Semantic Web & Web 2.0

Emerging Science of the Internet: Some New Opportunities (Ron Brachman, Yahoo)

Ismail Fahmi — Tue, 05 Jun 2007 15:16:06 +0000

This article is a free writing from the last keynote speaker at the ESWC 2007, Ron Brachman, from Yahoo. I write any ideas during the presentation, which reader might find not complete or structured. Anyway.. enjoy this note.

Knowledge Representation (KR) is a very important ingredient for the semantic web movement. And in the Semweb, KR meets the Web. Semweb can be said as “addding logic to the Web.”

Ron’s thought is that Semantic Web is now transforming the industry. He explained why Yahoo needs the SW, why KR is great, and how we can get profit wildly. However, major web company seem to take different direction.

We should tanks to Wayne Gretzky’s father speak: “don’t skate to where the puck is, but rather to where it will be.” So, not about where we are now or where we have been, but where the puck will be. The discussion is then will be lead to the ideas for future.

The idea behind Web 2.0 are making a huge impact in the industry (see Mark Greaves’ article Marc/April IEEE Intelligent System).

In Yahoo, they focuse on the underlying knowledge needed to move ahead (the science behind Web n.0). Ron thinks there are opportunities here.

The Internet

What is the Internet? According to Wikipedia, we get technical and very dry definition of the Internet. But this is not what most citizen would like to tell you. According to techies Internet is a distributed, heterogenous repository of tools, data, and people — all kind of people. It is a user centered view, where user can have many functionalities to find, use, share, etc “stuff” in the Internet. Stuff here can be any data, knowledge, help, things, people, etc.

Some example of user needs are search/find, communicaton through email, socializing/sharing, buying, selling, matchmaking, experiencing media, etc. A key question for compay like Yahoo is: how to satisfy the needs? NOw, internet is already far far away from ftp, telnet, gopher, etc. Now, we have instance messeging, search engine, e-commerce, social media, networked games, ads, podcasting, blogging, wikis, voip, etc.

There are some important concepts to watch for. Fragmentation: everyting are composed from various fragments. Relationships: linkages of all sorts are critical things.
User-generated: sell, help, to be help, contact, blog, are strong drivers compared to the need of companies. Social media/”wisdom of crows”. Rich media: putting tv, radio, video, etc on the Internet. CPC/CPM/CTR: is very very important. Cost per click, click through rate.. now become very very important for a large number of companies or bussinesses which rely on the Internet.

The biggest problem is still: how to find stuffs. It is finding and answering problem. What we really care is not searching, but finding stuff and people that matched. Very significant problem: how do you know that you find stuff that you really need? How to know that two things are really different or the same? This is a problem of Object identity. Problems of integration, mash up, retrofit, synthesis. How do we deal with putting all things together? Lightweight: most people don’t care wheter it is semantic-based or whatever, but care about how they get information, and they can use it. The company and system should be lightweight in order to people can use it easily.

World has changed, from TCP/header,DNS, TELNET, dll to one where the worthwords are things like “mushup”. Thus we need to move from the “Old” computer science. Institution like ACM tagsonomize what they do into hardware, software, data, computer, etc. The study and curriculum also still address their learning following the taxonomy.

We need a new science of the Internet. From that “old” CS to the new one, so the graduated students are ready to joint company like Yahoo to face the problem in the Internet. Some areas:

* “Finding” science and systems
* Community science and system, which drive the Web 2.0
* Algorithmic advertising
* Computational micro-economics. Collaboration between CS and economics. The impact is so greate
* Media experiences
* Data science. Wall Mart dealing with 70 Terrabytes of data per day. How to deal with this?

We need to bring together existing disciplines and nurture the creating of new ones.

Ron bring a high level view of some new areas of thinking, key issues for consideration, and research challenges.

Toward a Science of Finding (moving beyond “search”)

It is more than just classical IR such as structure, index, and context. We should move to an information supply model, from query-based retrieval to context-driven supply. There are huge number of people which like to help you to find what you want: social search. Allow poeople, not search engine.

There is an evolution of commercial web search engine. In the first generation, we use only “on page”, text data such as word frequency and language. In the second generation, we use off-page, web-specific data such as link analysis, click-through data, and anchor-text (how people refer to tis page). The third generation answers “the need behind the queries”. There are semantic analysis (try to understand the meaning of the web), integrate multiple source of data, understanding tthe context (spacial, geography, religion, etc), help the user using spell checker, query refinement, etc; and integration of search and text analysis.

There are potential interesting opportunitis for SW technology for the first two items in the third generation. Also difficult challenges.

