How to make sense of the Leximancer analysis: a quick guide to interpreting concept maps

stemswedishAI and Robotics

Oct 15, 2013 (3 years and 7 months ago)

275 views

1


H
ow to make
sense

of the Leximancer
analysis
: a q
uick
g
uide

to interpreting concept maps

This
quick
guide has been developed
to
support

the
research

process
es

of those who
are new to
the

semantic analysis software
Leximancer and would like to make a prompt
s
tart with
interpreting

the

data.
The guidance is mainly
focused

on
making sens
e

of
the

concept
maps and
gives

some tips on what to look for when exploring
the data.
Although
a couple
of


how to do


bits of
advice

are
also
included,
generally
we assume

that
the
resea
rcher

would

know how to reach
the

output


(actual concept map) stage
. This guide
is based on the
latest
Leximancer 4
v
ersion
,
so
the
interface could be slightly different
from the
early
version
s

of
the

software
,
but
the
general
principles of
the

analysis
and core functions of
the

software remain
the

same
.



1.

What is Leximancer?


This is a
semantic
analysis
software

that
h
as been developed at the
University of Queensland, Brisbane
in 2001

(
https://www.leximancer.com/
).


2.

Why
/
w
hen

would
you want to use the software?

Leximancer is a u
seful instrument for
researcher
s
/analyst
s

who need
to

explor
e

a
large
text
-
based data
set

where
manual analysis
and
coding
would be too time consuming
,
e.g. qualitative survey

data
,
multiple interview/focus group transcripts
, lengthy reports or

web
-
based textual information
.



3.

How does it work?


The process is called
unsupervised semantic mapping of natural
language
.
The
method can also be thought

of as a form of text mining
.



Leximancer

employs two stages of

information extraction
:
semantic
and
relational
,
using a different algorithm for each

stage.
It

computes
the frequency with which each
word

is used

and then
calculates

the
distance between each of the terms
(co
-
occurrence
)
.
The algorithms
used are statistical, but they employ non

linear dynamics and machine
learning
.
The results of computations are displayed
as a concept map

that can be explored
on in
divi
dual concept level
s

and
also by looking
at
the
family of associations

between

different
concepts

(themes)
.



4.

What are
the

benefits of using
the

software?


L
eximancer
provides a fairly unbiased method of revie
wing complex
textual data sets
and a clear

process of justifying decisions about text
selection
.
It
make
s

the
researcher
aware of the

global


context and
helps

to
d
iscover

hidden


structures in text

that fall outside of
his/her
preconceived

framework.


2


T
he a
utomated analysis
demonstrates a
stability of measurement over
time

and what is most important
-

it
allow
s

a
more rapid and frequent
exploration of

text

with r
educed cost
.
Tailored

(researcher driven)
analysis also could be done
if
the
researcher

would like to explore
specific
topics/
concepts.


O
ne of
the

attractive features of
the

software is its ability to identify

sentiments by

showing

probability of a concept being mentioned in
a
favourable

or unfa
vourable
context.
Leximancer

could be used on its
own or

as a complementary tool for other methods of analysis.

5. Anything to
be aware of
/any limitations
?

There are some limitation
s

to

purely automated


analysis.
Some
c
oncepts emerge strongly where they are represented by a narrow
vocabulary.
Others

will be
identified from

a broader pool of terms and
ha
ve

a greater likelihood of being diluted as concept
s

in the map.

This
can be mitigated
by the
manipulations

with

a

thesaurus
(e.g. adding
concepts that haven

t make a relevance threshold to
the

concept map,
cre
ating combined
/
compound

concepts

etc)
and should be
acknowledged as a researcher
-
driven analysis.

Leximancer findings might benefit from being combined with outcomes
from other type of analysis such as traditional thematic analysis or
content analysis for gaining a more comprehensive picture.


Some more insight
s into the software and how it works

could be found
on
http://www.textinsight.net/sites/default/files/files/What%20is%20Lexim
ancer.ppt

T
he most exciting part is
actually using the tool!


First of
all...

First of all you need to run a preliminary analysis
of
your

unspoiled


data set,
with no editing or
configuration
,

to get a

feel


of
the
data.

If you are using
the

software for
the

first time, f
or a step by step
guidance

of how to
make a start
look at the
PowerPoint

presentation
created by
Julia Cretchley

and

Mike Neal

http://www.textinsight.net/sites/default/files/files/First%20Leximancer%
20A
nalysis.ppt


6.
How to
make sense of your first
concept map




Concept level
exploration

3


1.

Take
some

time to visually explore the concept map

generated
(you might want to
set
the
Theme Size scrolling Bar
on

0%

(
see Fig
. 1), so that you can
focus
on
the
concept
s
first.


Fig 1.



2.


