Other References:
OLAP On-line analytical Procressing with TM/1 EF Codd & Associates 1994.*
J.
Goldstein et al., "Using Aggregation and Dynamic Queries for Exploring
Large Data Sets," Human Factors in Computer Systems, Boston, MA USA,
Apr. 24-28, 1994, pp. 23-29.
M. Spenke, et al., "Focus: The
Interactive Table for Product Comparison and Selection," GMD--German
National Research Center for Information Technology, pp. 41-50.
OLAP, On-Line Analytical Processing with TM/1, E.F. Codd & Associates, 1994.
Execu-View: Visualizing the Multidimensional Enterprise, A Practical Solution for Mangerial Investigation, Comshare, 1992.
Computers & Accounting, Essbase at Symantec, Management Acctg., Jun. 1994.
Essbase in Action, Arbor Software, 1994.
F. Hayes, "Data Staging Suits Quantum," Open Systems Today, 1994.
R. Finkelstein, "MDD: Database Reaches the Next Dimension," Database Programming & Design, vol. 8, No. 4, Apr. 1995.
E. F. Codd et al., "Beyond Decision Support," Computerworld, Jul. 26, 1993, pp. 87-90.
R. C. Bolt, "Essbase to the Rescue," DBMS, vol. 8, No. 3, Mar. 1995.
Manojit
Sarkar et al. "Graphical Fisheye Views of Graphs" Proceedings of the
1992 Conference on Human Factors in Computer Systems, May 1992
Monterey, California pp. 83-91.
J. Goldstein et al., "Using
Aggregation and Dynamic Queries for Exploring Large Data Sets" Human
Factors in Computer Systems, Boston, MA USA, Apr. 24-28, 1994, pp.
23-29.
OLAP, On-Line Analytical Processing with TM/1, E.F. Codd & Associates 1994.
Execu-View: Visualizing the Multidimensional Enterprise, A Practical Solution for Managerial Investigation Comshare 1992.
Computers & Accounting, Essbase at Symantec, Management Acctg., Jun. 1994.
Essbase in Action, Arbor Software 1994.
F. Hayes, "Data Staging Suits Quantum" Open Systems Today, 1994.
R. Finkelstein, "MDD: Database Reaches the Next Dimension" Database Programming & Design, vol. 8, No. 4, Apr. 1995.
E.F. Codd et al., "Beyond Decision Support" Computerworld, Jul. 26, 1993, pp. 87-90.
R.C. Bolt "Essbase to the Rescue" DBMS, vol. 8, No. 3, Mar. 1995.
Sarkar et al. "Stretching the Rubber Sheet: A Metaphor for Viewing Large Layouts on Small Screens" ACM, Nov. 1993, pp. 81-90.
R.
Finkelstein, "Understanding the Need for On-Line Analytical Servers,"
White Paper, URL: http://www.arborsoft.com/papers/finkTOC.html (1995).
Arbor
Software, "The Role of the Multidimensional Database in a Data
Warehousing Solution," URL:
http://www.arborsoft.com/papers/rolapTOC.html (1995).
Michael
Spenke, Christian Beilken, and Thomas Berlage, "Focus: The Interactive
Table for Product Comparison and Selection," In Proceedings of the ACM
Symposium on User Interface Software and Technology (UIST), Nov. 6-8,
1996, pp. 41-50, Seattle, WA.
Description:
FIELD OF THE INVENTION
The
present invention relates generally to the field of data access and
interaction, and more particularly to the analysis and visualization of
multidimensional datasets.
DESCRIPTION OF THE RELATED ART
The
need for organizing large bodies of data has lead to the development of
many types of data management systems, ranging from simple files to
relational databases. Although these systems provide suitable storage
and query features, they are not very well suited for exploratory
analysis, and using them requires a non-trivial degree of technical
expertise.
