TAG CLOUDS FOR SITUATED INTERACTION

AND PLACE PROFILING

Rui José, Bruno Silva

Information Systems Department, University of Minho, Guimarães, Portugal

Fernando Reinaldo Ribeiro

Informatics Department, Polytechnic Institute of Castelo Branco, Castelo Branco, Portugal

Keywords: Tag clouds, Situated interaction, Public displays, Bluetooth, Places, Pervasive computing.

Abstract: Tag clouds have become very popular as visual representations of the main topics in document sets or as

navigation tools that can provide quick access to resources related with specific topics. However, their

ability to represent the information environment associated with any meaningful reality in a way that is

collectively visible, actionable and easily understood may also be very relevant, even when the reality being

represented is no longer a set of documents or resources, but a stream of interactions occurring within a

particular ubiquitous computing environment. In this paper, we explore the use of tag clouds within the

context of situated displays and services. We hypothesise that such tag clouds may have a role as dynamic

representations of place and also as interaction controls, supporting the same comprehension and navigation

functions of classical tag clouds. We describe two case studies in which this concept of situated tag cloud

has been experimented in real-world settings. The case studies demonstrate two different applications of the

tag cloud concept as the basis for place description and situated interaction. The results obtained from the

case studies suggest that situated tag clouds can indeed provide valuable representations of place and

situations and can also support simple interaction models, allowing people to reason about the system

behaviour and how it is being influenced by new interactions.

1 INTRODUCTION

A tag cloud is a visualization of a weighted list of

words in which those weights are associated with

visualization attributes, typically size and colour.

Tag clouds have become very popular as a

visualization mechanism for the main topics

associated with a web site or set of documents. The

tags are obtained from the words in the text and the

size corresponds to the relative frequency of each

word. They offer a visual representation of the main

topics in that text, providing a simple and yet

powerful perspective of the respective content. Tag

clouds are also extensively used in crowdsourcing

systems where people can freely tag photos, web

sites, videos or other resources. In this case, the

generation of the tag cloud becomes a collective

process of knowledge creation and the tag cloud

may become a navigation tool, providing quick

access to resources related with specific topics.

What is very powerful about the concept of tag

clouds is their ability to represent the information

environment associated with any meaningful reality

in a way that is collectively visible, actionable and

easily understood. They may facilitate

comprehension by offering an alternative and instant

illustration of the main topics associated with a

particular entity, such as a document, a web site, or a

person’s interests. They may also facilitate

navigation by providing an alternative to more

conventional navigation patterns, such as those

based on menus. In other words, they provide a

social affordance in the sense that they convey social

interaction of fellow users and potentials for social

interaction (Bielenberg and Zacher, 2005). These are

valuable features that may still be useful, even if the

reality being represented is no longer a set of

documents or resources, but a stream of interactions

occurring within a particular ubiquitous computing

environment. As stated by Joe Lamantia (Lamantia,

2006) , “tag clouds are revolutionary in their ability

to translate the concepts associated with nearly

anything you can think of into a collectively visible

and actionable information environment”.

296

José R., Silva B. and Reinaldo Ribeiro F..

TAG CLOUDS FOR SITUATED INTERACTION AND PLACE PROFILING .

DOI: 10.5220/0003337502960301

In Proceedings of the 7th International Conference on Web Information Systems and Technologies (WEBIST-2011), pages 296-301

ISBN: 978-989-8425-51-5

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

Our aim is to explore the use of tag clouds within

the context of situated displays and services. In these

environments, people may interact with physically

situated services, such as public displays, Bluetooth

hotspots or location-based services. This type of

interaction has the potential to generate words,

which in turn may be used to dynamically generate

an evolving tag cloud that somehow represents those

situated interactions.

We hypothesise that such tag clouds may have a

role as dynamic representations of place and also as

interaction controls, supporting the same

comprehension and navigation functions of classical

tag clouds. They may support comprehension by

offering a representation of the topics associated

with a place. They may also support navigation by

offering an aggregate interaction mechanism,

whereby keywords in the tag cloud provide an

immediately available and dynamically evolving list

of interaction suggestions. The tag cloud concept

may thus be used as a unifying concept for the

association between keyword inputs and dynamic

behaviours. It can provide a visualization that may

be relevant in itself as a particular characterization

of place and at same time work as a driver for the

display behaviour by enabling the selection or

generation of situationally relevant content.

In this paper, we describe a system for

generating place-based tag clouds and two case

studies in which this concept of place-based tag

cloud has been experimented in real-world settings.

