Tillägg till "Intelligent Information Filtering" 
96-12-20

3.5 User studies

3.5.1 Think-aloud-study

In the years of 1993 and 1994 Ann Lantz made two studies that should make 
clear how experienced users of the Usenet News chose the information they 
want to read (Lantz ,1995). In the first study she let the subjects think aloud 
while they were looking for the contributions they wanted to read. In the 
other study Lantz sent an electronic inquiry to members in Swedish 
speaking news groups.  The results show that the subjects usually made 
their choice on keywords in the subject line.

3.5.2 Field investigation

During 1995 and 1996 Eva Fåhræus conducted a user study to find out the 
user´s opinion about the PEFNA filtering system and to search for possible 
improvements. At the same time the study produced a sample of classified 
texts to use as test material in the experiments with the filtering system.
 
The study was conducted as a field investigation. Three users used the 
PEFNA prototype to read Usenet News and placed the contributions in 
different categories or "baskets", depending on content and user's interest.

The behaviour of the users and the result of their manual filtering were 
studied through interviews and then analyzed by inductive methods, based 
on Grounded Theory  (Strauss & Corbin, 1990).

The goal was not to formulate a theory as such, but to try to get new ideas 
about how to support users in search of interesting texts.

The analyse lead to the following formutation of the core of the studied task:
 
It is about how a user, by throwing an eye on a part of a text, make a brief judgement what it is 
about and if (s)he finds it worth reading, now or later, and, in that case, in which situation. 
The judgement could also be that she/he needs to read more before making the decision.

This formulation can be translated into a more abstract level:

Perception of elements in a contribution give, if they are recognized, an association that 
results in a text classification which decides whether the text is saved for later use.

3.5.3 Generated ideas

Grounded on these core categories and inspired of ideas born during the 
interviews and meetings we formulate how the computer could support the 
reader.

The perception can be supported by the system presenting the elements in 
the contribution more clearly. The subject line and the keywords can e.g. be 
displayed brighter,  in bold font or in different colours. You could also show 
questions and answers together and perhaps in a different colour or type. 
Connections between contributions in a  chain should also be shown 
graphically. By showing the complete contribution in a smaller scale, one 
can quickly give an impresion of the contributions outline, format and size. 
These measures will make it easier for the reader to observe those elements 
that they are likely to recognise and that are valuable for the classification.

To support the association the system could show the readers' baskets next 
to the contribution so that the reader can be reminded of the different kinds 
of classifications she has chosen. The readers shall also be able to easily enter 
a basket and control the contributions they have put there and this way 
remember their strategy of classification.

Saving the contribution can be made simpler by direct manipulation, i.e. 
you can click and pull the contribution over the screen.

3.5.4 Conclusions

The results from the field investigation confirms the conclusions drawn by 
Ann Lantz (1995), that is that keywords in the subject line are very 
important as to how the subjects judge the contributions, and completes it 
with a more detailed image of how the subjects act.

Through this study we have received support for the assumption that the 
computer should be used to present the contributions in a clear way and that 
different readers have different wishes and needs. There are differences 
between different readers and the same reader on different occasions. 

We have not got any support for the theory that even a RATHER good 
sorting of the contributions can be of great help. The study lasted too short a 
time to make the subjects really able to use the prototype efficiently. 

The study does not give any reason to reject the filtering technique chosen 
but the subjects say they doubt it is possible to "teach" the computer what is 
interesting or uninteresting. Instead they suggested some criteria to use.

Filters can very well be used by groups in co-operation, by letting different 
kinds of contributions be guided to different individuals that this way can 
watch different areas for one another. The thought that  letting readers put a 
judgement on contributions, that others can benefit from (so called social 
filtering) is not supported in this study.

Sid 33 Chapter 4 Epilogue ¿

---
This study has strongly supported the idea of graphical presentation as a 
means to help readers find interesting texts. The combination of computer's 
possibility to sort and display information graphically and the human ability 
to recognize patterns and discriminate graphic properties will open new 
posibilities to help us take advantage of the information now sunk in the 
ocean of Internet.

Future work on this topic ought to focus on different ways to display the 
properties of a text that carry meaning to the reader and on the human 
ability to visually discriminate them.

It is however important to take in consideration that readers are different. 
Physiologically they have different ability to discriminate different kinds of 
perceptions and intellectually they find different properties in a text more or 
less important for judging the interest.

To test our conclusions it would be interesting to conduct a longer user test 
with a further developed prototype containing graphical presentation of 
different properties in texts. The prototype should be flexible so that the user 
can adjust it to his or her actual situation and preferences. If the study goes 
on for at least two months, preferably more, there is a chance that the user 
really adapts his/her behaviour to the tool and finds out the best way to use 
it. If different users, working under different conditions, take part in the test 
the flexibility of the prototype could provide a good picture of useful support 
functions.

The prototype could for instance have the following functions:

The user can give keywords that are prefered or not prefered in texts. 
Prefered words are shown green, not prefered are shown red.

The user can give maximum and minimum length of texts.

The user can give maximum (or minimum) ratio of cited text.

The texts are shown graphically as circles or cylinders with their size 
illustrating length, the colour telling if there are prefered or not prefered 
keywords.

The user can chose to look closer to one part of the presentation, e.g. one or 
two categories.

When pointing at a text symbol, the closeness to a category is shown by 
arrows and the layout can be examined by displaying the text in "stamp size" 
where the subject line is readable and the keywords shown in colour.

From any of these presentations it should be possible to click and read the 
text, to draw a text to a folder for later reading, to drop it in the waste paper 
basket or to send it to some other person, adding a comment.


Eva R Fåhræus