Motivation

My exposure to Semantic Web started off by building a vertical portal ITTalks. Initially I considered this work as a part of my responsiblity as a Ressearch Assistant but as I got more involved, I really got interested in the Domain of the Semantic Web. For the last 1.5 years I have been working on different aspects of Semantic Web, be it Application Prototyping, ontology engineering or context-aware systems or semantic web services. As with lots of researchers in this area, I do think that Semantic Web has to become a reality some day but there are various steps that need to be taken before this becomes true. For example, we will have to build tools for converting large amounts of data present in databases, in powerpoint files, in excel spreadsheets into formats suitable for the web. Another approach might be to augment these custom data formats by META-DATA so as to perform more meaningful search over the data space. Thereafter I do think that we need many more sophisticated tools for the present web development community to start building applications and systems for the semantic web.

I find a great degree of overlap between agents and semantic web and it seems that the true vision of semantic web can only be realized with agents. Having profound interest in Agents, I find myself increasingly interested in the notion of semantics of data and how it can be interpreted in a practially meaningful manner. Though the word agents here refer to the traditional concepts of agents, I could think that instead of these, simple software processes could provide the enabling architecture for information processing.

Areas of Interest

OfCourse, my primary interest is DAML . However I have worked with RDF to a fair extent. Having questioned a lot of people, one of the questions that I find most intriguing is "Can you explain the transition between XML, RDF and DAML". I undertook a small exercise to be clear myself. Anyone interested in the Semantic Web would normally approach our group members and I tried to explain to them these concepts. Having done that, now I clearly understand how each one is distinct in its own rights.I am primarily interested in the use of DAML by agents in representing knowledge. Our implemenation experience says that the toolset for DAML is not yet mature and there is still a couple of years left before the toolset for it becomes mature ands stable. At the same time, industry is focussing hard on building toolset for RDF. Hence, constructing tools for manipulating these semantic data models does interest me a lot.

Another idea that interests me is Ontology. Which is the best way to construct ontologies, Bottom-up or Top-down? Should ontologies be domain dependant or domain independant. How do two ontologies contructed independantly but belonging to the same domain, ensure compliance with concepts(Ontology Mapping Problem). Another facet of ontologies that attracts me is this. If we build highly application specific ontologies, what steps are required for us to move to the broader vision of the semantic web.

Proposed/Accomplished Work

My work in Hewlett Packard Labs was a perfect blend of Context-Awareness and Semantic Web. The question we were trying to answer was how can semantics of data be used to intepret context of a user. To this end, we used RDF for expressing contextual information and contructed ontologies for generalizing the idea of context of an entity.

I have been involved with ITTalks to a great extent. Though I am more on the implemenation side of this vertical portal, but in the process did come across various issues in constructing ontologies, using them and making instances of these ontologies. Right now we are concentrating on Agents interacting with ITTalks to intepret semantics of data to be used in interesting reasoning processes. Ofcourse prolog and XSB have been an intergral part of this development process. During the course of this work, we did figure out means for doing efficient reasoning based on semantic markup. It is a three step process based on the language schema constructs, ontology constructs and application specific rules.

I prototyped ITTalks Event Notification Service as an illustrate case for the services in the AgentCities Network. This work was done primarily to investigate the use of semantic web concepts for service description, service composition and service availability. Current work at HP Labs involves the use of RDF and information management. I have also been involved in building the Damlator tool. It provides server side transforms for converting DAML into other interesting formats like N3, PNG. Right now its very simple and is representative of the class of helper services that will be required on the Semantic Web.

UMBC's DAML EFFORT

Semantic Web - Is It Here to Stay - Article written by Sovrin Tolia There is a strong buzzword going on in the AI, WWW, Database and KR research community - The Semantic Web. This fever has got a stronger hold in Europe than in US. I don't want to concentrate on what semantic web is. Tim Berners Lee has defined his view of what the semantic web should be and what the essential components of it are. I would agree with him when he outlines the principles underlying the semantic web.

Lots of money is being invested into the realization of the semantic web, but the key question that remains is "Is It Here to Stay". I find it difficult to believe that it will be real soon. I am no expert in this field and my views will be limited to that of a student who has worked in this area for 2 years and aspires to work further to somehow realize this dream. So what are the key problems that need to be tackled for the semantic web to become true?