In response to this problem, the field of On-line
Analytical Processing (OLAP) has emerged. The basic premise of OLAP is
that end users think of their data in terms of a number of dimensions
and would like to be able to explore the data by manipulating these
dimensions in a number of ways. For example, a business manager may
think in terms of Products, Channels, Regions, Years, and so on. He may
then want to perform operations such as rearranging views using
different dimensions, slicing and dicing the data along various
dimensions, drilling down into subgroups within a dimension, or rolling
up subgroups into aggregate totals. An OLAP server allows users to
perform these operations on data accessed from relational databases.
Typically, a spreadsheet provides users an exploratory environment for
visualization for slices of data. A number of quite powerful front-end
tools suited to this kind of exploration of multidimensional data have
been developed. However, these tools are still largely based on textual
views and don't leverage powerful human perceptual abilities that could
support exploration.
U.S. Pat. No. 5,632,009 to Rao et al.,
which is hereby incorporated by reference, describes a tool that allows
users to view multivariate datasets (i.e., datasets having two
dimensions, one dimension of variables and the second of cases
associated with the variables) using a mixed graphical/textual
representation of the data. This "focus+context" technique allows for
the visualization and manipulation of large two-dimensional tables
(roughly 30-100 times as big in the same screen space as a conventional
spreadsheet or table browser). Because it displays much more of the
table at once by using graphics to show values, a user can examine
patterns in the whole table as well as zoom in on specific content
without losing global context.
Although the focus+context
technique, as described in the Rao patent, is a powerful tool for
viewing and manipulating multivariate data, it does not allow for
effective interaction and manipulation of multidimensional data. There
is, therefore, a need in the art to provide efficient analysis and
visualization of multidimensional data, such as the data manipulated by
OLAP systems.
SUMMARY OF THE INVENTION
Objects and
advantages of the invention will be set forth in part in the
description which follows, and in part will be obvious from the
description, or may be learned by practice of the invention. The
objects and advantages of the invention will be realized and attained
by means of the elements and combinations particularly pointed out in
the appended claims.
To achieve the objects and in accordance
with the purpose of the invention, as embodied and broadly described
herein, a first aspect consistent with the present invention includes a
method of visualizing a multidimensional data set. The method includes
the steps of (1) storing the multidimensional data set using an
abstract data model partitioned into dimensions; (2) converting
portions of the data set stored in the abstract data model into a
visual model having dimensions of the abstract data model organized as
at least one hierarchical tree; and (3) displaying the visual model to
a user as a tabular representation on a computer display screen,
wherein a first portion of the data of the data set is displayed in a
first level of detail in the tabular representation and a second
portion of the data in the data set is displayed in a second level of
detail in the tabular representation, the first level of detail using
more screen space per data value than the second level of detail.
A
second aspect of the present invention is directed to a computer system
comprising a processor, a display, and a memory. The memory includes
computer instructions and computer data, the computer instructions when
executed on the processor causing the processor to perform the
functions of: (1) storing the multidimensional data set using an
abstract data model partitioned into dimensions; (2) converting
portions of the data set stored in the abstract data model into a
visual model having dimensions of the abstract data model organized as
at least one hierarchical tree; and (3) displaying the visual model to
a user as a table in which a first portion of the data in the table is
displayed in a first level of detail and a second portion of the data
in the table is displayed in a second level of detail, the first level
of detail using more screen space per data value than the second level
of detail.
A third aspect of the present invention is directed
to a method of invoking an operation on a data set having three or more
dimensions comprising the steps of: (1) converting portions of the data
set into a two-dimensional visual model; (2) displaying the visual
model on a physical medium; (3) detecting a user's interaction with the
data represented in the visual model; (4) initiating an operation on
the data set based on the detected user interaction with the data; and
(5) updating the visual model to reflect the operation initiated by the
user.
