Loading...

Link Prediction using Dynamic Graph Neural Network with Application to Call Data

Sajadi, Nafiseh Sadat | 2022

187 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55734 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Jafari Siavoshani, Mahdi
  7. Abstract:
  8. In network science, link prediction is one of the essential tasks that has been neglected. One important application of link prediction in telecommunication networks is analyzing the user's consumption pattern to provide better service. This project aims to predict future links with applications to call data using the users' call history. In previous research, there are two main approaches: 1) heuristic-based approach, and 2) deep-learning-based approach, such as graph neural networks. These methods are mainly used for processing static graphs, and therefore, we cannot generalize them to dynamic graphs. But there are many graphs which are dynamic in nature. For instance, call data records and biological data will change over time; therefore, their graph's topological structure will evolve. In order to process dynamic graphs, one of the proposed ideas is take snapshots from the graph and then predict the links using a combination of graph neural networks and recurrent neural networks. In call data records, each link is only valid for a certain period. As a result, graphs corresponding to snapshots are sparse, and the classification problem is imbalanced. A proposed framework for processing dynamic graphs is Temporal Graph Network. This framework is developed based on graph neural networks and will continuously process the graphs. In addition to that, the node embedding will be updated with graph evolutions, such as edge addition or deletion. In this method, the embeddings of target nodes are concatenated, and the output values indicate the probability of future link creation. According to previous research, using graph structural information may improve performance. Therefore, in the upcoming study, the Temporal Graph Network was combined with a static link prediction method, which was developed based on graph neural networks and extracting enclosing subgraphs. Finally, the results show that using this method will improve the average precision by 2.6%.

  9. Keywords:
  10. Dynamic Graphs ; Deep Learning ; Graph Neural Network ; Representation Learning ; Link Prediction

 Digital Object List

 Bookmark

...see more