A New Approach to Detect Communities in Multi-Weighted Co-authorship Networks

Abstract

Co-authorship graphs related to publications (bibliographical production) are generally represented by undirected graphs, where each edge maintains a single value for publications produced by the authors. A variation of this type of graph is related to use of different weights associated with different types of publications. In this context, within the field of scientometrics, 'journal papers' or 'books' may have a higher priority than the 'conference papers', and the 'conference papers' might have higher priority than the 'extended abstract'. In this paper we present a simple weight combination to detect communities in co-authorship networks, which allows simultaneous consideration of multiple types of collaborations. Our preliminary results show a good performance in the detection of communities considering real bibliographical production graphs.

Preliminary Results:

For the tests the following initial conditions have been used:

There are three priority levels and each of them has a specific weighting

Level 1 (weighting = 3):

• Articles in scientific journals
• Book published/organized
• Book chapter published

Level 2 (weighting =2):

• Complete works published in proceedings of conferences

Level 3 (weighting =1):

• Articles in newspapers/magazines
• Expanded summary published in proceedings of conferences
• Summary published in proceedings of conferences

 

Example 1:

For this example, a collaboration graph generated by scriptLattes has been used, they have taken their data from Department of Computer Science - IME - USP and here you can find more information about this graph.

Figure 1: Input: IME - USP collaboration graph

Figure 2: Output graph using a single value for the edges.

Figure 3: The evolution of the modularity with the iterations using a single weight.

Figure 4: Output graph using multiple values for the edges.

Figure 5: The evolution of the modularity with the iterations using priority levels.

Description: In this example you can notice that some vertex has been moved from a community to other when a multi-weight graph is used, as you can see in the ouputs, the final real weight of each edges has changed.

Example 2:

For this example, a collaboration graph generated by scriptLattes has been used, they have taken their data from USP - São Carlos and here you can find more information about this graph.

Figure 6: Input: USP-SC collaboration graph.

Figure 7: Output graph using a single value for the edges.

Figure 8: The evolution of the modularity with the iterations using a single weight.

Figure 9: Output graph using multiple values for the edges.

Figure 10: The evolution of the modularity with the iterations using priority levels.

Description: In this example you can notice that the number of communities detected has changed when a multi-weight graph is used, as you can see in the ouputs, the final real weight of each edges has changed.