BRIEF DESCRIPTION OF DRAWINGS
The accompanying
drawings, which are incorporated in and constitute a part of this
specification, illustrate several embodiments consistent with this
invention and, together with the description, help explain the
principles of the invention. In the drawings,
FIG. 1A is a
block diagram of an exemplary computer system which may be used to
implement a method consistent with the present invention;
FIG. 1B is a block diagram showing the interaction of software components consistent with the present invention;
FIGS. 2-4 are diagrams illustrating the relationship between dimensions and dimension keys;
FIGS. 5 and 6 are diagrams illustrating N-dimensional data sets;
FIG. 7 is a diagram illustrating grouping of sticks and slabs of an N-dimensional data set;
FIG. 8 is a diagram illustrating dimensional planes perpendicular to the record dimension;
FIGS.
9 and 10 are diagrams illustrating a dimension hierarchy for a three
dimensional data set of vault size 3.times.2.times.2;
FIG. 11
is a diagram illustrating the arrangement of two hierarchical trees
associated with the horizontal and vertical screen axis;
FIG. 12 is a diagram illustrating tabular representation of the dimensional hierarchy shown in FIG. 11;
FIG. 13 is an illustration of an exemplary table image of a multidimensional data set;
FIG. 14 is an illustration of a typical business multidimensional data set;
FIG. 15 is an illustration of a multidimensional data set after performing the select-slice operation;
FIGS. 16A, 16B, and 17 are illustrations of a multidimensional data set on which the promote/demote operations are performed;
FIG. 18 is an illustration of a multidimensional data set after performing the aggregate operation; and
FIG. 19 and 20 are illustrations of a multidimensional data set after performing the repeat-variables operation.
DETAILED DESCRIPTION:
Reference
will now be made in detail to embodiments of the invention, examples of
which are illustrated in the accompanying drawings. Wherever possible,
the same reference numbers will be used throughout the drawings to
refer to the same or like parts.
A software visualization tool
consistent with the present invention integrates OLAP functionality
with focus+context based techniques for navigation through and
inspection of large multidimensional datasets. The tool supports a
number of operations including: select-slice, aggregation,
promote/demote, repeat-variables, and sort.
I. The Physical System
FIG.
1A is a block diagram of an exemplary computer system used to implement
the visualization tool. Computer system 102 includes a chassis 110,
which holds the computer's main processor and main memory (in which the
visualization tool may be stored); input devices such as keyboard 112,
and a pointing device such as a mouse 113; a secondary storage device
such as floppy or hard disk drive 114; and a display such as monitor
116. Computer system 102 is optionally connected to a network 18, and
may be operated directly by a user or through network 118.
Many
variations of computer system 102 are possible. For example storage
device 114 may additionally include storage media such as optical
disks, and mouse 113, may additionally or alternatively include other
pointing devices such as a trackball, a lightpen, a touch-sensitive
pad, a digitizing tablet, or a joystick.
FIG. 1B is a block
diagram showing the software components of the visualization tool. The
user interacts with the visualization tool through user interface 120,
which is displayed on monitor 116. Formula processor 121 parses
arithmetic expressions input by the user and may create new variables
for addition to the data model. File layout component 122 parses the
input data and feeds it to data model module 123, which contains an
abstract data model (described in more detail below). Visual model
module 124 contains rules for converting the abstract data model in
data module 123 to a form representable on display 116.
Context
object 130 acts as an intermediary between components 120-122 and
123-124. More specifically, context object 130 performs methods which
convert information flowing between elements 120-124, if necessary.
Thus, context object 130 allows for simplification from both a
technical and a practical point of view. For example, formula processor
121 does not need to know what the interface to the data model is, even
though its results may eventually be integrated into the data set.
Instead, it calls a method in context object 130, which later calls the
appropriate method in data model 123. Similarly, the user interface 120
communicates with the visual model through context object 130.
II. The Abstract Multidimensional Data Model
In
one embodiment, the visualization tool stores and retrieves data for
visualization using an abstract data model, such as a multidimensional
data cube. The word "abstract" signifies that the data model does not
directly correspond to how data is represented in memory or visualized
on the screen. Based on the user's commands, the abstract data model is
reduced to a visual model which may then be viewed on computer monitor
116.
The following defintions apply to the abstract data model and the visual model:
A
value is the smallest unit of data which continues to have meaning in
the physical world. There can be even smaller units of data, like bits
and bytes, but they are only relevant to how a value is represented,
not what it means.