The results suggest that situated tag clouds can

indeed play a role as dynamic representations of

place and also as interaction controls, although there

are some open issues with the interaction models.

2 RELATED WORK

Many researchers have reported alternative

applications of the tag cloud concept to represent

realities outside their original context, including

personal profiles, social dynamics in meetings or

trends in political discourse.

Steinbock et al. studied the use of wearable tag

clouds in face-to-face interaction (Steinbock et al.,

2007). Within the context of an academic meeting,

participants were given a large badge with a tag

cloud of the most common words in their published

documents. This was expected to represent a

synopsis of the respective interests and facilitate

interaction between participants. McNaught and

Lam (McNaught and Lam, 2010) explored tag

clouds to analyse the spoken and written responses

of informants in focus groups transcripts. The tag

clouds facilitate the study of the social dynamics in

those groups, but this analysis is only conducted at a

later stage and provides no feedback on the social

interaction as it unfolds. Viégas and Wattenberg

report on some of the uses of IBM web site Many

Eyes, where people upload and visualize data in a

variety of ways (Viegas and Wattenberg, 2006). One

of those visualizations enables people to represent

their profile through a tag cloud generated from the

set of blogs that person normally reads. Tagline

Generator (Mehta, 2006) supports the generation of

chronological tag clouds from time-based text data

sources. This work is an interesting example of a

system that deals with the time dimension. In these

tag clouds, the colour in the words is associated with

usage variations. Words whose usage is increasing

will be brightened, while words whose usage is

decreasing will be fading away. The use of tag

clouds for content recommendation has been

described by Pessemier et.al (Pessemier et al., 2009).

Tag clouds are generated from user ratings to create

a form of personal profile. These tag clouds are then

used to recommend movies to that person. In our

work, we also suggest the use of tag cloud for

recommendation purposes, but in our case this

corresponds to a place or situation profile, and not to

the profile of a single individual.

This related work demonstrates the variety of

applications for the tag cloud concept. However, the

use of tag clouds as a situated representation of place

remains to be explored.

3 SITUATED TAG CLOUDS

A situated tag cloud is a tag cloud generated from

the keywords extracted from implicit and explicit

interactions observed in the context of a particular

situated system. The most distinguishing property in

the case of situated tag clouds is the way the input

words are obtained. Instead of counting the words

within a particular set of documents, we have a

continuous stream of words being generated by

various types of situated interactions, such as

Bluetooth names, Obex exchanges or SMS/MMS

messages. Additionally, we have a specific physical

and social setting within the context of which we

need to interpret those words. As part of this

research, we have developed a system for creating

tag clouds based on the digital footprints generated

from presence and interaction events associated with

a public display. The tag cloud generation

mechanism is part of a larger system, called instant

places, in which people can use their Bluetooth

device name for managing their presence and

TAG CLOUDS FOR SITUATED INTERACTION AND PLACE PROFILING

297

activating interactive features. The details of instant

places are outside the scope of this paper, but the

reader is referred to (José et al., 2008) for a more

detailed account of the system characteristics.

3.1 Situated Identities

The instant places system recognizes the presence of

Bluetooth devices and generates persistent identities

that may evolve with the history of presence and

interaction in that particular place. The system also

recognizes parts of the Bluetooth device name as

explicit instructions, allowing people to use their

Bluetooth name to trigger specific behaviours on the

situated display, such as the presentation of specific

content. The resulting data model, composed by the

observed identities and their actions, is constantly

being shaped by new interaction and presence events

and is used as the main driver in the behaviour of the

situated display. Identities that are currently present

are the major source for the generation of tags. The

presence of tags represents an additional dimension

that is not normally included in traditional tag

clouds, but may be key to introduce the situatedness

of a situated tag cloud. Even tag clouds with a large

time span may thus favour tags that are currently

present, albeit less popular, instead of popular tags

that no one is currently generating.

3.2 Token Data Model

The list of currently present identities constitutes a

dynamic keyword source that can be sampled

periodically to generate an evolving collection of

place keywords. The token data model is the result

of this sampling process in which keyword

snapshots are periodically being generated.

A keyword snapshot is made of a list of

keywords and their corresponding presence

intensity. The generation of this list can be tailored

using a presence level map that associates specific

command types with various weights in their

contribution to the intensity of presence. Because

situated tag clouds are generated as part of specific

interaction events and also because they can be

created to support specific forms of adaptation, they

can be optimised for different purposes and to

represent multiple dimensions of place. For

example, a music tag cloud may be created with

words and interactions related with music and be

used to support music recommendation for the place.

3.3 Tag Cloud Specification

The Tag cloud module supports the creation of tag

clouds that will be associated with a particular place.

