Google has switched from a rule-based method of detecting useful data to an AI machine learning technique. This, plus the knowledge graph, produces better search results.
For a while it has been obvious that the next step for search has to be the incorporation of AI methods that can weigh up the relevance of a web page to a particular request. Google Research has moved in this direction, albeit in a very niche area of search. Google Fusion tables have offered a table search facility for some time, but a recent makeover has attempted to improve the quality of the results it returns.
The big problem with the web is that while we should be using HTML as semantic markup - that is to tell you about the structure of a page; most web pages still use HTML to control the layout.
One of the most noteworthy culprits of this confusion is the <table> tag. This is supposed to be used to present tables consisting of rows and columns of data complete with headings. However, many web pages use <table> to create a tabular layout without the content having anything to do with data. Indeed a typical table tag is likely to have data cells that contain complete articles and pictures.
In short, the table tag is not a sure sign that you have found some data in table form.
Google's problem is that some queries ask for data tables and, for the reasons already explained, it can be difficult to distinguish what constitutes a data table. In the past Google used a set of rules to try and weed out general pages from true data tables. Now, however, it has moved to using machine learning in the form of a Support Vector Machine SVM. The SVM that Google uses is a little special in that it uses multiple kernels and a method of combining kernels invented by Google Research.
Once the SVM has determined that a table is indeed a data table, Google's recently introduced Knowledge Graph is used to identify topic and context. This improves the quality of the response to any query asking for data of a particular type.
A less important, but more directly noticeable, improvement is that you can now import the data tables you find directly into Google Drive as Fusion Tables. You can then work with the data and draw charts.
The TransportLAB Hackathon is a worldwide event giving teams and individuals the opportunity to wrangle transpiration and logistics datasets. Register now and be ready for a two-week sprint running fr [ ... ]