A variable is an attribute of a physical
system in consideration which gives rise to data values. For example
heart rate, atmospheric pressure, gross domestic product, and revenue
are all variables.
A dimension is an independent partitioning
of the set of all values according to some aspect of the state of the
system incidental with each value. Each partition is labeled by a
dimension key. For example, a set of revenue values may belong to a
variable in the company marketing system. The set can be partitioned
according to the product responsible for a particular revenue value,
and also according to the year when such revenue was generated.
Therefore, "year" and "product" are two dimensions of the company
marketing system.
A multidimensional data set comprises data
from a single physical system. A single physical system can be
described in terms of multiple dimensions, one of which defines
multiple variables. For example the description of the company
marketing system may include several dimensions such as "distribution
channel," "product," and "year," and contain several variables such as
"profits" or "units sold."
In summary, each value in a
multidimensional data set belongs to a variable within a physical
system. The context of each value observation corresponds to the state
of the system incidental with that observation.
A. Dimension Key Structure
A dimension key is a characterization of a relevant aspect of the system's state. A key can be simple or compound
A
simple key represents the finest interesting partitioning of the data
set along its respective dimension. For example, the "male" and
"female" keys of the gender dimension are simple because they represent
the finest possible partitioning according to gender.
A
compound key is in fact a hierarchy of keys which label finer
partitions. For example, as shown in the dimension key hierarchy of
FIG. 2, the partitions associated with the keys "Contact Lens" and
"Vitamin Tablets," of the product dimension, could be partitioned
further into "Disposable Contact Lens" and "Non-Disposable" contact
lens, and "Vitamin Tablets for the young" and "Vitamin Tablets for the
elder." Attributes such as "Disposable," "Non-Disposable" cannot be
taken out of the context of the key hierarchy, because they only apply
to one dimension key, namely "Contact Lens."
There is also an
intermediate key between simple and compound, called a pseudo-compound
key. Consider the "Year" dimension of the company marketing system. A
"Year," as shown in FIG. 3, can be further partitioned into four
quarters.
The difference between a pseudo-compound key and a
compound key is that the sub-partitions of a pseudo-compound key repeat
over all the keys of that dimension. For example, all years in the
"year" dimension have four quarters. This situation occurs very often
especially with dimensions which quantify time and space. In such
cases, even though it seems that a quarter is an inherent part of a
particular year, it is reasonable to speak of a quarter outside the
context of that year. For example, studying yearly fluctuations in the
company's marketing might include examining how variables such as
"profits" and "revenue" behave as fictions of the quarter regardless of
the actual year. Thus, for all practical purposes, years and quarters
behave as the keys of two independent dimensions, as illustrated in
FIG. 4.
B. Record and Set Dimensions
Integrating
values from different variables into a single data set can be handled
uniformly, consistent with the multidimensional data model, by
designating a special dimension whose keys correspond to the existing
variables. This special dimension is called the record dimension and
all other dimensions are called set dimensions. Thus, a set of values
whose dimension keys are all equal, except for the record dimension
key, are incidental with the same system state and represent a record
of observations; hence the name "record dimension." On the other hand,
a set of values with the same record dimension key are a set of
observations of the same variable; hence the name "set dimension."
The
keys of a record dimension are variable descriptors. Each variable
descriptor consists of a variable label, e.g. "profits," and a value
domain. The value domain characterizes the type of operations that are
meaningful to the values of the variable referred to by the variable
descriptor.
C. Value Domains
Depending on the variable
to which a value belongs, it may participate in certain data operations
and not others. For example, the values of the "revenue" variable can
be added together or ordered, whereas the values "red," "green" and
"blue" of the "color" variable cannot.
Values belonging to the
same variable are able to participate in the same data operations.
Depending on the data operations of the current data model, it may be
possible that the same set of operations is applicable to the values of
several variables. We can classify values into value domains according
to the set of operations on the values which are meaningful.
Exemplary value domains include:
Nominal.
