CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0

BY MEANS MULTIAGENT SYSTEMS

Knowledge Engineering Notes

Gonzalo A. Aranda-Corral

, Joaqu

ın Borrego-D

ıaz

and Jes

us Gir

aldez-Cru

Department of Information Technology, Universidad de Huelva, Palos de La Frontera, Spain

Department of Computer Science and Artiﬁcial Intelligence, Universidad de Sevilla, Sevilla, Spain

Artiﬁcial Intelligence Research Institute (IIIA-CSIC), Campus Universidad Aut

onoma de Barcelona, Barcelona, Spain

Keywords:

Mobile web 2.0, Formal concept analysis, Concept lattices, Tagging, Knowledge extraction.

Abstract:

Increasingly, users connect to the Internet by mobile devices and they are generating massive content through

them. The lead-off projects in Mobile Web 2.0 offer the opportunity to add semantics in order to obtain

structured knowledge. In this paper, we present speciﬁc challenges for tagging reasoning, into the SinNet

project. SinNet is based on user generated content (UGC) by mobile devices, as well as how to solve them by

means of combining multi-agent systems and formal concepts analysis.

1 INTRODUCTION

Mobile Web 2.0 (MW2.0) represents the new revolu-

tion in both social networks and digital convergence.

Despite this convergence -it seems to be the ”ulti-

mate” one- more work on knowledge interoperability

among heterogeneous devices and technologies (mo-

bile phones, Web 2.0 mashups, Semantic Web or In-

tranets) will be needed, however several difﬁculties

will arise. On the one hand, commercial, geopoliti-

cal and sociological interests have made the interop-

erability and knowledge organisation a challenge that

mobile companies are not considering in their new

projects. On the other hand, mobile phones have been

designed for more activities than many of the tradi-

tional Web 2.0 (W2.0) ones, and are often more re-

ﬂective.

Roughly speaking, Mobile Web provides to web

experiences ubiquity and mobility. These features

show signiﬁcant differences between MW2.0 and

Web 2.0, because users are able to generate content

with explicit (geo)spatial, temporal, contextual and/or

personal features. Some metadata associated with

them could be automatically generated while others

could be intentionally generated by users.

The marriage of social mobile networks and the

Semantic Web (SMW2.0), which is unique because

of its current features and mobile device limitations.

In a mobile computing context, these special features

and mobile device limitations can make the con-

vergence of MW2.0 and SW technologies (aka.

SMW2.0) difﬁcult. These may lead to a new digi-

tal divide between mobile networks and SW (or In-

tranets).

Unfortunately, MW2.0 has been developed on

special features networks which depict a completely

different framework from the neutral WWW. Some of

these features cause difﬁculties for MW2.0 and con-

tribute to the rising of a new digital divide, between

mobility and WWW computing, whereas other con-

ditions make hard the semantic interoperability:

1. Mobile Web is supported by private telecommu-

nication channels with different terms of use and

technological features, as opposed to the global

nature of conditions in Web 2.0.

2. Mobile Internet is not a neutral net. Network own-

ers can set limitations, or access conditions to se-

lected mobile services (e.g. VoIP services).

3. Mobile Telecomms companies can develop their

own ontologies for selling or distributing mobile

services, without the need to interoperate with

competitor’s ontologies.

4. Socio-economic aspects. In some countries, mo-

bile internet is expensive or the infrastructure has

not been deployed. Social networks based on pri-

mary channels (MMS) represent the main option.

5. Mobile user’s behaviour differs from desktop

176

A. Aranda-Corral G., Borrego-Díaz J. and Giráldez-Cru J..

CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0 BY MEANS MULTIAGENT SYSTEMS - Knowledge Engineering Notes.

DOI: 10.5220/0003736001760183

In Proceedings of the 4th International Conference on Agents and Artiﬁcial Intelligence (ICAART-2012), pages 176-183

ISBN: 978-989-8425-96-6

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

users. MW2.0 encompasses classic W2.0 tasks

with others related to diary needs that it may turn

into new mobile services.

6. Usability, hardware architectures, costs...

7. Mobile Communications provide robust com-

mercial infrastructures (subscriptions, rates, SM-

S/MMS channels) for designing feasible business

models for MW2.0 (Martignoni and Stanoevska-

Slabeva, 2007).

