Abstract
State-of-the-art in network science of teams offers effective recommendation methods to answer questions like who is the best replacement or what is the best team expansion strategy, but lacks intuitive ways to explain why the optimization algorithm gives a specific recommendation. To tackle this problem, we develop an interactive prototype system, EXTRA, as the first step towards addressing such a sense-making challenge. Through the lens of the underlying network where teams are embedded, EXTRA explains team recommendation results. The main advantages are: (1) Algorithm efficacy: we propose an effective and fast algorithm to explain random walk graph kernels, the central technique for networked team recommendation; (2) Intuitive visual explanation: we present intuitive visual analysis of the recommendation results, helping users better understand the rationality of the underlying algorithm.
Introduction
In organizational management and collaborative research, building a "dream team" has traditionally been treated as an art. However, network science has increasingly turned it into a data problem, using complex algorithms to suggest who to hire, who to let go, or how to expand. The primary hurdle for adoption is that these algorithms are often "black boxes"—they might identify a candidate as the best fit but rarely explain the underlying reasoning.
Presented at RecSys ’18, EXTRA (Explaining Team Recommendation in Networks) is designed to move team recommendation from "what" to "why." By leveraging the mathematics of graph kernels, EXTRA provides intuitive, visual explanations for team-based decisions.
The Sense-Making Challenge
Modern team recommendation systems rely heavily on Random Walk Graph Kernels (RWGK). These are powerful mathematical tools that measure the similarity between two networks by analyzing how "walks" (paths) through the graphs overlap. While effective at finding a candidate who structurally and skillfully mirrors a departing team member, RWGKs are mathematically dense and opaque to end-users.
Without a "sense-making" step, human leaders are unlikely to trust algorithmic suggestions. A team leader needs to know if a recommendation is based on a newcomer's specific technical skills, their previous collaborations with existing members, or their central position in a professional sub-community.
The EXTRA Framework
EXTRA decomposes the recommendation process into three explainable perspectives: Edges (relationships), Nodes (team members), and Attributes (skills). The system adapts these explanations based on the specific team optimization scenario: replacement, expansion, or shrinkage.
The technical core of the system is the use of Influence Scores. Instead of just calculating a final similarity score, EXTRA calculates the derivative of the graph kernel function with respect to specific network elements. This sensitivity analysis identifies which components most heavily impact the recommendation:
- Edge Influence: Measures how much the overall team similarity would change if a specific collaboration were added or removed.
- Node Influence: Aggregates the influence of all edges connected to a specific person to determine their importance to the team's structural integrity.
- Attribute Influence: Measures the impact of specific skills (e.g., "Data Mining" or "Machine Learning") on the recommendation.
Interactive Visual Analysis
The authors implemented these influence algorithms into an interactive UI that allows users to query team members and perform influence analysis in real-time.
In a case study using a co-authorship network, the system was tasked with finding a replacement for a prominent researcher, Dr. Yu. The system identified Dr. Jiawei Han as the optimal candidate. EXTRA justified this by visually demonstrating that Han’s most influential edges overlapped almost perfectly with Yu’s, proving he could step into the same collaborative role. Furthermore, the system identified "databases" and "data mining" as the critical shared attributes justifying the replacement, presented via intuitive pie charts on the nodes.
Conclusion and Future Work
EXTRA provides a significant leap in transparency for networked recommendations. By defining "explanation" as mathematical sensitivity, it bridges the gap between complex graph math and human decision-making. Future work includes improving the scalability of random walk graph kernels for massive social networks and conducting formal user studies to evaluate how these visual explanations impact the speed and quality of human-led team management decisions.
Summary updated