These values belong to a finite set. They can only be tested for
equality but cannot be ordered. For example, the three primary colors
are nominal values.
Quantity. These values represent an amount
of something. They can be tested for equality and can be ordered. In
addition, they can be added together or multiplied by a fraction. For
example, height, pressure, profits, etc. are variables with quantity
values.
Textual. These are values which support typical string
operations, like concatenation and search. For example, the values of
the variable "remarks".
III. The Abstract Data Model Used By The Visualization Tool
The
previous section described the general notion of the multidimensional
data model and its components. The visualization tool uses an instance
of the multidimensional data model having particular operations, value
domains, and terminology. Even though the data model used by the
visualization tool is more specific than the general multidimensional
data model described in the previous section, it will continue to be
referred to herein as the "abstract" data model to distinguish it from
the "visual" model. One of ordinary skill in the art will recognize
that alternate instances of the multidimensional data model other than
the one described here could equivalently be used.
An
N-dimensional data set used by the data visualization tool can be
expressed in terms of spatial relationships. In such case the data
values are organized into an N-dimensional array, such as the
N-dimensional array 501 shown in FIG. 5. Vault 501 is a
multidimensional array of cells, where a single cell 502 is a
placeholder for a single value. The context in which the value is
observed is encoded by the spatial position of the cell.
The
structure consists of N dimension objects. A dimension object has of a
list of unique keys and a label. As shown in FIG. 6, label 601
("Product") is the name of the dimension. Keys 602 correspond to
spatial coordinates. Therefore, since each cell in the vault is
specified by an N-tuple of spatial coordinates, it can be also
specified by an N-tuple of dimension keys. The correspondence between
spatial coordinates and dimension keys is what establishes the
correspondence between the cell's spatial position and the
interpretation of its value.
Stick 603 is a unidimensional
group of cells parallel to a dimension. The cells within a stick share
all but one spatial coordinate, i.e., they share all but one dimension
key.
Slab 604 is a two dimensional group of cells parallel to
two of the dimensions. The cells within a slab share all but two of the
dimension keys. A slab is basically a grouping of parallel sticks. FIG.
7 illustrates the grouping of sticks into slabs, and slabs into
3-dimensional blocks. The groupings shown in FIG. 7 correspond to
successive levels in a data consolidation path.
As shown in
FIG. 8, cells whose values belong to the same variable form (N-1)
dimensional structures perpendicular to the record dimension.
IV. The Visual Model
The
visual model is the projection of the abstract data model on a visual
medium, such as two-dimensional computer display screen 116. The visual
model contains components similar to the components in the abstract
data model. Because of the constraint that every visual component must
have a straightforward two-dimensional representation, however, the
visual model loses some of the simplicity of the abstract data model.
For example, in order to visualize multiple dimensions, a dimension
hierarchy is introduced.
In order to avoid confusion between
the dimensions of the abstract multidimensional model and the
dimensions of the screen, the screen dimensions will be referred to as
axes. A screen has two axes: a horizontal axis and a vertical axis.
A. Linear Dimension Hierarchy
Before
introducing the idea of a planar dimensional hierarchy, such as that of
a computer screen, it is useful to describe the simpler concept of a
linear dimensional hierarchy.
In a linear dimensional
hierarchy, the visual vault is a collection of sticks, aligned along a
single screen axis. The dimensions are arranged in a chosen order of
seniority. For purposes of illustration, and without loss of
generality, assume that the order is chosen to be: D.sub.1, D.sub.2, .
. . , D.sub.N ; where D.sub.1, D.sub.2, . . . D.sub.N are each
dimensions from the set of N dimensions. Given this order of
dimensions, the top level of the hierarchical tree corresponds to
D.sub.1 and the branches correspond to the keys of D.sub.1, the level
of nodes below corresponds to D.sub.2 and so on. This concept is
illustrated in FIG. 9. Leaves 902 of the tree store the actual data
values. Furthermore, the leaves of any node from the lowest level of
the tree contain data values which share all but one dimension key and
therefore those data values form a stick, such as stick 901.