1.1 Social Network on SMS/MMS

Channel

SinNet platform, was built as an evolution of

Mowento (Aranda-Corral et al., 2009). Mowento was

designed for publishing testimonial documents about

events, witnessed by the user, at the time they are oc-

curring, and it is possible to use with any mobile de-

vice, even it it only has basic features. These events

are captured using mobile devices as photos and short

videos. Digital documents are sent to the Mowento

platform, via MMS, in order to be published onto the

WWW. An alert about this new event is also broad-

cast -via SMS- to the Mowento user’s neighbourhood,

in different ways and depending on the user’s ac-

count. Thus, the aim is to provide a social relation-

ship among mobile users -usually without Mobile In-

ternet access- by means of publication on Mowento’s

WWW site for photos and short videos, as well as

the micro-dissemination of these in the user’s social

neighborhood on Mowento’s mobile network (see ﬁg.

1). The choice of the MMS/SMS channel (justiﬁed

bellow) limits user management and participation and

it leads UGC management to a Multiagent Platform

(see ﬁg. 1). SinNet extends Mowento with new fea-

tures and with a new WWW portal with more classi-

cal Web 2.0 features.

1.2 Aim and Contribution of the Paper

The aim of this paper is to describe our original work

about how to increase the tagging task as a seman-

tic annotation method in a more hostile scenario than

above described, namely for non-sophisticated mo-

bile devices. We show the solutions we have de-

signed, as well as These techniques have been im-

plemented in a SMW2.0 experimental platform called

SinNet.The above restrictions have been considered

both, scientiﬁc and economic features for SinNet

project. E.g., The choice to use of MMS/SMS chan-

nel only, is based on it is neutrality and universal ac-

cessibility (solving restrictions (1),(2),(4) above), and

provides a pay channel, a basis for a business model

(feature (7)). Moreover, ontologies are based on

user’s tagging instead of company-owned ontologies

(bridging the gap considered in (3)) and, ﬁnally, Sin-

Net mobile application is designed for mobile devices

which satisfy minimal requirements and concrete user

needs (dissemination of testimonials on events) solv-

ing potential difﬁculties described in (5) and (6).

Why the MMS Channel? There exist strong rea-

sons for developing a platform by using the MM-

S/SMS channel, as SinNet is.Neutrality of the SM-

S/MMS channel is a key feature, because they are

revealed as universally accessible channels for mo-

bile phone users. In MW2.0, socio-economic as-

pects of networks must be particularly considered.

Although MW2.0 cannot be fully tested in devel-

oping countries, there exist opportunities for using

the MMS channel instead of Mobile Internet when

economic, social or geographic characteristics make

the use of Mobile Internet difﬁcult. Another im-

portant barrier is purely economic and local. If

we wish to ensure that your social network will

get a wider spread the relative high cost of Mo-

bile Internet services, e.g. in Spain, discourage

them as a primary channel. Recently, interesting

initiatives have exploited this channel to bridge the

mobile internet divide, such as Microsoft’s Oneapp

(http://www.microsoft.com/oneapp), a software ap-

plication that enables basic mobile phones access to

the W2.0 universe. In emergent countries such as

South Africa (where initially Oneapp was adopted)

or India, the impact of mobile based relationships on

socio-economic activities is stronger because these re-

lationships are new and often are not augmenting pre-

existing social networking. The choice of this chan-

nel also provides the project a robust business model

based on premium SMS/MMS.

The most important limitations of this choice are

the limited size of the document generated by the user

and the existence of so many mobile device hard-

ware/software architectures. However it more impor-

tant non-advanced mobile phones can also run the

SinNet mobile application. This allows us to work

with a great scope for the SinNet application, but the

design and usability of the Mobile application will be

a key issue. In fact, the SinNet mobile application is

available to every mobile phone with basic minimum

requirements (see Sect. 3).

Another requirement -observed from

experiments- is the need for a quick publication

when the user acts as a journalist for an ongoing

concrete event. Thus, the mobile application must

simplify the process to few clicks. The design of

SinNet’s Multiagent System (MAS), with speciﬁc

agent’s behaviours, is strongly limited by this re-

CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0 BY MEANS MULTIAGENT SYSTEMS - Knowledge

Engineering Notes

177

Figure 1: Mowento (precedent of SinNet) in action.

quirement. It is clear that, based on this scenario,

documents with deﬁcient tagging, will often be

generated. SinNet’s efforts are particularly addressed

to ﬁxing four main barriers: automatic enrichment

of poor tagging, tagging app for mobile phones that

allow basic tagging, by means of clicks, enhance-

ment for older tagging with new frequent tags and

knowledge conciliation among content generated by

different users.

The paper can be divided in two parts. The ﬁrst

part is devoted to describe the environment where the

project is realized and to justifying design decisions.

