updating and structuring

[helm.git] / helm / mathql / doc / mathql_introduction_avsets.tex

\subsection {Sets of attributed values.}

The data representation model used by {\MathQL} relies on the notion of 
\emph{set of attributed values} ({\av} set for short) that is, in practice,
the only data type available in {\MathQL}.4. In this sense {\MathQL}.4 is a
statically untyped language.%
\footnote
{A type system that fits {\MathQL} as an {\RDF}-oriented query language,
should be driven from the {\RDFS} class system. This may be a future
improvement.}
Each {\av} in an {\av} set consists of a string% 
\footnote{When we say \emph{string}, we mean a finite sequence of characters.}
(that we call the \emph{head string} or \emph{value}) and a (possibly emty)
multiset of named attributes whose content is a set of strings.
Attribute names are made of a (possibly empty) list of string components, so
they can be hierarchically structured. 
Moreover the attributes of a value are partitioned into a set of \emph{groups}
({\ie} subsets) to improve its structure.

In the above description a \emph{set} is an \emph{unordered} finite
sequence \emph{without} repetitions wheras a \emph{multiset} is an
\emph{unordered} finite sequence \emph{with} repetitions.

In the present context repetitions are defined as follows:
two {\av}'s are repeated if they share the same head string without any
condition on their attributes, two groups are repeated of they contain the
same attributes (equal both in name and content), two attributes of a group
are repeated if they share the same name without any condition on their
content, and two strings are always compared in a case-sensitive manner.%
\footnote
{The Author's experience with {\MathQL} seems to show that the above
definition of an {\av} set is just the right one among the many alternatives
that were tried.} 

As we said, {\MathQL}.4 uses {\av} sets to represent many kinds of
information, namely:

\begin{enumerate}

\item
A pool of {\RDF} triples having a common subject $r$, which in general is a
{\URI} reference \cite{URI}%
\footnote 
{A {\URI} \emph {reference} is a {\URI} with an optional fragment identifier.},
is encoded in a single {\av} placing $r$ in the head string.
The predicates of the triples are encoded as attribute names and their objects
are placed in the attributes' contents.
These contents are structured as multiple strings with the aim of holding the
objects of repeated predicates.
Moreover structured attribute names can encode various components of
structured properties preserving their semantics.

\begin{figure}[ht]
\begin{footnotesize} \begin{verbatim}
The RDF triples:
("http://www.w3.org/2002/01/rdf-databases/protocol", "dc:creator", "Sandro Hawke")
("http://www.w3.org/2002/01/rdf-databases/protocol", "dc:creator", "Eric Prud'hommeaux")
("http://www.w3.org/2002/01/rdf-databases/protocol", "dc:date", "2002-01-08")

The corresponding attributed value:
"http://www.w3.org/2002/01/rdf-databases/protocol" attr
             {"dc:creator" = {"Sandro Hawke", "Eric Prud'hommeaux"}; "dc:date" = "2002-01-08"}
\end{verbatim} \end{footnotesize}
\vskip-1pc
\caption{The representation of a pool of {\RDF} triples} \label{AVOne}
\end{figure}

\figref{AVOne} shows how a set of triples can be coded in an {\av}.
Note that the word \emph{attr} separates the head string from its attributes,
braces enclose an attribute group in which attributes are separated by
semicolons, and an equal sign separates an attribute name from its contents
(see \subsecref{Textual} for the complete {\av} syntax).

In this setting the grouping feature can be used to separate semantically
different classes of properties associated to a resource (as for instance
Dublin Core metadata, Euler metadata and user-defined metadata). 

\item
A pool of arbitrarily chosen {\RDF} triples is encoded in an {\av} set 
placing different {\av}'s the subset of triples sharing the same subject.

Note that the use of {\av} sets to build query results allows {\MathQL} queries
to return sets of {\RDF} triples instead of mere sets of resources, in the
spirit of what is currently done by other {\RDF}-oriented query languages.

