Files and links (1)
CC BY V4.0, Open Access
Dissertation
ITeM: Independent Temporal Motifs For Representing and Analyzing Temporal Networks
Washington State University
Doctor of Philosophy (PhD), Washington State University
01/2022
DOI:
https://doi.org/10.7273/000005493
Handle:
https://hdl.handle.net/2376/119305
Abstract
A graph is a flexible and natural data modeling approach used to represent entities and relationships in many domains. Since the rate of data generation has increased many-fold in recent years, it is imperative to efficiently extract knowledge from underlying real-world systems to develop scalable artificial intelligence-based solutions. Temporal graphs are specialized graphs that can be used to model an evolving real-world system. Many real-world domains such as social networks, communications, travels etc., can be modeled as a temporal graph. When extracting knowledge from temporal graphs, observations about a small set of entities (or relationships) within a shorter time duration are early indicators of larger patterns in the graph. These local properties can be measured using small subgraphs called motifs. We propose Independent Temporal Motif (ITeM) to extract temporal-spatial knowledge from real-world domains. We show that ITeMs are efficient to compute and scalable indicators of the temporal evolution of the graph. We present the motivations behind using ITeM, key definitions, algorithms, and software tools to support our hypothesis. We use ITeM to analyze multiple synthetic and real-world graphs for graph comparison, generation, sampling, and trend analysis tasks. We benchmark ITeM with state-of-the-art motif-based approaches and show that ITeM performs better at measuring the changes in the temporal graph than the other approaches. We also use ITeM as node and graph embeddings to show that it outperforms state-of-the-art deep learning embedding techniques to perform standard graph analytic tasks. Additionally, we show the impact of ITeM in different real-world applications such as analyzing the change in graphs representing Covid-19 and scientific publications.
Metrics
11 File views/ downloads
32 Record Views
Details
- Title
- ITeM: Independent Temporal Motifs For Representing and Analyzing Temporal Networks
- Creators
- Sumit Purohit
- Contributors
- Lawrence B. Holder (Advisor)Ananth Kalyanaraman (Committee Member)Assefaw Gebremedhin (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- School of Electrical Engineering and Computer Science
- Theses and Dissertations
- Doctor of Philosophy (PhD), Washington State University
- Publisher
- Washington State University
- Number of pages
- 166
- Identifiers
- 99900592359301842
- Language
- English
- Resource Type
- Dissertation