The second part details the Knowledge Engineering

behind the system, especially the MAS that supports

Knowledge representation and reasoning. The next

section is focused on describing the main character-

istics of tagging from mobile devices that we have

considered. In section 3, Formal Concept Analysis is

introduced as a key mathematical tool for managing

ontologies based on concepts, and how it is applied

by SinNet. Section 4 is dedicated to describe Sin-

Net project, its main components and the role of the

Multiagent System embedded into SinNet as a seman-

tic assistant for users. The four main semantic prob-

lems about Knowledge Organization Methods based

on FCA are described in Sect. 5, and how these have

been solved in SinNet in Sect. 6. The paper ends with

some considerations and future work.

2 TAGGING IN MOBILITY

In order to design tagging methods for SinNet users,

several considerations have been discussed. Tagging

is a universal method for annotating ﬁles, documents

and urls. Historically, it was the primary semantic

activity in the WWW, because it represented a so-

cial method for categorizing and classifying docu-

ments and allows users to navigate through document

databases. Web 2.0 Tagging is a task that different

Web sites consider in many different ways. In MW2.0

the main motivation is to collect and to share digital

objects. Therefore, User Generated Content (UGC)

from mobile devices has to be smartly organized. So-

ciological and psychological context of UGC in mo-

bility differs dramatically from Web 2.0 experience

through desktops. In fact, tagging from mobile de-

vices may be dangerously considered by users as a te-

dious task (particularly when user has a non-advanced

phone).

An interesting solution could be to design and to

implement an ontology for representing speciﬁc fea-

tures of the content that is generated, or transformed

by the social network. However, such ontology -that

allows users to tag digital objects generated from a

mobile application- should be a consensus ontology,

for both users and recommender systems. This must

be so because there is a real danger that ontologies de-

signed by engineers are not well understood by (non

specialist) mobile users. Thus, it seems more natu-

ral to provide a consensus ontology extracted from

users’ folksonomies, cognitively sound for the social

network. In these kind of ontologies, it does not

consider features associated with the user’s mobility

which can be automatically provided by the system/-

platform. Lastly, in SinNet project the tagging pro-

cess must be easy, cheap and quick, therefore usabil-

ity decisions are key aspects to consider.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

178

In order to conciliate tagging with ontologies, sev-

eral methods to integrate this kind of knowledge or-

ganization and SW realm have been considered. It

can be classiﬁed according to formal semantics as-

sociated to tag sets (or folksonomies). A ﬁrst class

are th methods based on ontological deﬁnition of tag-

ging (see (Kim et al., 2008)). A second one are the

methods based on transformation from folksonomies

to ontologies (Damme et al., 2007), included ontolo-

gies designed to deal with folksonomies as for exam-

ple (Gruber, 2007). Finally, methods based on con-

cept mining from folksonomies (J

aschke et al., 2008).

The latter is very useful for tag-based services

In order to use mobile phones for sharing content

in SMW2.0, it is necessary to accept that semantic

annotations of digital documents from mobile appli-

cations suffer several limitations. On the one hand, as

already commented, mobile phones are often tedious

tools for writing (some how) large content, speciﬁ-

cally in basic phone models. Therefore mobile appli-

cations should simplify the tagging task, producing an

automated set of tags, such as geo-location and tem-

poral metadata. On the other hand, a user appreciates

both the immediate generation of digital documents

on an event that he/she aims to report, and their fast

publication in their social network. Therefore, a care-

ful balance between sound annotation and usability is

necessary. All these considerations have inﬂuenced

tag management in the SinNet Platform.

Several difﬁculties for tagging through mobile de-

vices have been detected in SinNet:

Incomplete Tagging According to Concept Lattice.

User tags are not enough to locate the document into a

lattice concept. This tagging behaviour does not mean

a mistake, the user could not be sure of all of the tags

that can be applied to the document. Solution is de-

scribed in Sect. 5

Outdated Tagging Menu in Mobile Application.

The application should offer to users a menu broad

enough to obtain high quality tagging. Thus the appli-

cation has to avoid the user not completing all tagging

for the document. This could happen if the current

menu pages are not up-to-date (synchronized) with

the current lattice. The concept lattice is automati-

cally by agents updated when new documents or la-

bels came into the platform. In the same way, agents

send to the mobile devices new tagging menu pages.

This process will be described in Sect. 5.

Re-tagging with New Tags. Mobile application al-

low to users add new tags. They can induce user to

complete old tagged documents with the new one.

This routine is not appropriate for SMS/MMS chan-

nels. Therefore, an agent in the SinNet platform

should extend old tagging to new ones in an automat-

ic/autonomous way. A behaviour have been imple-

mented for SinNet’s agents to extend the tagging set

in a proper way.

