ASWC 2009—- Some Interesting Experiences and Stories
(by Yi Zeng)
Today I just came back from ASWC 2009 in Fudan University, China. Generally speaking, it was a great experience to attend this conference. For the first day of the conference, we have 2 workshops and 3 tutorials. The 1st Asian Scalable Semantic Data Processing (AS2DP) workshop was organized by Zhisheng Huang from our LarKC (VUA) team, and Hyun Namgoong from Seoul National University, Seoul, Korea. More than 20 persons attended the workshop. To me, the experience is very special, since I’ve never had 2 talks in one workshop.
The first one is an invited talk titled “LarKC: The Large Knowledge Collider”. It was originally for Ning Zhong (Director of Web Intelligence Consortium and LarKC China team), but for some reason, he couldn’t make the trip. Hence, I prepared and present the talk as a representative of our LarKC project (with help from Frank and Spyros’ previous slides about LarKC). The slides can be found below. Many participants think that LarKC is with a great vision, and we have done a lot, but still will have many new challenges. Several detailed questions can be summarized as following:
(1) What are the unsuccessful stories for this project, or what kind of problems you met for your proposed methods? Because there are no projects which have never met any problems that cannot be solved very well.
My answer: (1) At the beginning stage of the LarKC project, we adopt the pipeline architecture. Later, we found that this kind of architecture is not flexible enough since plugins that are within the pipeline cannot invoke each other as one likes. Later, we decided to use workflow instead of pipeline. (2) When we try to create the context for Web reasoning from the end user perspective (mainly on acquiring users’ previous interests), we found that results for semantic similarity calculation cannot always satisfy our needs. For example, by calculation using google distance, we can make the decision that user interests on “search” and “retrieval” can be merged together since they are highly related from the semantic perspective. At the same time, the distance of “search” and “pagerank” is even closer than “search” and “retrieval”, but we cannot merge “search” and “pagerank”, since we cannot say these two are approximately the same.
(2) Stopping rule may diverse among different users, some times a specific user may prefer to wait for more results than other users.
My answer: I am not familiar with this, and our colleagues at MPG may give better answers. But it seems that as a first try, it is not a bad idea to find the rules that are applicable to most of the end users by statistical analysis.
The slides for this talk can be found Here.
My second presentation is titled “Social relation based Scalable Semantic Search Refinement”, the idea is that we use the social relations of a specific user to find a user’s interests since a group of friends of yours may have some common interests that are the same as yours. This serves as another strategy to select an important subset for reasoning (The first strategy is users’ own retained interests as introduced in the slides and the paper). One question from Harold Boley is that we did not consider the diverse relationship of the interests terms. Namely, even the users may have the same interests, the connections among interests can be different. This is true, and we are going to investigate on it in our future studies.
The slides for this talk can be found Here.
Papers from other authors can soon be found on the AS2DP workshop homepage (or may be you can ask Zhisheng for the papers). The other workshop “Asian Workshop on Social Web and Interoperability” focuses on the social perspective of the Semantic Web. The authors made very good efforts on the analysis of many existing social networks such as Japanese and Korean twitter and twitter-like social networks. Many efforts focus on the analysis of influencers in these networks. One of the studies focuses on the “friendship degree” in these networks. One of the invited lecturers for this workshop is by Paolo Bouquet from the OKKAM project.
The main conference began from the second day. The keynote was given by Nigel Shadbolt on “Government Linked Data: A Tipping Point for the Semantic Web”. It is interesting to see that the UK government release its data to the public and the integration of these data makes public service to the citizen seamless. To me, the most interesting thing is that integration and services produced based on these data are not provided to the citizen only online, but also offline in the form of printed version, since many of the citizen’s cannot access the Internet all the time. The second invited talk is from Aoying Zhou on “China Web Infrastructure”. He was invited as a friend from the database community, showing that how the “Web Infrastructure” was designed (mainly focuses on Web information from online Forums, Blogs, etc.) and how efficient queries can be made based on the infrastructure. Integration of the advantages on MapReduce and big table is considered, and a SQL-like query language is adopted. Some of the ideas can be found from one of their previous slides:http://deke.ruc.edu.cn/fdm2007/resource/FDM07.pdf.
One of the interesting stories is that on the poster session, there is a study on “Research Profilling Based on Semantic Mining”. It is done by National Science Library/Chinese Academy of Sciences and Indiana University. It crawls website information and identifies main content blocks, key concepts, events and their relationships. Their use case is using LarKC website! As you can see in the attached picture. One of the example is that the term “Large Knowledge Collider” is related to “massive distributed incomplete reasoning”, and “Semantic Web”. I am more interested in how they extract the relatively longer “massive distributed incomplete reasoning” as an object. Later, the author of the poster and I had a short discussion on LarKC since she was not aware of the recent development of our project.
It was a great experience to meet many people from other EU FP-7 projects, such as NeOn, OKKAM, DynaLearn, and SEALS. Most of the projects have some introduction materials for the participants, while LarKC does not. I blame myself about this. And we hope to do more collaboration with other EU projects. It reminds me of the Semantic Week 2009 at VUA this summer, where we learn from each other a lot!