User needs results not documents. 60% of the search need are informational, what to learn e.g. definition of hemoglobin. 25% want to go to a page or navigational. Such as we already know the australian air line, but we use google to naviage to to that page. Other need is transactional, want to do something, e.g. buy a device (mediated by web), about 35%. Of course there are gray areas such as need of finding hub, and knowing what’s happen there.

The third generation search focus on user need and answer it directly using text analysis, semantic analysis. Either use hard or soft (partial) mathces, etc. For example, search for britney spears. This could be name of person, headline, company worked with her, etc. Yahoo knows this. So it gives ordinary search results also interesting shortcuts such as buy CDs, get pothos, etc. Searching hotel, also provide the availability of hotel rates. Searching person names and cities, usually need phone numbers and detail address.

Challenges of the third generation of search engine: answer user needs. For example, a search “sea food san francisco”. We can find the category in our data is restaurant, and then give shortcut/suggestion related to this category. We can apply machine learnig to classify and extract structures from users’ context, content, etc. Problem: the data are not always pre-classified.

From IR to Information Supply model

Information driven by user query, and the systrem provide information they need. Another example, after we book a flight, we are suggested to rent a car. In microsoft word, recommend “you seem to be writring a bussiness letter..” etc.

The following notes are only key ideas. Ron’s gave the presentation for last pages so quick because no more time for him.

Technical challenges: representation of context, information, and users. How we match in the right context.

The power of social media: flickr. People tags thousand of photos. The problem: tags are uncontrollable and organic. “the wisdom of crowds”.

We need community science and system works. myspace, flickr, etc. HOw we support people there?

There are four types of communities: social network, knowledge social search, enthusiast/affinity/music, marketplaces.

Nice example: DBLife Portal, extract information created by members of groups.

Book: the wisdom of crowds

Advertising algorithms. Problems in this area:
*match ads to query/context/user
*order the ads
*pricing on a click-through
*ambiguity

What this all about?
*For large open web companies, they move things strongly toward “web 2.0″ applications. Relation and social element are very important.
*Huge and growing among of content is user produced.
*dont forget ads: huge matching and inference problems
* new science are evolving

Toward a science of aggregated organic knowledge
* extraction to integration/rationalization
* wrapper induction
..

Lets skate to where th puck is going to be
Yahoo mission: connect peopleto their passions, their communities, and the world’s knowledge.

SWHi Demo and Poster at the ESWC 2007

Ismail Fahmi — Tue, 05 Jun 2007 15:11:12 +0000

In the ESWC 2007, I presented a poster and a demo about my work at the Library, in collaboration with Junte Zhang (previous master student) and two supervisors, Gosse Bouma and Henk Ellermann. Lately I was helped by Peter Scholing and Rene v.d. Ark to improve the program.

The paper was published in the LNCS on “The Semantic Web: Research and Application” and can be downloaded here: http://www.eswc2007.org/pdf/eswc07-fahmi.pdf . The poster it self can be downloaded here: …odur..(UPLOAD)

From 7 to 9.30 p.m. there are around 10 system descriptions (poster + demo), 17 demos, and 37 posters presented on the Kristal Foyer floor of the Innsbruck Congress. My paper is entitled “SWHi: A Case Study in Information Retrieval,k Inference, and Visualization in the Semantic Web.:

I got more than 15 visitors in this session which some of them are interested in the implementation of semantic web for culture and history domain, want to know how the ontology populated, how I use GATE to get name entities and relations (which I haven’t really integrate with this system), and give comments and suggestions for the improvement of the system.

The following are some important notes from some notable visitors:

Andy Seaborne [HP Labs, http://www.hpl.hp.com/personal/afs/] asked how if a user want to add their comments, or correct relatinships generated by the system. These features are not yet implemented in the SWHi’s berrypicking or bookmarking model. His questions actually underlined what I have gotten the day before during a workshop that community-based feature is a must in a future semantic digital library. Like the Plos One application which use Fedore as its main engine, also provides annotation feature to users. To each scientific paper, a user can add his comments and share it to other readers.

Danny Ayers [http://dannyayers.com/] came to my booth to meet Andy when I just started to answer Andy’s questions. He quickly recognize the Network Graph on my poster. “I know it, TouchGraph!” Then, he gave his opinion about the tool, that according to him TouchGraph’s code is somehow messy. However, the creator gets a lot of benefits from selling it to companies.