I strongly believe that Semantic Web is an extension of the existing web. However, in this case, the web is marked up with Semantic Information so that machines processes can "understand" the content. For the Semantic Web to exist, annotations have to come in more strongly. I refer to two aspects of Annotations, one in which there are central Annotations servers where users/processes can post their comments on a piece of information. These annotations servers can serve as a dictionary of information semantics. For example, if a machine does not understand the concept "name" derived from a particular namespace, they might ask these annotations servers for more information on the concept hoping that some other process had discovered it earlier, and then it might post this understanding there. I sometimes wonder how the DNS Servers across the Internet help you find relevant host locations so easily. On similar lines, if we have a hierarchical organization of these annotations servers, the processes would exactly know the hierarchy to follow when trying to look for information. The way this hierarchy is organized is non-trivial. It might perhaps be organized as domain concepts. For example, a particular annotation server might hold information about concepts related to humans and other might host concepts related to devices. These annotation servers should not be viewed as the ones hosting information, but rather hosting information about information, meta-information. Moreover, they might be expected to host information related to concepts. Assume some machine process finds out that concept A from namespace X is the same as concept B from namespace Y. This information helps other processes looking for similar kind of mappings. There are additional questions that come in a scenarios like this like that of how can I trust this fact? More research is needed in order to investigate the anatomy of such annotation servers and how would one organize such servers to facilitate efficient retrieval.

The second aspect of annotations that I feel is required is to that of existing web information. When I go to http://www.gl.umbc.edu/~stolia1, there is an implicit assumption made that it's my homepage. When I provide a URL in my resume, again the assumption is that there are more details about me present in that location. When machines start processing such information, it may be important to provide meta-data about information. Meta-Data for the web perhaps might the way to go for automatic processing of this information. More Research is required to investigate further into how this meta-data will be structured, how will existing web-infrastructure be transformed to have meta-data, how search engines will take into account this meta-data when evaluating search queries provided by the user.

Semantic Web will represent an extreme form of distributed and diverse information stores. The success of the Semantic Web perhaps might be very closely coupled with the Success of Web-Services Infrastructure. I do support the view of Open Services Environments like Agentcities, hosting the so-called Service Networks. These network of services will provide semantically rich information to machine processes (At this point I would like to reassert, that I don't believe in Conventional Agent Theories like Intelligence, Sociability etc. primarily because I think realizing such abstract notions into practical systems is very difficult. I however do believe in Agents in the sense that they offer a different programming paradigm and help realize distributed systems in much more effective manner. (I have to be very frank here that my view of Agents would be heavily biased towards infrastructures perspective as I have been involved in using JADE framework quite exhaustively)). So when an entity is looking for information, the first source would be the Meta-data enabled Semantic Web and thereafter it would look into Open Service Networks like the ones provided by Agentcities type of environments.

One of the greatest problems I have encountered so far is that of building systems with the assumption of "Shared Ontologies". I personally feel that this is a very strong assumption to make in systems geared towards the semantic web. In the present web, there are no such restrictions. It looks to me that mapping of different Ontologies is an extremely difficult problem to solve. It is not quite clear to me how Ontologies and their notion of being shared helps in automating the information processing on the web. My question to the research community would be that could we have a technique devised that allows automated information processing without human intervention or more so partial intervention.

When an entity is looking for information, it might get bits and pieces from diverse places. Aggregating these bits of information to form a complete picture will be one of the most challenging problems for the semantic web community. Further, in this area, we might have questions like what if the information is contradictory; information is misleading and so on. Recently I attended a talk on google and its infrastructure and the interesting aspect of their system is that you cannot evaluate them against the correctness of answers (In a google search) there is no such thing as the right answer. I agree with this notion, because there might be potentially vast areas of the web that might not have been explored by google crawlers. Posting of experiences might be another key aspect of enhancing the semantic web. For example, if I am looking for information about Mike Franklin. Some information store might give me his academic background and some other might give me his personal information. If I can aggregate this information to make a valid answer for my query, I might want to pose this experience in lets say some annotation server stating that I was able to infer from the bits of information gathered from the following stores. This brings us to a very important paradigm in query-response systems. Today most of the query-response systems work this way. When a query is issued, the data stores respond back with certain pieces of information and the querying entity filters that to use the information it wants. It is assumed that the response from the data stores will have the information that was asked for. In queries on the semantic web, this might not be the case. Partial query results will be returned to the requestor that then would have to combine together in "some" manner to construct the right answer it was looking for. It would be a good issue to investigate as to how the querying entity constructs an expectation for the answer and how it merges different bits of information to come to the answer it was looking for. What I mean to say is that in the world of the semantic web, getting "perfect and precise responses" might be an exception rather than a rule.

Search Engines will have to play a key role in information gathering. If they are able to mine the semantics of data, the end-user will be able to find information that is much more relevant. Moreover the machine processes looking for information would find it helpful as an additional source where information can be found. What I am trying to suggest is Cooperating Architecture. Individual elements of the web (meta-data enabled), Service Networks and Search Engines could really make semantics of information accessible to humans and machine processes. I am a strong proponent of the idea of the Semantic Web and truly believe that it will catch on some day. However, I do have to agree that tools and technologies present today are perhaps insufficient to realize the true potential of the semantic web. More needs to be done on the technology front for the semantic web to kick in. To the end-user or the machine processes, they might never be aware of the existence of the semantic web. It has to be an Inconspicuous Adaptation. Semantic Web should "Dissolve" into the systems that we build and use.