The
example shown in FIG. 9 illustrates the dimension hierarchy for a
three-dimensional data set, with an abstract vault size of
3.times.2.times.2. The hierarchical tree describes the arrangement of
sticks along a particular screen axis. As shown, stick 901 is
associated with the horizontal screen axis.
A more compact way
of representing the dimensional hierarchy of FIG. 9 is shown in FIG.
10, in which the. branches are represented by boxes and the nodes are
omitted. This is referred to as the tabular representation of the
dimension hierarchy. Each level of the hierarchy is labeled by the
label of the corresponding dimension.
B. Planar Dimensional Hierarchy
In
the case of a planar dimensional hierarchy, the visual vault is a
collection of slabs tiled in the plane of the two screen axis. The
visual schema consists of two hierarchical trees associated with the
horizontal and vertical screen axes. These hierarchies determine the
arrangement of the slabs on the screen, as shown in FIG. 11.
The
two hierarchical trees contain two complementary sets of dimensions.
Within each tree the dimensions are arranged in a seniority order.
There is no such ordering defined between dimensions from different
hierarchical trees. As shown in FIG. 11, dimension D.sub.3 is senior to
dimension D.sub.4 and dimension D.sub.1 is senior to dimension D.sub.2.
Similar to the linear dimensional hierarchy shown in FIG. 9,
the leaves of the dimension trees in FIG. 11 contain data values.
However, unlike the hierarchy in FIG. 9, each leaf contains an entire
row or column of data cells depending on whether the hierarchical tree
is associated with the horizontal or vertical axis. Since a data cell
can belong to exactly one row-column pair, a cell is completely
specified by a pair of leaves from each tree.
The tabular view representation of the dimensional hierarchy from FIG. 11 is shown in FIG. 12.
To
provide a compact representation of the abstract data model, the visual
model may suppress the representation of data cells with nonexistent
values (i.e., values not defined) whenever it is possible to do so
without creating inconsistencies. For example, if all cells in row 1201
addressed by k.sub.4 [2] in FIG. 12 contained nonexistent data values,
then row 121 would be eliminated and the gap created would be closed by
shifting the bottom of the table up, without altering the spatial
relationships among the rest of the cells.
On the other hand,
if only half of the cells of row 1201 had nonexistent values, e.g. the
cells addressed by k.sub.4 [2] and k.sub.1 [1], then these cells 1201
would remain represented in the visual model.
The rules for suppressing the representation of cells in the visual model may be summarized as follows:
To preserve the spatial relationships between cells, only entire groups should be eliminated.
In
order to represent all essential data, only empty groups can be
eliminated, i.e., groups of cells with nonexistent data values.
C. Operations on the Visual Model
The
user invokes operations on the abstract data model by operating on the
visual model. The result of those operations are conveyed to the user
through the visual model. The set of operations on the visual model act
as an interface to the operations on the abstract data model.
FIG.
13 illustrates a table image 1300 of a multidimensional data set
showing a focus+context representation. More particularly, table 1300
is a visualization of stock market data, which contains the following
values for each of a number of days for various securities: volume
(column 1301), highest price (column 1302), lowest price (column 1303),
and closing price (column 1304). Each of values 1301-1304 is
partitioned by the dimensions date (columns 1310), security name
(column 1314), and stock market (column 1315). In other words, for any
given combination of date, stock name, and stock market, there is not
more than one possible value for volume, highest price, lowest price,
and closing price.
The data model represented by table 1300
has four dimensions: three set dimensions for date, stock name, and
stock market, and one record dimension for the variables volume,
highest price, lowest price, and closing price. The date dimension is
actually a compound dimension consisting of the pseudo-compound keys
year (column 1311), month (column 1312), and day (column 1313).
Altogether there is one record dimension and three set dimensions, one
of which has a 3-level compound structure. The three levels of the date
dimension may be treated as separate dimensions.