When you
have
located the most
re
levant

concept (largest circle)
,
you
will
instantly
know what is of the most importance to
either
your
re
spondents/research participants

or to the text creators.

P
ay attention to the
concepts
positio
ned closely to
the
most
relevant

concept
-

t
hose with direct links/connections
would
indicate that these two
words
are
often used

together

and
worth
y

of detailed exploration
.

Things to note:

Wa
tch
out for language your research participants use.
S
tudents
often
use different terms to describe the same concept
-

it
might give you an indication of a specific connotation
or
a context
in
which
they choose to use
a

particular word.

For example

while
describing
academic

staff, students use different wor
ds such as

staff,
lecturers, tutors
,
teachers

etc. W
hat is interesting
is that
these
concepts/themes are not always located close to each other and
associated with a different
contextual and often
emotional
background.


4


3.

Explore other concepts (
those
that

constitute ‘nodes’ in the concept map)
. Their
relevance is reflected by size

(more relevant ones have bigger size)
. Note
where

they are positioned
on
the

map
and which
concepts they are
surrounded

by
. This
could

give you an idea of
other
significant

themes
and
how they are connected
.


4.

Look for
concepts
that are
less
important/could be excluded from

your analysis,
(these
, for example,

could be frequ
ently mentioned words
such as

certain

,

have


etc). Some concepts
could be
merge
d
(e.g. placement and placements)
or

ma
de
compound

(e.g. work +
experience
)
.
You could later make these

changes to

make
your concept map more readable



5.

E
xplore

the
Concept Ranking Table

located to the right from
the

concept map

(Fig 2)
.
T
he
relevance of the

concepts will be
indicated
in
the

table

numerically

(
number of
instances
/actual quotes

and
relevance
%)
.


Fig 2.




6.

By clicking on a concept on
the

map you can
see

how this concept is related to
others. It will be shown via
graphical

links and

also as a table indicating the likelihood
of
other

concept
s

being
mentioned together wi
th

the concept in question

(Fig 3)
.



5


Fig 3.




7.

When you
familiarised yourself with
the

layout of

the map and
have
a broad
understanding

of the

concepts
that are coming out from
the

data

and how they are
connected
,
it's

time
to explore the direct quotes/instances tha
t

formed
these concepts.

Go to the
Concept Ranking Table
(
located to
the

right from concept
map)
and under
the
Count

heading
c
lick on
a number
that

corresponds to the concept you would like
to explore

(Fig
4
)
. You will have instant access to a
ll the quo
t
es that
contributed to
the creation of

the concept.
The
quotes
will

give you an ind
ication of

the
meaning
(s)

behind the concept and
essentially

how ‘
semantically clean

is the
concept.

For
example

word

feedback



would
most likely
have
a
singular meaning

(Fig
5
)
, while

work
’ might
have multiple meanings,
since
students
, for example,


could talk about
their course

work, part
-
time work, work related learning, work as verb


e.g. ‘work it
out’ etc.
In case of multiple meaning
s

you might want to explore only specific
links/connections with
other

concepts

that ar
e of interest for your research or make
some adjustments
on
the

next, advanced stage of
the

data processing.








6


Fig
4
.


Things to note:

Not all
possible
quotes
where the concept (word) was
mentioned in
the

text
wil
l

be
included in the list. Only those that
passed the relevance threshold will be listed.


Fig 5
.


7




8.

If you are interested in

quotes/instances where two specific c
oncepts were
mentioned together
,

c
lick on a particular concept (under
WordLike
)

i
n the Concept

Ranking table
,

and when a list of rela
ted concepts appeared with
the icon
,

click
on
the

icon

attached to the concept
that
you would like to explore together with the
first concept.
(Fig
6
)



Fig
6
.


Exploration of
themes

Themes are
concept

clusters

that
represent the most
semantically
connected groups

of concepts.
Theme name is
the
most prominent
concept
in
the

cluster.

1.

Set your Theme size
scroll bar to 100%
-

to see
the
concept
(s
) with

the

highest level
of connectivity


these

will be the
most important

themes
that are
coming out from
your text

(Fig 6)
. There might be one theme, two

or
,

sometimes
,
more



they all are
worth paying close attention. You could explore relevance of
the

them
es and, as in

the

case of concepts,

have access to all quotes that are illustrative of the theme.

Things to note:

If you

ve got more than one theme at 100%
resolution
,
l
ook for
the

concepts that are positioned
in

the
intersected area
(
s
)



they might give you some insights of what
topics

belong


to both
themes.



8


Fig
.

7
.


2.

When
reducing
percentage
on
the

scroll bar
,

more
themes (concept clusters) will
appear
on
the

map

(Fig.8)
.
Look carefully at
the
dyn
amic
s

of transformation
-

e.g.
how smaller themes merge into bigger ones,
what

themes
disappear

and what
remain.

3.

