Representation and execution of human know-how on the Web
View/ Open
Date
02/07/2018Author
Pareti, Paolo
Metadata
Abstract
Structured data has been a major component of web resources since the very beginning
of the web. Metadata that was originally mostly meant for display purposes gradually
expanded to incorporate the semantic content of a page. Until now semantic data on
the web has mostly focused on factual knowledge, namely trying to capture “what humans
know”. This thesis instead focuses on procedural knowledge, or in other words,
“how humans do things”, and in particular on step-by-step instructions. I will present a
semantic framework to capture the meaning of sets of instructions with respect to their
potential execution. This framework is based on a logical model which I evaluated in
terms of its expressiveness and it compatibility with existing languages. I will show
how this type of procedural knowledge can be automatically acquired from human-generated
instructions on the web, while at the same time bridging the semantic gap,
from unstructured to structured, by mapping these resources into a formal process description
language. I will demonstrate how procedural and factual data on the web can
be integrated automatically using Linked Data, and how this integration results in an
overall richer semantic representation. To validate these claims I have conducted large
scale knowledge acquisition and integration experiments on two prominent instructional
websites and evaluated the results against a human benchmark. Finally, I will
demonstrate how existing web technologies allow for this data to seamlessly enrich
existing web resources and to be used on the web without the need for centralisation. I
have explored the potential uses of formalised instructions by the implementation and
testing of concrete prototypes which enable human users to explore know-how and
collaborate with machines in novel ways.