Recent Events

New Graph Searching

New Graph  Searching

New Graph Searching

Greg Notess, Team Leader at Montana State University and author of Search Engine Showdown noted that a huge transition in web search is occurring: searching keywords and matching words on a page is moving to a focus on the semantic web and an emphasis on presenting more direct results instead of lists of abstracts, like this one for the hours of the National Air & Space Museum.

Typical Knowledge Graph

Typical Knowledge Graph

Google describes this as a knowledge graph of real-world things and connections to give more meaningful results to users. Sources of knowledge graph information include Wikipedia, Freebase, DBPedia, structured data, and Google search data. Knowledge graphs generally appear on the right side of the screen.

Carousel

Carousel

Sometimes a carousel appears that shows related topics (see above illustration).  The appearance of the carousel depends on the search and the country where the searcher is located. Bing has similar displays and data sources.

Graph searching originated with a 2001 proposal in Scientific American by Tim Berners-Lee who proposed that structured documents read by software would create a “web of data”. The web originated with unstructured data, but now there is a much higher incentive to structure data for the web.

Other systems employing graph searching include Wolfram Alpha and Microsoft Office Delve.  Since it is part of the Office platform, Delve allows users to use graph searching on their own documents. (Facebook used to have graph search, but it is no longer available.) Structured graphing requires links to related data; schema.org is a markup system used by Bing, Google, Yahoo!, and Yandex that is used to create the display. Creative works, embedded non-text objects, events, organizations, places, and similar data can be tagged using schema.org.  Google’s Guide for Developers describes some opportunities for structured data, a tool for testing the data by showing the markup, as well as examples of promoting events and customizing your knowledge graph. When a site changes, the knowledge graph can adapt, and errors can be corrected.

Sometimes one does not see the knowledge graph, simply the answer to a question. Google calls these “Quick Answers”. See moz.com/blog/how-we-fixed-the-internet for an example of how answers can be changed. In about 25% of the answers, there is no attribution to the data source. The knowledge graph results are sometimes also pulled into the inline results, but condensing knowledge sometimes creates interesting situations that can be hard to interpret. The best way to check the accuracy of the retrieved information is to use data for which the correct answer is known by the searcher.

Google is moving into medical information and has hired licensed illustrators to develop images for the knowledge graphs. The medical information includes typical symptoms, treatments, frequency of occurrence, and ages affected.  It is also building “knowledge vaults”, which may be the next generation of knowledge graphs. Its knowledge vault is built from 1.6 billion facts in its database, of which 271 million are described as “confident facts with more than a 90% chance of being true”. However, Greg noted that this statement should give librarians pause as to the overall accuracy of the information from the knowledge vault in search results. Google has published a recent research paper on the technique to automate the identification of factually accurate websites.

Knowledge graphs are a relatively recent innovation, so we can expect to see their further development and growth.

Don Hawkins
Conference Blogger

Comments are closed.