If the {\av}'s of an {\av} set share the same attribute names and grouping
structure, this set can be represented as a table in which each row encodes
an {\av} and each column is associated to an attribute (except the first one
which holds the head strings).
\figref{Table} shows an {\av} set describing the properties of two resources
``A'' and ``B'' giving its table representation, in which the columns
corresponding to attributes in the same group are clustered between
double-line delimiters%
\footnote{A table with grouped labelled columns like the one above resembles a
set of relational database tables.}.   

%Another possible use of a {\MathQL} query result is for the encoding of a
%relational database table: in this sense the indexed column is stored in the
%subject strings, the names of the other columns are stored in attribute names
%and cell contents are stored in attribute values.

\begin{figure}[ht]
\begin{footnotesize} \begin{verbatim}
"A" attr {"major" = "1"; "minor" = "2"}, {"first" = "2002-01-01"; "modified" = "2002-03-01"};
"B" attr {"major" = "1"; "minor" = "7"}, {"first" = "2002-02-01"; "modified" = "2002-04-01"}
\end{verbatim}
\begin{center} \begin{tabular}{|c||c|c||c|c||}
\hline   & {\bf ``major''} & {\bf ``minor''} & {\bf ``first''} & {\bf ``modified''} \\
\hline ``A'' & ``1'' & ``2'' & ``2002-01-01'' & ``2002-03-01'' \\
\hline ``B'' & ``1'' & ``7'' & ``2002-02-01'' & ``2002-04-01'' \\
\hline
\end{tabular} \end{center} \end{footnotesize}
\caption{A set of attributed values displayed as a table} \label{Table}
\end{figure}

The above example gives a spatial idea of the geometry of an {\av} set ({\ie}
a query result) which fits in 4 dimensions: namely we can extend independently
the set of the head strings (dimension 1), the attributes in each group
(dimension 2), the groups in each {\av} (dimension 3) and the contents of each
attribute (dimension 4).

The metadata defined in the table of \figref{Table} will be used in subsequent
examples.
For this purpose assume that \TT{first} and \TT{modified} are the components
of a structured property \TT{date} available for the resources ``A'' and ``B''.

\item
The value of an {\RDF} property is encoded in a single {\av} distinguishing
three situations:

\begin{itemize}

\item
If the property is unstructured, its value is placed in the {\av} head
string and no attributes are defined.

\item
If the property is structured and its value has a main component%
\footnote{Which is set by the \emph{rdf:value} property or defined by a
specific application.},
the content of this component is placed in the {\av} head string and the
other components are stored in the {\av} attributes as in the case 1.

\item
If the property is structured and its value does not have a main component,
the {\av} head string is empty and the components are stored in the
attributes.

\end{itemize}

\begin{figure}[ht]
\begin{footnotesize} \begin{verbatim}
First example, one instance:
"" attr {"major" = "1"; "minor" = "2"};  no main component
"1" attr {"minor" = "2"};                main component is "major"
"2" attr {"major" = "1"}                 main component is "minor"

Second example: two separate instances:
"" attr {"major" = "1"; "minor" = "2"}, {"major" = "1"; "minor" = "7"}; no main component
"1" attr {"minor" = "2"}, {"minor" = "7"}                            main component is "major"

Third example: two mixed instances:
"" attr {"major" = "3", "6"; "minor" = "4", "9"} no main component
\end{verbatim} \end{footnotesize}
\vskip-1pc
\caption{The representation of the structured value of a property}
\label{AVTwo}
\end{figure}

\figref{AVTwo} (first example) shows three possible ways of representing in
{\av}'s an instance of a structured property \TT{id} whose value has two
fields ({\ie} properties) \TT{major} and \TT{minor}.
In this instance, \TT{major} is set to ``1'' and \TT{minor} is set to ``2''.
The representations depend on which component of \TT{id} is chosen as the
main component (none, \TT{major} or \TT{minor} respectively).
Several structured property values sharing a common main component can be
encodes in a single {\av} exploiting the grouping facility: in this case the
attributes of every instance are enclosed in separate groups.
\figref{AVTwo} (second example) shows the representations of two instances of
\TT{id}: the previous one and a new one for which \TT{major} is ``1'' and
\TT{minor} is ``7''.

