Semantic networks in cognition and computing


Internet infrastructure currently operates in many networking modes; however, to most people, it is the web: a technology stack of files, their unique identifiers, and transfer protocols to serve to user agents. However, now that the web has evolved to include ubiquitous web services, search functions, and other indirect queries or manipulation, the the underlying infrastructure makes it difficult for machines to handle content of files served. A semantic network approach to the web may help re-train the computerized infrastructure to behave in a more connectionist way and better serve an increasingly automated, emergent Internet.

Human cognition, especially in memory, often uses spreading activation to identify the piece of information sought after from related concepts. For example, if someone were trying to figure out the name of a vegetable that’s orange and crispy, they would likely think of those concepts and explore related items, eventually converging on the concept of a carrot. To achieve the same thing on the web, the best a user could do is query a search service for a string like “vegetable orange crunchy.” A Google search now yields a fairly unsatisfactory result, seeking documents that include instances of the string itself.  Instead, if a semantic web were in place, the user could search via properties for a file that isa ‘vegetable’ hascolor ‘orange’ and hastexture ‘crunchy’ to arrive at a web page for a carrot. Recipe website Yummly achieves this to some success.

Even if a page for a term the user searches for were not in the network or database, there may be similar results via spreading activation. For example, say a user were searching a site detailing local food systems for organic carrots within 20 miles of Georgia Tech. The system may not find a page for ‘organic carrot,’ but it did find the string ‘carrot’ mentioned in a page for a farmer 15 miles away. The system may then afford several degrees of fan by returning a map clustering pointers to the farmer’s house, the land covered by the farm, and the nearby zones. The user can surmise from the surrounding dirty industrial zoning that the carrots are likely not organic, wary of the potential runoff from the industrial sites. This sort of guided search is one way that the web is turning into more of a semantic network. Similarly, many database querying services use sentence and word verification to accommodate for typos or related concepts. If the user accepts a verification suggestion, the system can be trained to more closely associate items to search terms.
Semantic web progenitor Tim Berners-Lee and his World Wide Web Consortium (W3C) have made strides in bringing the semantic web to fruition via health care and government projects, while Semantic MediaWiki enables ordinary Internet users to build their own platforms. Eventually, a semantic web can be a significant step toward Internet as an artificial neural network, capable of more advanced computing and interconnectivity.