Each tag in a tag cloud has four key parameters: A

name corresponding to the tag itself; a popularity

corresponding to the accumulated presence of that

tag as observed in the keyword snapshots within a

particular time scope; a presence level indicating the

current presence level for that tag; and a rank that

arranges the most popular keywords, the ones that

are actually making it into the tag cloud, in a number

of categories according to their popularity.

The creation of a new tag cloud is specified by

setting the key parameters that will determine the tag

cloud behaviour. These parameters may be arranged

in four major blocks: General, time dynamics, place-

making and visualization.

3.3.1 General Parameters

General parameters essentially serve to identify and

describe the tag cloud. They include a name that is a

unique identifier within the respective place. There

is also a title that describes the overall concept of the

tag cloud in terms of its source and dynamic

properties. The main purpose is to enable people to

interpret the tag and reason about it, and thus it

should be presented when the tag cloud is visualised.

Additional parameters include the number of tags to

be included, the rank Algorithm that defines the

model used for distributing the N most popular tags

among the ranks, the number of rank levels, and the

minimum word length.

3.3.2 Time Dynamics

Tag clouds are normally generated from a

reasonably static resource set, meaning that their

frequent update is not a major concern. With situated

tag clouds, time gains an increased importance

because the source for words is a continuous

sequence of presence and interaction events. Time is

therefore a key part of the underlying data model for

situated tag clouds. After some time, data becomes

less relevant and eventually it should be discarded.

A particular memory span defines for how long the

tag cloud will retain the sighted tags before the

information is discarded.

Time dynamics parameters specify the tag cloud

behaviour in relation with time and the deprecation

of the keyword observations. Managing time

involves defining a strategy for two main issues: the

minimum time scale for keyword aggregation and

the deprecation policy for discarding or reducing the

weight of older tag sightings. The combination of

these two policies will determine whether the system

is very reactive to the presence of new tags, or is a

more stable representation of place. They can be set

WEBIST 2011 - 7th International Conference on Web Information Systems and Technologies

298

through the following parameters: The Period Type

determines the time aggregation level from which

the tag cloud is calculated. The Number of Periods

defines the number of time periods of type Period

Type. Information deprecation is controlled by

setting the Period Decay parameter that

progressively reduces the weight of older periods in

the final aggregate information that is going to

constitute the tag cloud.

3.3.3 Place-making

The situated tag cloud must integrate specifications

that align its evolution with the place-making goals

of whoever installed the display and should support

an implicit negotiation between the place owner

perspective and the input from multiple visitors.

Place-making parameters allow a place owner to

provide additional characterization and specify

adaptation boundaries for a tag cloud. Even though

these specifications are not meant to determine the

final outcome of the tag cloud, they provide a way

for aligning the display behaviour with the general

expectations of appropriateness of the display owner

and its place-making objectives. The main part of

these specifications is a set of keyword lists that

enable some control of the tags in the tag cloud,

including a blocked words list or a list of seed

keywords that serve to initialise the tag cloud and

maintain a number of place keywords when there are

not enough keywords being generated. Seed tags are

defined with a minimum popularity value that

determines how visible they remain when other tags

begin to emerge. A Seeds Only parameter can be

used to determine that a particular tag cloud will

only accept seed words. This works as a white list

that restrains the accepted words to those on the list.

This closed tag cloud model may be useful for

thematic tag clouds, e.g. a tag cloud with sport

teams, emoticon symbols, or music styles. A tag

cloud based on a white list promotes aggregation

around particular tags. Finally, there is also a list of

contextual keywords that can be used to provide

additional context to the words in the tag cloud. For

example, Sports and Football could be added to a tag

cloud representing football teams. This is

particularly important if the tag cloud is to be used

for selecting web content.

3.3.4 Visualization

Visualization is an integral part of any tag cloud. In

its essence, the way a place-based tag cloud may be

presented is basically the same as for any other tag

cloud. In the basic format, tags are primarily sorted

alphabetically with the most important items being

shown with larger font sizes. The main difference is

that current presence is highlighted through the use

of colour. Another difference is the possibility, in

the case of closed tag clouds, of using images

instead of words, even if each image is directly

associated with a specific word.

4 EVALUATION

We have explored this concept of situated tag cloud

in two separate case-studies, which we will now

briefly describe.

4.1 Scheduling for Public Displays

In our first case study, we have used a generic tag

cloud to represent a dynamic and evolving view of a

place that could then be used as the key input to a

recommender system that would select information

feeds for presentation in a public display. The full

details of this case-study can be found in (Ribeiro

and José, 2010), but for the convenience of the

reader, we briefly summarise the key points of that

case study.

