Seyed's Blog

10 Jun 2021

ESWC 2021 Summary

Here is a brief overview of experiences, keynotes, and presentations of ESWC 2021.

The conference was held online from June 6 to June 10. Presentation sessions were in Zoom. In addition, there was a Slack workspace and a Gather.town space for further communications. Apart from being a participant, I attended this conference as a volunteer. We were about 20 volunteers mostly young Ph.D. students and we were supposed to help the organizers and chairs for hosting the sessions and recording them, managing questions in the Zoom chat and the Slack channels, etc. This kind of volunteer works created a sense of responsibility and paying more attention to the conference and multiple coordination meetings with the chairs helped to build communications with people.

Day 1 – KGC Workshop

The first two days, Sunday and Monday, were for workshops and Ph.D. symposium. My paper had to be presented at the Knowledge Graph Construction workshop on the first day along with thirteen other papers.

My presentation was the first one and all went well. There were good questions about the meaning of topical subset, extraction of T-Boxes, and the reason for choosing two different dumps for extracting subsets.

After that, we had other presentations about construction with mapping and RML language. one of the papers was about creating an open drugs Knowledge Graph by gathering information from Wikidata and some other drug-related knowledge bases which used a form of subsetting.

We had also Jesús Barrasa keynote from Neo4j talking about property graphs and the Eccenca project And the fact that the capabilities of RDF and property graphs can be used together, and these two are not competitors and they are complementary.

KGC workshop had between 35-40 audiences, which I think was more than other workshops even though it was on Sunday. Another thing I felt was that in the workshops, the sessions were more interactive than the main conference, there was more participation in the discussion, more questions, considering we had less time than the main conference presentations.

Day 2 – Ph.D. Symposium

On the second day which was Monday, I was in the Ph.D. symposium. The keynote speaker was Paul Groth from the University of Amsterdam, and talked about Similarities and differences between continuing the research in academia and industry and presents some good points based on his own experience.

We had Luca Costabello, who spoke about ten important points that can create a mistake in the research path, again emphasizing the continuation of the path in the industry. I mean 10 wrong prejudgments. We had two presentations on the use of the knowledge graph in the management of home appliances and the Internet of Things. Another interesting idea was to use the knowledge graph to detect and break the chain of propaganda and political advertisements.

Day 3 – Natural Lang. Processing & Problems

Days 3 to 5 were the main conference. The main trend of the conference was on machine learning (Natural Language Processing, Question Answering, and Deep Learning) and also Data Quality issues.

On Day 3, the most important presentation for me was Thomas Tannon's on "Neural Knowledge Base Repairs". This paper won the Best Research paper award and it really deserved that. He described a deep learning approach that uses the edit history of the knowledge base to correct the shortcomings and constraint violations of the knowledge base. I was familiar with Thomas’s work on the edit history of Wikidata so I planned to chat with him after his presentation. I had a short chat with him about using Wikidata edit history in research (bot activity detection). It turned out that he had closed his Wikidata edit history endpoint due to the large volume (Wikidata edit history XML dump now has more than 1TB data compressed!!).

There was another session on the first day called "Problems to solve before you die". One of the interesting works on that was the Open Linked Code which is an approach to build a semantic knowledge base of programming code snippets to help the reusability of the million lines of codes. Personally, I very enjoyed the idea. I think having a knowledge base of codes with rich metadata is a key requirement of programmer robots. Age of Ultron is coming!

Day 4 – Posters/Demos & Machine Learning

Day 4 was the hardest day of volunteering for me so I cannot be in the sessions, but I managed to be in posters and demos session on Gather.town which I saw potentially useful approaches of using relation extraction and entity linking which is relevant to my work.

I also was on the keynote of the day which described the Amazon Product Knowledge Graph as an effort to use KGs in industry, showing what are the challenges in starting with a few amount of data and successfully transfer the KG into the billions of records scale.

Day 5 – Knowledge Graphs

On the last day, I was in the Knowledge Graph session. I can mention two interesting and practical presentations.

The first was about building a knowledge graph using data from operating systems logs and application logs like Linux journal logs, systemd logs, etc., and using this knowledge graph to analyze user activities and guide the user for better experiences. For example, they considered a case of security, their knowledge base was used to obtain a pattern of attacks and threats that could be used in intrusion detection systems.

The second interesting presentation was Philip’s talk from USC that described the construction of a Commonsense knowledge graph. They use an interesting tool called KGTK for subsetting Wikidata and other knowledge graphs. I talked to him after the presentation and he liked the idea of checking the quality of references in Wikidata. he said that the KGTK tool can do the most complex subsetting from the current dump of Wikidata in less than 10 minutes by creating a basic indexed dataset is not more than 4 hours, which will be very helpful compared to WDumper. They decided to add reference indexing to their toolkit so that we might be able to collaborate on new use cases on Wikidata.

Summary

Overall, the conference was very useful and I saw a lot of interesting questions. There were I think more than 70 presentations, I was in about thirty of them and I tried to choose sessions that were more related to my Ph.D. I found good friends. I got familiar with the challenges of holding such an important conference online. I am more motivated to continue now and I can see some good trends related to my Ph.D. to follow.


Please share your comments with me via email (sh200 [at] hw.ac.uk) or Twitter.