Armando Stellato [http://ai-nlp.info.uniroma2.it/stellato/] with his Semantic Turkey poster was sitting near my table. His application, Semantic Turkey, a plugin of Firefox for bookmarking important information based on an ontology system, is more or less has similar feature with my SWHi: on bookmarking and using TouchGraph for visualizing the network. Both of us use LinkBrowser version of the TouchGraph. His visualization is somehow basic and can be improved like the one I did by adding icons, modifying colors, font, hints, and borders. He want to know if someday the SWHi application will be ready for download.

Osma Suominen from Finland has a nice faceted search implementation for his Citizen’s Health Portal. Since the first day of the conference, I have met him and share our works as well as some personal information like families. We have interests on each other works. I gave him the address of the SWHi site (developmental version) and from him I got ideas for exploring faceted search features more than before, and to use Solr/Lucene for searching the ontology to get better and faster performance. At this moment, generating a ClusterMap data may take 1 or more minutes and still only 100 search results that are clustered. These are caused by the use of direct queries to the Sesame. In the future, I will index the ontology using Solr and send queries to it, especially for incomplex queries, for example to get properties of an instance.

Wrap up:

[1] The idea of berrypicking is fancy. Many visitor impressed with it. However, more work to be done in the future:
* provide more features on personalization and should be more community-based
* beside provide a grapical visualization, there should be an option to provide suggestion in a textual-base presentation

[2] Provide indexes of contained subjects in the front page (as suggested by Peter) automatically either from ontology or from faceted search.

[3] Provide textual representation for the ClusterMap, which can be combined with the faceted search feature.

[4] Improve the performance by exploiting Solr/Lucene.

Good bye Old Digital Library: Community-based Collaboration is a MUST (From the ESWC 2007)

Ismail Fahmi — Tue, 05 Jun 2007 15:06:18 +0000

Hi all. I’m now in Innsbruck attending the ESWC 2007 conference. The first day, I attend a tutorial about the Semantic Digital Library. This is a short note from this tutorial.

There are three systems presented and demonstrated in a full day tutorial about the Semantic Digital Library (DL), on Sunday, 3rd of June 2007. The three systems are JeromeDL (DERI Galway), Fedora (Cornell University), and BRICK (Univ of Vienna). I followed this tutorial because I want to know the current development and the future vision of the digital library, w.r.t to the semantic web.

The explosion of the involvement of communities into the Internet, for example indicated by the huge number of blogs and blogospheres, have shifted the old paradigm of the digital library. The old digital library are not enough any more, because it is a one way communication, from librarian to users. Library and librarian have information, and users just use them. The future of digital library should allow community of users to give contribution to the system.

There is a new coined term, at least new for me, named ’social semantic digital library’. In this model, a user can give comments to any resources presented in the system, create bookmarks and shared their contributions with other users. Moreover, a semantic digital library tool like Fedore allow user to create a blog on top of a digital library application.

We are not really ’say good bye’ to the old digital library. We have to use their power such as providing search, browse, harvesting, etc. But, we also have to provide community-based services in order to they can join the ‘new generation of the web’. This generation based on providing answer to user needs, profiding findings, not search results. More about this can be read here (from the presentation of Ron on the new science of the internet).

Text based search on the semantic web, a survey of existing systems

Ismail Fahmi — Wed, 18 Apr 2007 21:06:16 +0000

Michiel Hildebrand compiled a list of systems that provide access to semantic web data through a graphical user interface. He was interested in systems that provide some form of free text search, in addition to those that include different forms of browsing. For each system he stated the intended purpose, intended users, the scope, the triple store and optionally the technique or software that is used for literal indexing.

Overview of existing systems that support semantic search

Secondly, he analyzed the systems in three different stages of the search process: input, processing and search results, as well as the role of user feedback within these. For each of these he considered the functionality that the system provides and how this is made available through the graphical user interface.

Analysis of systems

Interested? Visit his survey page here.

For me, his survey is very interesting, especially when I face difficulty in implementing free text search on the semantic web, if it should be fully depended on the SPARQL or native semantic query languages. For my SWHi project implementation, I use Lucene/Solr to index the ontology and provide free-test and faceted search as well. Apparently, many of the surveyed systems also use Lucene to index their literal (textual) values, such as: HybridSearch, KIM, Longwell, SemSearch, Squiggle, and SWSE.

Solr and Lucene: Flexible, Easy, and Powerful

Ismail Fahmi — Wed, 18 Apr 2007 06:16:56 +0000

I found that Solr and Lucene are powerful tools for indexing and searching. I used them for the Semantic Web for History project. This week I played again with them for making a toy site: Search Engine for Song Lyrics. What I like from them are:

Easy to install, configure, and run
Easy to create an index, delete or update the index
Flexible and powerful search features (faceted search and flexible sorting are among others)
Fast and scalable.