A context-aware scheduler considered both the

weight and current presence level of the represented

tags to select which content it would show next. The

tag cloud was specified by the place owner, but

place visitors could publish their own interests

through tag commands in their Bluetooth device

name. The combination of multiple contributions

from place visitors was then used for content

recommendation on the public display.

The evaluation of this system was based on a 3

weeks experiment in which the tag cloud was seeded

with 20 words representing topics related with

Informatics and Engineering as well as location

related keywords associated with the town and

region. During these 3 weeks we collected usage

logs and conducted a total of 15 structured

interviews with people who had previously tried to

use the system. The results obtained with this case

study suggest that this is a viable approach to the

problem of selecting relevant content for a dynamic

view of place. In particular, the visual nature of the

tag cloud seemed to facilitate the interpretation of

the system behaviour in a way that influences

positively the user perception, even when the

selections are not perfect. The results have also

shown that place visitors recognize the sensitivity of

the system to their demands and that a place tag

cloud can provide an important element for the

interpretation of place and for the combination of the

interests expressed by the place owner and the mul-

TAG CLOUDS FOR SITUATED INTERACTION AND PLACE PROFILING

299

tiple place visitors.

4.2 Group Expression at School

In our second case-study, we deployed a public

display in a school in which tag clouds were used to

represent the popularity of local communities, such

as students’ classes and teachers’ unit departments.

We had two separate tag clouds: one for students,

with 56 seed tags, and the other for teachers, with 12

seed tags. The seeds only property was set to true, so

that only the seed tags appeared in the display. The

popularity was calculated from the sightings of the

last two weeks, with the current week being counted

100%, and the previous one only 50%. The display

was divided in four panels, as shown in figure 1.

Figure 1: Group presence representation.

The left panel displays the Bluetooth device

names that are currently detected in the place. The

top right panel represents the students’ classes tag

cloud and the low right panel represents the

professors’ unit departments tag cloud. At the

bottom, instructions for using the tag command were

provided.

Although this experience was part of a broader

study lasting twenty-one weeks, this specific

experience was run for a period of four weeks.

During those four weeks, 313 different devices were

detected and 25 of them used the tag command,

which represents 7.98% of the total device detection.

In this period, there were 3226 sessions, in which

259 were produced by the devices that used the tag

command, which represents 8.02%. In terms of

individual tag cloud interaction, 23 devices in 44

session interacted with the students’ tag cloud and 2

devices in 215 sessions interacted with the

professors’ tag cloud.

This particular type of situated tag cloud seems

to have generated an interaction pattern that was

radically different from the previous case study.

After an initial period with no interactions, the first

interactions appeared and seemed to have sparred a

significant set of additional interactions. What we

have observed from these numbers and also from the

interviews was that this was a case in which early

adopters were hard to get. Only when someone

started pushing their group name, would others,

from the same or other groups, would follow, almost

as a reaction.

4.3 Analysis

The main objective of this work was to explore how

we can leverage on the tag cloud concept to support

situated interaction. One way to answer this question

is to consider to what extent our notion of tag cloud

is in fact similar to the traditional notion of tag

cloud, and to what extent existing tools for the

generation of situated keywords may still be used to

create situated tag clouds.

Clearly the underlying data model for generating

and storing words is very different, given the

substantial differences in the words generation

processes. However, independently of the data

model being used for managing the keywords

obtained from situated interaction, it would be trivial

to generate an equivalent text based containing those

same words, or at least the most popular ones, with

each word being represented as many times as any

internal frequency equivalent. This would allow

existing tools to take that document as their input

and generate the tag cloud, which means that at least

from the perspective of generating the visualizations

there is some potential for leveraging on existing

tools. There are however, some specificities that do

not map well into this model. In particular, presence

information represents an additional information

attribute that cannot be captured by tag frequency

alone and that is not considered by most tag cloud

tools. While many tag cloud generators support the

use of colours, for example, as an additional visual

element, this is only used for aesthetic purposes and

is not linked to any variable.

Regarding the case-studies, they demonstrate

various ways in which tag clouds may be used to

represent place and support situated interaction. In

the first case-study the tag cloud is continuously

reflecting the social setting around the display, being

sensitive to immediate indications of interest and

providing a balanced combination between the

content suggestions expressed by multiple place

visitors and those expressed by the place owner. In

the second case-study it represents the flow of

groups around the display setting.