Note that if the attributes of the two groups are encoded in a single group,
the notion of which components belong to the same property value can not be
recovered in the general case because the values of an attribute form a set
and thus are unordered. \newline
As an example think of two instances of \TT{id} encoded as in \figref{AVTwo}
(third example).

\item
A natural number is stored, using its decimal representation, in the head
string of a single {\av} with no attributes.

\item
The boolean value \emph{false} is stored as an empty {\av} set, whereas
a non-empty {\av} set may be interpreted as the boolean value \emph{true}.
The default representation of \emph{true} is a single {\av} with an empty
head string and no attributes.

\end{enumerate}

{\MathQL} defines five binary operations on {\av} sets: two unions, two
intersections and a difference. The first four are defined in terms of an
operation, that we call \emph{addition}, involving two {\av}'s with the same
head string.
The result is an {\av} with the same head string of the operands but there are
two ways to compose the attribute groups:

\begin{itemize}

\item
With the \emph{set-theoretic} addition, the set of attribute groups in the
resulting {\av} is the set-theoretic union of the sets of attribute groups in
the operands.

\item
With the \emph{distributive} addition, the set of attribute groups in the
resulting {\av} is the ``Cartesian product'' of the sets of attribute groups
in the two operands. 
In this context, an element of the ``Cartesian product'' is not a pair of
groups but it is the set-theoretic union of these groups where the contents of
homonymous attributes are clustered together using set-theoretic unions.

\end{itemize}

\figref{Addition} shows an example of the two kinds of addition.

\begin{figure}[ht]
\begin{footnotesize} \begin{verbatim}
Attributed values used as operands for the addition:
"1" attr {"A" = "a"}, {"B" = "b1"}
"1" attr {"A" = "a"}, {"B" = "b2"}

Set-theoretic addition:
"1" attr {"A" = "a"}, {"B" = "b1"}, {"B" = "b2"}

Distributive addition:
"1" attr {"A" = "a"}, {"A" = "a"; "B" = "b2"}, {"B" = "b1"; "A" = "a"}, {"B" = {"b1", "b2"}}
\end{verbatim} \end{footnotesize}
\vskip-1pc
\caption{The addition of attributed values}
\label{Addition}
\end{figure}

Now we can discuss the five operations between {\av} sets that we mentioned
above:

\begin{itemize}

\item
The two unions ocorresponds to the set-theoretic union of their operand where
the {\av}'s sharing the head string are are added either set-theoretically or
distributively as explained above (thus we have a set-theoretic union and a
distributive union in the two cases). In this context the empty {\av} set
plays the role of the neutral element. 
These operations play a central role {\MathQL} architecture and allow to
compose the attributes of the operands preserving their group structure.

\item
The two intersections are the dual of the above unions: they contain the
{\av}'s whose head string appears in each argument where {\av}'s sharing the
head string are added either set-theoretically or distributively as before.

The distributive intersection has the double benefit of filtering the
common values of the given {\av} sets, and of merging their attribute groups
in every possible way. This feature enables the possibility of performing
additional filtering operations checking the content of the merged groups.

\item
The difference of two {\av} sets contains the {\av}'s of the first
argument whose head string does not appear in the second argument.

\end{itemize}

\figref{Binary} shows how the above operations work in a simple example.

\begin{figure}[ht]
\begin{footnotesize} \begin{verbatim}
Sets of attributed values used as operands for the operations:
"1" attr {"A" = "a"}; "2" attr {"B" = "b1"} 
"2" attr {"B" = "b2"}

Set-theoretic union:
"1" attr {"A" = "a"}; "2" attr {"B" = "b1"}, {"B" = "b2"}

Distributive union:
"1" attr {"A" = "a"}; "2" attr {"B" = {"b1", "b2"}}

Set-theoretic intersection:
"2" attr {"B" = "b1"}, {"B" = "b2"}

Distributive intersection:
"2" attr {"B" = {"b1", "b2"}}

Difference:
"1" attr {"A" = "a"}
\end{verbatim} \end{footnotesize}
\vskip-1pc
\caption{The binary operations on sets of attributed values}
\label{Binary}
\end{figure}