Semantic search has two legs

The discussion of semantic search has gradually become popular. Just not long time ago, semantic search was thought to be barely a little bit more than a dream. At present, optimistic researchers have started to believe its possibility in the near future. Very recently at Read/WriteWeb, Dr. Riza C. Berkan, the CEO of Hakia (a company declared to perform "semantic search"), posted an article about semantic search that attracted much attention. Despite of agreeing with the post, here are more thoughts about semantic search.

Semantic search has two legs: semantic understanding and proactive collaboration. Until now, however, most semantic search articles only have focused on the first one. Including Hakia, an "ideal" semantic search engine is popularly thought to be alike a "semantics-enhanced Google." This is, however, a narrowed thought. The intension of semantic search is more beyond "semantics + search."

In order to better understand these two legs, we may watch a regular semantic search scenario in human society that is, however, often overlooked. We humans have daily practised a type of semantic search very successfully for centuries. We ask questions; everybody asks questions, from children to adults. We ask questions to look for answers. These question-and-answer behaviors are typical semantic search activities.

When we are young, we look for answers from parents, whose words are oracles to us. When we grow older, we look for answers from teachers, whose words are oracles to us. When we grow even older, we start to realize that there are indeed no oracles. We start to look for answers by ourselves. In particular, we make friends with various specialities. These friends become our sources of question answering when we get troubles in particular realms. At the meantime, we ourselves also become such a type of sources to our friends. These links constitute a delicate, complicate, and successfully executed network of semantic search in our human society.

If we take a closer look at this successful semantic search network, we can find two fundamental factors that support its execution. First, its success relies on the ability of semantic understanding at each but not some of its nodes. It is generally believed that the set of human knowledge is too rich and too complicated to be executed in a centralized way. For instance, Mor Naaman at Yahoo! Research very recently said that "there is no way that we can engage the masses in annotating media with 'semantic' labels" in a WWW2007 panel. Therefore, representations of global semantics are better to be distributed widely other than be accumulated onto only a few special nodes. In consequence, every node in this semantic search network has its ability to perform a certain level of search depending on its own capability of semantic understanding. This is the basis of a successful semantic search network.

Beyond the local semantic understanding on every node, a successful semantic search network also requires proactive collaborations. In a search network, some nodes (such as professors) may have much greater capability than others (such as first-grade students). But even the node with the greatest ability is still very much limited in its search capability when the search space is about the whole set of human knowledge. A successful semantic search network demands well collaboration among individual search nodes. Moreover, such a type of collaborations appeals to being proactive.

Proactiveness is a unique factor in the network of friendship. The network of friendship is not only a regular social network, but also a search network. When we get troubles, we used to get to our friends for help. Nevertheless, we often make friends on purpose, i.e. in contrast to randomly or aimlessly. A successful semantic search network in human society is priorly built on the joint or depending interest of individuals. For example, both John and Mary love music; so John actively make friend with Mary. Another example, John play piano and does not know to tune a piano; Kate, however, is good at tuning pianos. For the sake of his future requirement of piano-tuning, John proactively make friend with Kate. The third example, Rose is good at history literature; John, however, does not like to read history literature. In consequence, John inactively make friend with Rose. These examples show that the establishing of a search network very much depends on the proactiveness (which in turn decided by the semantic understanding of interests) of these nodes to make connections.

In summary, semantic search naturally contradicts to the centralized web search strategy. In order to activate semantic search to the practical level, we need a search network that is participated by all web users beyond the few independent and aggregated semantic search nodes such as Hakia. The entire web search strategy must experience some revolutionary change other than simple makeups. In the second part of my article about web evolution, I have more discussions about the collaborative search for the future semantic web.

The initial draft of this post is published at SemanticFocus.