YOU MUST EAT THIS GRAPH NEWS, GRAPH OMAKASE. 3weeks July

Link prediction for ex ante influence maximization on temporal networks

[https://arxiv.org/pdf/2305.09965.pdf]

Introduction

I’m sure you’ve all used SNS at least once. By any chance, when should I post a post so that it exposed like views?”Didn’t you think about it?” When I was young, I was able to infer the “time zone that students often see” by repeating activities such as writing diaries and leaving Ilchonpyeong on Cyworld, and I was happy to receive a lot of exposure and attention accordingly. This graph, Omakase, is also sent out at this time of the week, judging that it’s the right time for you to be interested in Sunday evening, night.

The experience I wrote at the beginning was entirely dependent on ‘gam’. There is a field that quantifies this ‘sense’ and studies when my information will be delivered well, that is, relatively amplified and delivered to others. It’s called “influence maximization (IM). I think the most familiar areas for us are “Viral Marketing of Social Networks” and “Guessing Corona Infected Persons.”

This paper reflects the aforementioned IM techniques and uses them for link prediction. The main point of this paper is to verify the hypothesis that link prediction performance will be improved if a node is set to seed node, the influence of seed node is transmitted over time, and the received temporal network view is reflected.

Preliminary

In the context of network science, the concept of “information maximization” can be related to the dissemination and utilization of information in a network. It pertains to maximizing the efficiency and effectiveness of information flow within a network. Here’s an overview of information maximization in the context of network science:

1. Diffusion and Spreading Models: Information maximization in network science involves studying how information, opinions, or behaviors spread through a network. Various diffusion and spreading models are used to understand and simulate this process, such as the epidemic models, influence maximization models, or rumor spreading models.
2. Information Cascades: Information cascades refer to the propagation of information through a network where individuals adopt a particular behavior or belief based on the actions or decisions of their neighbors. Maximizing information in this context involves studying how to trigger large-scale cascades or to influence the behavior of individuals in a network to maximize the spread of information.
3. Influence Maximization: Influence maximization is the task of identifying a small subset of influential nodes in a network that can maximize the spread of information or influence. The goal is to strategically select nodes to initiate information cascades and ensure the broadest dissemination of information within the network.
4. Network Resilience: Information maximization also considers the resilience of a network to disruptions or attacks that can hinder the spread of information. Maximizing information flow involves studying strategies to enhance network robustness, identify critical nodes for information dissemination, or design efficient communication protocols.
5. Optimal Routing: Information maximization can also be viewed in the context of optimizing the routing of information in a network. This involves finding the most efficient paths or routing strategies to maximize the throughput, minimize delays, or ensure reliable information transmission.

Overall, information maximization in network science focuses on understanding and optimizing the flow of information, influence, or behaviors within a network. It involves studying various models, techniques, and strategies to enhance the spread of information, identify influential nodes, improve network resilience, and optimize information routing.

Summary

We verify the hypothesis by comparing the model performance before and after applying IM technique to seven Real-world datasets. With data from each static / temporary network, we deduce the time-varying network topology and apply IM techniques to the derived new network topology. The key here is the various combinations of perspectives applied by IM techniques and link prediction techniques, respectively.

Insight

I think it would be very helpful to read the paper with questions about the ‘why’ of solving the chronic problem of Static & Temporary IM and how it was applied to the link prediction task.

Building Your Own Schema.org

[https://medium.com/@Tonyseale/building-your-own-schema-org-7600a90e690a]

It’s rare in Korea. This is written by Tony Seale, who works as a knowledge graph engineer. Basically, why do we use knowledge graphs? I think we use it to infer the connection between knowledge that we don’t know, to derive and express potential and intuitive meanings. I came across knowledge graph for the first time while reading KGAT paper. Here too, we propose KGAT to verify the hypothesis that a model with an additional application of Semantic Information from Knowledge Graph will show good results in inference performance. To put it simply, for example, we have a different interpretation of each chef, samurai, and craftsman for the object of ‘knife’. Because each subject approaches the ‘knife’ from a different perspective, the perspective is considered semantic and applied to the recommended system field. I’m out of the conversation for a while.

Back then, when I first learned about knowledge graphs through the recommendation system, and now the view of knowledge graphs is similar in the big context of discovering knowledge between data, but from where? The little context of how? is very different. Fragmented, how do you define the schema before SPARQL queries now, contrary to the previous view that you can just ‘query well’ with SPARQL? The view was added. It’s an ontology design perspective.

In this post, we have written a good view of the design in a short article. How to design ontology? Those who had questions like this must have had a hard time because of the difficult data on the Internet. I am confident that this post and the medium-related posts I made will be of great help. I strongly recommend this post because it contains a lot of useful materials.

It’s a good book to listen to. The book is expensive, but I think it’s very cheap compared to the knowledge inside. I’ve done Knowledge Graph & Ontology! It’s a must-read book that you need to study in advance to talk to people who do. Recommend.