I don’t think I need to give more description here on how to create index and search using Solr. The Solr tutorial is short and enough.

Past Works

Ismail Fahmi — Mon, 02 Apr 2007 13:39:55 +0000

Hi all, It has been two months I didn’t post any article on this blog. A lot of works has been being done, especially in developing the Semantic web application for history (SWHI).

I made a converter from MARC21 metadata into an ontology based on the PROTON ontology.
The ontology is now rich with relationships.
I created patches to the ClusterMap applet in order to when users click a cluster, its cluster members (documents, persons, etc) will be listed in another HTML iframe.
Last night I just modify my Lucene indexer for the ontology to be more configurable and efficient.
I also would like to experiment with scoring and ranking ontology search using a technique similar with the PageRank.

I’ll make posts about these topics later.

Profiles in Terror (Semantic Web Apps)

Ismail Fahmi — Wed, 10 Jan 2007 12:34:32 +0000

I got this site when searching for web sites fully comply with the semantic web technologies. It is profilesinterror.minswap.org, the Semantic Web Terrorism Knowledge Base, a project of the Semantic Web Research Group of the MIND LAB at the University of Maryland Institute for Advanced Computer Studies. This website was built in order to explore how the Semantic Web can be used to analyze terrorist activity. Research on the site is also supported by the Joint Institute for Knowledge Discovery.
You can browse its ontology (structure) and rules used to infer information such as: affiliation of terrorrists, facilities, residence, nationality, parent, uncle, cousin, and sibling.

ProgrammableWeb: A Mashups and APIs Reference

Ismail Fahmi — Wed, 27 Dec 2006 12:08:31 +0000

When I looked for websites labelled as Web 2.0 compliants, I found the ProgrammableWeb as an interesting one. Flagged as a Top 500 Blog of Feedster, this site provides a list of new APIs and Mashups updated everyday.

If you want to build an application using Web as the platform, you can take a look to this site, reviewing similar applications to yours, and read what people said about them. At the time of this post written, ProgrammableWeb has recorded 1392 Mashups and 352 APIs.

Other thing that we can learn from this site is that if we want to have a highly rated blog, make a reference blog about anything we want to focus on. The ProgrammableWeb “somehow became one of Google’s highest rated links on the topic.”

Web 2.0, shall we learn from it?

Ismail Fahmi — Wed, 27 Dec 2006 00:48:23 +0000

Sir Tim Berners-Lee, in a podcast interview for IBM almost one year ago (9/1/06), indicated that the term “Web 2.0″ is a piece of jargon that we might not need it. Further, he said, “nobody even knows what it means.” He might be right. The meaning of this term, since the first introduction by Tim O’Reilly in 2004, is keep evolving until the end of 2005 (and now?).

What I can understand from the Web-2.0-term debates is that there are some visionary people trying to give a name to the evolving new technologies which bring new excitements, benefits, and experiences to users, while on the other hand some visionary people see the evolution of user experiences in the internet is just consequences of the maturity of underlying technologies they developed. Indeed, Web 2.0 applications use the same technology (TCP/IP, HTTP, SMTP, IMAP, LDAP) developed since the beginning of the internet. The incredible AJAX which enriches the Web also uses HTTP.

If I look at again to the history of the Web 2.0 term invention, I would be easily suspicious to the intention of their inventors. This term was first arose during a brainstorming session between O’Reilly and MediaLive International, a producer of technology tradeshows and conferences. If I would be there, of course I would think of some thing big, bigger, great publicity, or if needed a buzzword, a hype which will successfully persuade people for attending and participating the shows and converences. But again, this is only my naughty and subjective feeling. Don’t take it into account.

However, for people like me, which has no benefit of being supportive nor critical, being open mind would be better for not “missing the opportunity to learn what the new visionaries out there think, those that do believe they are onto something different, building something different, onto a different era (Stowe).”

I learn one wisdom while reading posts and comments about the Web 2.0:

Organizations that fail to embrace the Web’s natural communication-oriented strengths will fail when put in competition that those that do.

People who stand hand-in-hand promoting the Web 2.0 are now trying their best to learn the evolusion or revolusion of the Net, and formulate strategies and wisdoms. Regardless we are a hype or an anti-hype toward the term, they are exist and keep sharing their though through journals and workgroups.

