumption: observed weight is the sum of weights at all levels.
The observed weight is equal to the sum of a geometrix series.
No. of times being cited: h*contribution
Weight of edge between A, B: contrA+contrB
Decomposition and Deconvolution
In this project, several goals were achieved.
Designed an approach to connect the scientists by their coauthoring publications
Improved the weighting criteria proposed by Newman (2001)
Infered the direct connections with network deconvolution method
Proposed a method for scientists and institutions ranking based on publications
Observations
Objectives
Method proposed by Feizi et al (2013)
Construct a collaborative network among the researchers and evaluate their relationships
Propose an approach to rank the institutions and researchers