As shown, the
values in columns 1301, 1302, 1303, and 1304, are sorted based on the
single dimensional hierarchy tree given by: market, stock name, and
date. Further, the user has focused on certain values in table 1300,
causing these values to become fully visible in their textual
representation. These values are labeled as areas 1330. Other values in
table 1300, such as the values labeled 1340, are not focused. These
values are in the "context," and are represented graphically. In one
embodiment, the user may choose either a focus view or a context view
via a simple manipulation of a pointing device, such as a mouse. As
previously mentioned, the focus+context technique allows values of
particular interest to the user to be displayed using a high focus
level, i.e., using the full textual representation of the value. The
human operator controls the areas to focus on using an input device
such as pointing device 113.
A user may manipulate the data
shown in table 1300 to, for example, "slice and dice" the data along
various dimensions, "drill down" into subgroups within a dimension," or
"rollup" subgroups into aggregate totals. A number of multidimensional
focus+context operations are supported by the visualization tool to
achieve these objectives. For example, the user may drag the tile
labeled "high" (which corresponds to the "high" record variable) in
table 1300 and place it between the tiles labeled "low" and "close,"
causing the visualization tool to redraw table 1300 with column 1302,
corresponding to the "high" variable, located between columns 1303 and
1304.
In addition, several other operations are supported by
the visualization tool. These additional operations, described in more
detail below, allow the user to move between multidimensional views of
the data set. These operations will be explained with reference to
FIGS. 14-20, which illustrate a multidimensional data set displayed as
tables using the focus+context technique.
Table 1400 is a
business multidimensional data set having six dimensions:
years/quarters (columns 1401), products (column 1402), product
distribution channels (column 1403), sale regions (column 1404),
salespersons (column 1405), and the record dimension (columns 1406).
The record dimension includes the line item values: units (column
1407), revenue (column 1408) and profit (column 1409). Each of the
dimensions shown in columns 1401-1404 has multiple keys. For example,
the product dimension (column 1402), contains multiple keys, such as
keys 1420-1423. Keys 1420-1423 may correspond-to products such as the
"ForeFinancial" product or the "ForeRecreation" product.
1. The Select-Slice Operation
The
select-slice operation slices a dimension D at a specific key value y
of the dimension to produce a data model with one less dimension and
which only contains the values addressed by V. Slicing the data set
allows the user to navigate to a subset of interest in the data set. In
one embodiment, the select-slice operation is preferably initiated by
the user with a simple gesture using pointing device 113, such as
pointing to a predetermined section of the column dimension to be
selected and "flicking" (i.e., quickly moving) the pointing device in a
predetermined direction, such as northeast.
Table 1500
illustrates the data set of FIG. 14 after performing the select-slice
operation twice successively on product dimension 1402 and on channel
dimension 1403. The other dimensions remain encoded in the slice and
continue to exhibit the cyclical patterns reflective of their nature as
dimensions.
As shown, product dimension 1402 and channel
dimension 1403 are no longer encoded in table 1500. These two
dimensions, and the subgroup of the dimension they are frozen at (i.e.,
the "ForeFinancial" product sold through the "Direct Sales" sale
channel), are shown in menu section 1501 of table 1500. The units,
revenue, and profit variables 1502 contain values for the ForeFinancial
product sold through the direct sales channel, and hierarchically
sorted based on year, quarter, region, and-salesperson.
The
slicing operation can be used to focus on a selected subset of the
data. Slicing n-2 times will zoom on one particular slab defined by the
remaining two dimensions. Slicing one more time will produce a stick,
and finally slicing again will produce one value addressed by the
sliced values for each dimension. This is often referred to as a
"hierarchical drill-down."
2. The Promote/Demote Operations
The
promote and demote operations are complements of one another. These
operations can be applied to either axis of the visual model.
The
demote operation reduces the dimensionality of the visualized data set
by merging the slabs at the lowest level of the dimensional hierarchy
of the target axis. Stated more formally, when a dimension is demoted,
the dimension is represented as a variable, and the variable is
populated with values that were the keys of the dimension.
FIG.