WEBIST 2011 - 7th International Conference on Web Information Systems and Technologies

300

5 CONCLUSIONS

Tag clouds as proposed in this work exhibit very

interesting capabilities for supporting a balanced

combination of information filtering and information

retrieval. They support information filtering because

even if no one is using the system, the tag cloud is

already there and capable to serve as a content

generator. Moreover, the tag cloud specification

defines an adaptation scope that limits the extent of

the content that can be displayed. They support post-

filtering because once the display is in place, people

passing-by will sort out the content deemed more

interesting to them. Moreover, the tag cloud

provides an interesting representation of the interests

of a crowd. It avoids merely determining averages

that are not representative. It can also deal with the

tension between place adaptation, as something that

can be learned over time, and situatedness, as the

ability to react quickly to the social dynamics around

the display. Another very positive point is the way in

which the tag clouds can be visualised and enhance

the perception that people may have of the

adaptation processes going on.

We also found that peoples’ expectations may

not be aligned with the interaction concepts upon

which the model is based. While the place-based tag

cloud is essentially designed as crowd interaction

mechanism, and, moreover framed within the

concept of place, people often expect the system to

exhibit an immediate reaction to their specific

interaction. Moreover, the context and the semantics

of tagging in this context are ambiguous. When

someone advertises a tag that is collected by the

system at a particular place, what are they tagging?:

the place, themselves or that particular situation?

Perhaps people do not even think of themselves as

tagging, but rather as interacting with a system that

accepts words as input. In either case, peoples’

perception about these issues and the tagging

patterns that may emerge will necessarily have a

major impact on the viability of this approach.

5.1 Future Work

Further research is needed to evaluate across

multimultiple settings the ideal values for some of

the system parameters. For example the decay of

user-suggested tags affects responsiveness and also

the balance between pre-defined and emerging

notions of place, while the size of non-repetition

queues affects the balance between content quality

and diversity. Results suggest that this may be a

valuable step towards the emergence of dynamic

place profiles that match the social expectations and

practices of their evolving social settings. Following

on this idea, we also intend to explore how the

similarity between places can be inferred from the

similarity between the respective tag clouds.

ACKNOWLEDGEMENTS

Fernando Ribeiro was supported by a Portuguese

Foundation for Science and Technology scholarship

(SFRH/BD/31292/2006).

The research leading to these results has received

funding from FCT under the Carnegie Mellon -

Portugal agreement: Wesp (Web Security and

Privacy (Grant CMU-PT/SE/028/2008).

REFERENCES

Bielenberg, K. and Zacher, M. 2005, 'Groups in Social

Software:Utilizing Tagging to Integrate Individual

Contexts for Social Navigation', Master’s thesis thesis.

José, R., Otero, N., Izadi, S. and Harper, R., 2008. Instant

Places: Using Bluetooth for Situated Interaction in

Public Displays. IEEE Pervasive Computing, vol. 7,

no. 4, Oct.-Dec., pp. 52-7.

Lamantia, J., 2006. Tag Clouds Evolve: Understanding

Tag Clouds. Joe Lamantia.com Retrieved from:

http://www.joelamantia.com/ideas/tag-clouds-evolve-

understanding-tag-clouds, viewed September 22.

McNaught, C. and Lam, P., 2010. Using Wordle as a

supplementary research tool. The Qualitative Report.

vol. 15(3), no. 3, pp. 630-43, viewed September 2010,

<http://www.nova.edu/ssss/QR/QR15-

3/mcnaught.pdf>.

Mehta, C., 2006. Tagline Generator. Retrieved from:

http://chir.ag/projects/tagline, viewed Sept. 2010.

Pessemier, T. D., Deryckere, T. and Martens, L., 2009.

Context aware recommendations for user-generated

content on a social network site. 7th European

interactive television conf. Leuven, pp. 133-6.

Ribeiro, F. and José, R. 2010, 'Autonomous and Context-

Aware Scheduling for Public Displays Using Place-

Based Tag Clouds', in Springer-Verlag (ed.), Ambient

Intelligence and Future Trends-International Symp. on

Ambient Intelligence, vol. 72, pp. 131-8.

Steinbock, D., Pea, R. and Reeves, B., 2007. Wearable tag

clouds: visualizations to facilitate new collaborations

8th Int. Conf. on Computer supported collaborative

learning. New Brunswick, New Jersey, pp. 672-4.

Viegas, F. and Wattenberg, M., 2006. Communication-

Minded Visualization: A Call to Action. IBM Systems

Journal, vol. 45, no. 4, pp. 801 - 12.

TAG CLOUDS FOR SITUATED INTERACTION AND PLACE PROFILING

301