[Paper Review] Graph Pattern Matching for Dynamic Team Formation

Shuai Ma, Jia Li+3 more|arXiv (Cornell University)|Jan 3, 2018

Data Management and Algorithms36 references6 citations

TL;DR

This paper proposes a graph pattern matching approach for top-$k$ team formation that integrates structural constraints and capacity bounds using pattern graphs. It introduces a unified incremental algorithm to efficiently handle continuous, simultaneous updates to both the pattern and data graphs, significantly outperforming batch re-computation in dynamic environments with up to 36% pattern updates and 34% data updates.

ABSTRACT

Finding a list of k teams of experts, referred to as top-k team formation, with the required skills and high collaboration compatibility has been extensively studied. However, existing methods have not considered the specific collaboration relationships among different team members, i.e., structural constraints, which are typically needed in practice. In this study, we first propose a novel graph pattern matching approach for top-k team formation, which incorporates both structural constraints and capacity bounds. Second, we formulate and study the dynamic top-k team formation problem due to the growing need of a dynamic environment. Third, we develop an unified incremental approach, together with an optimization technique, to handle continuous pattern and data updates, separately and simultaneously, which has not been explored before. Finally, using real-life and synthetic data, we conduct an extensive experimental study to show the effectiveness and efficiency of our graph pattern matching approach for (dynamic) top-k team formation.

Motivation & Objective

• To address the gap in existing team formation methods that fail to model specific collaboration relationships among team members as structural constraints.
• To formulate the dynamic top-$k$ team formation problem for handling continuous, incremental updates to both the team pattern and the underlying data graph.
• To develop a unified incremental algorithm that efficiently processes pattern updates, data updates, and their simultaneous occurrence without full re-computation.
• To optimize performance through techniques like radius-varied balls and density-based filtering in the batch computation phase.
• To empirically validate the effectiveness and efficiency of the proposed approach on real and synthetic datasets under various dynamic update scenarios.

Proposed method

• Introduces 'team simulation' as a revised graph pattern matching semantics that incorporates structural constraints (via pattern graph topology) and capacity bounds (via node-level limits).
• Develops a batch algorithm for top-$k$ team formation using team simulation, with optimization via radius-varied balls to limit search space and density-based filtering to prune low-quality candidates.
• Proposes a unified incremental algorithm that maintains auxiliary data structures to support efficient updates to both the pattern graph and data graph.
• Employs a lazy update policy in the incremental framework to reduce overhead, especially under continuous update workloads.
• Supports three update modes: pattern-only, data-only, and simultaneous pattern-data updates, with incremental propagation of changes to maintain consistency.
• Uses a hybrid evaluation strategy combining real-world (Citation) and synthetic datasets to assess performance across diverse update frequencies and scales.

Experimental results

Research questions

• RQ1 How can structural constraints—specifically, required collaboration relationships among team members—be effectively modeled in top-$k$ team formation?
• RQ2 What is the impact of capacity bounds (minimum and maximum team member counts per skill) on the complexity and feasibility of team formation?
• RQ3 Can an incremental algorithm be designed to efficiently handle continuous, simultaneous updates to both the team pattern and the expert data graph?
• RQ4 How does the performance of the incremental approach compare to batch re-computation under varying update frequencies and data sizes?
• RQ5 What is the memory overhead of maintaining auxiliary structures for incremental updates in large-scale graphs?

Key findings

• The proposed graph pattern matching approach effectively captures practical team formation requirements, including structural collaboration constraints and skill capacity bounds.
• The batch algorithm achieves high efficiency, completing top-$k$ team formation in 116 seconds for a dataset with 1.39 million nodes and a pattern with 10 nodes.
• The incremental algorithm outperforms the batch approach when pattern updates exceed 27% and data updates exceed 24% on average, depending on the dataset.
• For simultaneous pattern and data updates, the incremental approach remains superior when update rates are below 22.5% (Citation) and 20% (Synthetic) per update cycle.
• The incremental algorithm incurs only 25MB to 130MB of extra storage on Citation and Synthetic datasets, representing 69.4% and 26.9% of the original dataset size, respectively.
• The lazy update policy in the incremental framework reduces processing time for subsequent updates, enabling efficient handling of continuous update sequences with up to 29% pattern and 26% data updates per batch.