16A illustrates an exemplary data set before applying the demote
operation. The demote operation is preferably initiated by the user by
moving a mark on the target axis. The mark is shown in FIG. 16A as the
thick vertical line 1601. As shown in FIG. 16B, mark 1601 has been
moved to the left of the "Year" and "Month" dimensions, thus demoting
these dimensions to the level of the record variables. That is, the
"Year" and "Month" dimensions are encoded in the table as data values.
The
promote operation is the reverse of the demote operation. In the
promote operation, dimension variables are represented as a dimension
in which the values of the dimension variables become keys of the
dimension.
The demote operation discussed above may be
reversed by promoting the "Year" and "Month" columns in FIG. 16B. This
operation may be initiated by moving the mark 1601 to the right of the
"Year" and "Month" columns. The result of this promotion is show in
FIG. 16A.
Table 1700 also-illustrates the promote operation,
in which the salesperson dimension has been promoted. In table 1700,
the salesperson dimension of table 1600 has been moved to the far left
(top of the dimension hierarchy) of axis mark 1701. The "Year,"
"Quarter," and "Region" dimensions have been demoted to the level of
the record variables.
In summary, the promote operation
converts variable values to keys of a dimension, while the demote
operation converts dimension keys to variable values.
3. The Aggregate Operation
The
variables may be aggregated over a promoted dimension to generate
summary values corresponding to the sticks or slabs referring to each
of the dimension keys. This operation is called the aggregate
operation. The summary information may include, for example, totals,
averages, or extreme values. Table 1800 illustrates the result of the
aggregate operation applied to table 1700, in which sticks of table
1700 referring to the salesperson dimension have been aggregated and
the aggregated values have been focused on and the non-focused values
removed from the visual display. Aggregated table 1800 may be used, for
example, in a final presentation document.
4. The Repeat-Variables Operation
FIG.
19 illustrates the use of the repeat-variables operation. The
repeat-variables operation takes the particular items of a dimension
and repeats all the non-dimensional columns (i.e., the items of the
variables dimension) underneath the items of the dimension. Table 1900
has had the select-slice operation performed on it, selecting the slice
corresponding to the "mail order" portion of the "channel" dimension.
The select-slice operation has left data for two salespersons, labeled
"Rebecca Greep" and "Norma Jones." The repeat-variables operation
duplicates the units, revenues, and profit variable for each of the two
salespersons.
FIG. 20 illustrates a summary table 2000. By
further aggregating on the products dimension and focusing on the
summaries of table 1900, table 2000 is generated, which is a summary
table similar to table 1800.
Functionally, the
repeat-variables operation is equivalent to a promotion on the
horizontal axis; and is thus a specialized version of the promote
operation.
5. The Sort Operation
The sort operation
sorts the sticks along a column or a row of the visual model. The sort
operation is preferably initiated by the user with a flick gesture
along the target row or column, the direction of which indicates the
direction of the sort (i.e., ascending or descending). When a stick is
sorted, the same permutation is applied to the sticks parallel to it.
One
use of the sort operation is to determine how well two parallel sticks
are correlated: if after sorting one stick, the other stick appears to
be more or less sorted then the two sticks are fairly correlated.
The
software visualization tool described above allows easy and intuitive
navigation of a multidimensional data set. Focus+context based
navigation techniques have been used to increase the clarity and
information content provided to the user. The visualization tool
supports a number of operations including: select-slice, aggregation,
promote/demote, repeat-variables, and sort.
It will be
apparent to those skilled in the art that various modifications and
variations can be made to the present invention without departing from
the scope or spirit of the invention. For example, although the above
aspects of the present invention were described using a two-dimensional
visual model, a three dimensional visual model could also be used to
present the data set to the user. The three dimensional visual model
would be organized as multiple levels of tables that form a three
dimensional rectangle.
Other embodiments of the invention will
be apparent to those skilled in the art from consideration of the
specification and practice of the invention disclosed herein. It is
intended that the specification and examples be considered as exemplary
only, with the true scope and spirit of the invention being indicated
by the following claims.