R
emember or better save the themes generated by the
automatic

analysis
. It is
possible that when you
made
some

custom

configuration

changes
,
the

thematic
picture
will also

change.

Fig.8



9



N
ext step


The n
ext
step
is making
necessary

adjustments

for a more
focused
/
customized

analysis
,

for
example
,

removing

concepts that are
not central for your research
,

merging or adding the
concepts,
activating sentiment analysis
,
adding
tagging

etc.


When
all
the

changes you think might be useful
for your research
are

done
,

the
analysis
should be run once more, following the same
stages that were
indicated/described above.

7.
Sentiment

analysis

Activating
the
sentiment
lens

can
be a bit tricky, so below is
a quick
step by step
guide how to do it.


1.

To enable sentiment analysis, in the Project Control Panel
select


Generate
Thesaurus


and then activate

Show Settings

.


2.

Click on
Concept Seeds


and
then
on

Edit


(Fig
9
).

3.

Choose

User Defined Concepts

from the top menu
and
then
click on
Sentiment
Lens

button that
is

activated
/highlighted
.
Favourable

(favterms) and
Unfavourable

(unfavterms) terms will appear under the
Concept list

(Fig
10
)
.


4.

S
ave
the

settings by pressing

ok


and then run your analysis from scratch so that
s
entiment analysis is
incorporated
.


Fig.
9



10


Fig.
10



Why

use
the
sentiment analysis
?

Sentiment analysis

gives you an instant indication of

probability

of a
concept
being mentioned in a favourable and
unfavourable

con
text
.
J
ust
click

on a concept in the
Concept Ranking Table
and see
the

results
.
T
he difference could be
relatively small

(as in Fig
11



just 1%

or under
) or quite
noticeable
.
The latter case will be an
evidence of a
particular
strong
disposition

of your research
participants
.


By clicking on

favourable
/
unfavourable

you will have access to
all
quotes that were selected as indicative of
the

particular sentiment
for this concept.

Things to note:

Don

t be surprised if
,

for example
,

the
percentage of
unfavourable

likelihood

is
higher, but number of quotes selected by
the

software
will
be

the same
than in case of
favourable

quotes. This
is not a direct/
proportional
relationship
,
as

it is

determined by
various
statistical procedures and indicative
of more complex relationships

between
the

concept

and strength of the sentiments expressed.

Things to n
ote
:

Pay attention to the location of

favourably

and
unfavourably

rated
concepts on
the

map.
Sometimes
the

same
sentiment

concept
s

cluster together
,
this would be an indication of a
strong
shared concern
s
/issue
s
.
It is also useful to explore themes in
relation to dominant sentiment background of
the

concept that form
the theme.






11


Fig
11
.



Final remarks

You could
use

some
advanced
function
s
useful

for the analysis

-

for
example t
agging
f
iles and groups of file
s

to
allow

for
comparison
,
c
reat
ing

D
ashboard Report
s
, explore word cloud
, look at your concept
map from

the


social
network’

perspective etc
.
I
nstructions how to do it
can
be found in
the

Help


section of
the

software.

Some advance
features
are
also
explained in
https://www.leximancer.com/dl/training/2011
-
12
-
V4194059105
-
AB
-
45
-
J/Leximanc
erIn
-
depth.ppt

The m
ain
and most time consuming
work will be about
delv
ing
into the

actual quotes
and
extract
ing
meaning
s of
the

concepts and concept
connections
, making sense of

themes and general landscape of your
text.

When you
have
had a chance to compare
the outcomes of

a
‘fully
automated’ analysis
and
a
researcher
-
driven
,
customized

one

you
might decide that automated analysis
is

more insightful. It is
easier

to
start everything from scratch (as a new project)
rather

than trying to

undo


the changes that were made.

Comparative analysis of concept maps

Leximancer could be
a helpful tool when you need to compare two
data sets or to look at the data
longitudinally
.
Look for
new, emerging
concepts

and concepts that
have
disappeared
,
explore
how
the
relevance of
the

concepts and
themes

have
changed,
what
happened

to direction and
strength

of
the
sentiments for the key concept, how
positioning of
the

main concepts and
their

neighbours changed.

Comparative analysis could be
especially

insightful!


12


Please send your questions or comments
related
to

this guid
e

to
Dr
E
lena
Z
aitseva (
e.zaitseva
@
ljmu.ac.uk
)


7.

References


1.

Smith, A.
&

Humpreys
.

M.
,

2006
,


Evaluation of unsupervised semantic mapping
of
natural language with Leximancer concept mapping

,
Behavioural Research
Methods
,
(
38
), pp.
262

79.


2.

Dodgson
et al
., 2008


Content analysis of submissions by
Leximancer

. Available at

http://www.innovation.gov.au/Innovation/Policy/Documents/LeximancerSubmissionA
nalysis.pdf

(accessed
15

April 2013
)