Robustness and stability of enterprise intranet social networks: The impact of moderators

TL;DR

This study examines how removing different types of nodes, especially moderators, affects the structure and content of enterprise intranet social networks over time, providing insights for community management and network analysis.

Contribution

It offers a detailed analysis of the impact of removing moderators, spammers, and peripheral nodes on network robustness and content stability in enterprise intranet forums.

Findings

Removing spammers and peripheral nodes has low impact on network structure.

Removing moderators significantly affects network connectivity and shared content.

Moderators have distinctive behavioral characteristics that can identify them.

Abstract

In this study, we tested the robustness of three communication networks extracted from the online forums included in the intranet platforms of three large companies. For each company we analyzed the communication among employees both in terms of network structure and content (language used). Over a period of eight months, we analyzed more than 52,000 messages posted by approximately 12,000 employees. Specifically, we tested the network robustness and the stability of a set of structural and semantic metrics, while applying several different node removal strategies. We removed the forum moderators, the spammers, the overly connected nodes and the nodes lying at the network periphery, also testing different combinations of these selections. Results indicate that removing spammers and very peripheral nodes can be a relatively low impact strategy in this context; accordingly, it could be used to clean the noise generated by these types of social actor and to reduce the computation complexity of the analysis. On the other hand, the removal of moderators seems to have a significant impact on the network connectivity and the shared content. The most affected variables are closeness centrality and contribution index. We also found that the removal of overly connected nodes can significantly change the network structure. Lastly, we compared the behavior of moderators with the other users, finding distinctive characteristics by which moderators can be identified when their list is unknown. Our findings can help online community managers to understand the role of moderators within intranet forums and can be useful for social network analysts who are interested in evaluating the effects of graph simplification techniques.