Questioned about the meaning of the term, one of Web 2.0 supporter, Dion, obliged to review or gather the best or most interesting definition of this term and review its status. And finally, the official definition of Web 2.0 is:

Web 2.0 is the network as platform, spanning all connected devices; Web 2.0 applications are those that make the most of the intrinsic advantages of that platform: delivering software as a continually-updated service that gets better the more people use it, consuming and remixing data from multiple sources, including individual users, while providing their own data and services in a form that allows remixing by others, creating network effects through an “architecture of participation,” and going beyond the page metaphor of Web 1.0 to deliver rich user experiences. (Tim O’Reilly)

Despite of all criticisms, people keep using the term as “an indicator that something different is going on with recent innovations on the web (Baloney)”. They believe that the term “is being adopted by a wide range of people, including marketing weasels and earnest technologists, each of whom have their own reasons for adopting the term.” Answering weak business model criticisms, Dion suggest some tactics to successfully monetize Web 2.0, and supporting new adopters, John Musser, a member of Web2.0 workgroup, collects interesting APIs and Mushups available in the Internet.

Indeed, Web 2.0 has no standard like the Semantic Web. However, we can learn many things from these guys efforts. What do you think?

Visualizing Similarity Sets

Ismail Fahmi — Mon, 25 Dec 2006 16:13:50 +0000

One interesting topic of the natural language processing field is finding word similarity in a document collection. For example, given a set of news documents (more is better), we can see what names which are similar to the word “Fiat” (the name of a famous car). Can you guess, what are the first seven words which are closely related (similar in some ways) to “Fiat”?

If you click this (open new window), you will get a graph as shown by the following figure:

In the news document collection, Fiat has a close similarity with Volkswagen, Toyota, Open, Nissan, Renault, Honda, and Peugeot. Using intuition, we know that they are the names of popular automobile. I will not explain how this set of similarity information can be extracted from a document collection. You can read my colleague’s paper here. Instead, I will show how can we visualize sets of similar words in an interactive way.

If you click a node, for example Toyota, you will get new words in the graph related to the word Toyota. Click again and again to any nodes interesting for you, an interesting graph will emerge showing how names or words related to each other, as shown by the following figure (click to enlarge).

In the above graph, we can see a path showing how a car company Fiat related to a software giant Microsoft. Using a distributional similarity method, we can get similarity information from documents.

I use TouchGraph to visualize that information. It is an open source Java applet application which generate an animated graph of a set of linked nodes. The second graph above is generated using the following dataset:

 <TOUCHGRAPH_LB version=“1.20″>
<NODESET>

<NODE nodeID=“Fiat”>
<NODE_LOCATION visible=“true”/>
<NODE_LABEL label=“Fiat” shape=“2″ backColor=“ffffcc” textColor=“000000″ fontSize=“12″ />
<NODE_URL url=“http://odur.let.rug.nl/fahmi/sets/explore.php?id=Fiat” urlIsLocal=“false” urlIsXML=“true”/>
NODE>
<NODE nodeID=“Nissan”>
<NODE_LOCATION visible=“true”/>

<NODE_LABEL label=“Nissan” shape=“2″ backColor=“ffffcc” textColor=“000000″ fontSize=“12″ />
<NODE_URL url=“http://odur.let.rug.nl/fahmi/sets/explore.php?id=Nissan” urlIsLocal=“true” urlIsXML=“true”/>
NODE>
<NODE nodeID=“Honda”>
<NODE_LOCATION visible=“true”/>
<NODE_LABEL label=“Honda” shape=“2″ backColor=“ffffcc” textColor=“000000″ fontSize=“12″ />
<NODE_URL url=“http://odur.let.rug.nl/fahmi/sets/explore.php?id=Honda” urlIsLocal=“true” urlIsXML=“true”/>
NODE>…

NODESET>
<EDGESET>

<EDGE fromID=“Fiat” toID=“Nissan” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Honda” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Opel” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Renault” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Toyota” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Volkswagen” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
<EDGE fromID=“Fiat” toID=“Peugeot” type=“1″ length=“65″ visible=“true” color=“99cc99″ />
EDGESET>
<PARAMETERS>
<PARAM name=“offsetX” value=“0″/>
<PARAM name=“rotateSB” value=“0″/>
<PARAM name=“zoomSB” value=“0″/>
<PARAM name=“offsetY” value=“0″/>
PARAMETERS>
TOUCHGRAPH_LB>

TouchGraph can be modified or adapted for new data set, as long as the data contain information about nodes and links between the nodes. I also modify a bit the code to display a network of persons as shown in my other post.