Knowledge Conciliation. The use of personal con-

cept lattices could be an obstacle: Linking labels from

different users to navigate through SinNet global con-

tent can be hard. Although the initial lattice is com-

mon, introduction of new tagged documents and the

addition of new tags can cause semantic heterogene-

ity. Hence, concept lattices must be reconciled to

reach an agreement and to accept labels from other

users. SinNet implemented solution was presented in

(Aranda-Corral and Borrego-D

ıaz, 2010). It is useful

in any tagging-based service (see Sect. 5 ).

Above deﬁciencies should be ﬁxed by agents at

platform, due to the cost of SMS communications.

3 ARCHITECTURE OF SINNET

SinNet has three main components: the mobile appli-

cation, SinNet Web and the MAS which bridges both.

The SinNet mobile application lets users to pub-

lish audiovisual documents (photos and videos) forth-

with, so the user should ﬁnd it attractive, fast and easy

enough, to create and to send documents to the plat-

form. The selection ﬁle picking, that it could be a

short video or a photograph, with a current 300KB

limitation due to MMS technology. Any other kind of

ﬁles have not been considered for this project. A ba-

sic document tagging system and some ﬁelds, such as

title, small description, date, etc. Finally, the sending

of full audiovisual documents by MMS.

The current Mobile Application version uses a

system of menu pages which is built automatically by

means an algorithm which extracts the menu from the

concept lattice. The user navigates among them by

clicks, where each click represents a new tag (ﬁg. 3

(left)), or a set of them can be selected in the last menu

page (ﬁg. 3 (right)).

The mobile application allows users to capture im-

ages, or videos, by means of the installed camera on

the device. Documents (images or videos) are pro-

cessed (size, codiﬁcation, etc) and saved temporarily.

Also it allows the addition of some basic information,

such as title, short description, private,...). But the

main task is the tagging of the document. Finally, the

documents and all the information/metadata about it

is packed and sent in a MMS format. This application

is portable, based on Java J2ME, and it accomplishes

usability and speed.

The SinNet Web portal has been built on Elgg

(http://elgg.org/), an open source social networking

CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0 BY MEANS MULTIAGENT SYSTEMS - Knowledge

Engineering Notes

179

Figure 2: Formal context, associated concept lattice and Stem Basis.

platform which provides basic Web 2.0 features for

users.

Figure 3: Mobile Application, nonterminal (left) and termi-

nal (right) screens.

The SinNet Multi-Agent System offers mobile

services. Roughly speaking, MW2.0 services can be

created in two different ways: By Mobile extension

of an existing Web 2.0 service or by a pure MW2.0

service speciﬁcally dedicated to mobile networks and

based on user-generated content. New tools for this

kind of generated knowledge engineering and man-

agement have to be designed. In our case, the sec-

ond way is considered, by the choice of a MAS to

execute several semantic tasks. Each document -

which is received in SinNet (via Vodaphone’s Red

Box platform)- leads to the creation of an agent as-

sociated with this document which performs some of

the classical Web 2.0 user tasks as well as other spe-

ciﬁc (reasoning) tasks.

SinNet is a SMW2.0 which is supported by a

multiagent system platform installed in a public

Internet system. Agents, which are living in this

platform, behave as facilitators, moving the hard task

from the mobile device to the platform, and therefore,

making the users life easier. Although the possibility

of agents embedded into mobile devices exists, it

was discarded based on the decision that the SinNet

mobile application only will require minimal features

from user’s mobile phones. The customisation of

every task for each user and document, which is sent

to the Platform, is supported by the autonomy(of

computing) that a MAS can provide. Future scalabil-

ity will also be solved by means of controlling agent

population. The agent platform can be composed

by a group of servers and the agents can move over

them without difﬁculties. Another advantage is to be

able to extend the functionality system without a big

effort. Developers only have to add a new behaviour

to agents, as indeed has been done already.

SinNet’s platform is built on JADE

(http://jade.tilab.com), a FIPA compliant MAS

platform and tools for fast development. JADE could

offer us, in the future, the opportunity to open SinNet

to others platforms and interact with other social

networks.

The social organization model chosen for SinNet

agents’ is hierarchical. Now it is 2-levels deep,

but this could be increased. Also, we can separate

the agents into two types: generic and specialized.

Both of them live in the platform and they can move

through the containers (servers). Generic agents, also

called Planners, are responsible for all high-level

tasks and the platforms performance. They are the

core of all semantic information management. When

a document arrives, they review it and organize all

the tasks to be performed. Within a platform there

can be several planner agents to avoid bottlenecks

in this task, because this could block the entire

system. This Planner agent has a timer behaviour

(TickerBehaviour) where it checks if a new document

has arrived to the platform and, if so, pre-analyses

this content. Depending on the nature/format of this

content, a set of tasks should be applied on it. Some

of these tasks will be semantic ones, but not all of

them(e.g. format converting). For this, the planner

agent will create some specialized agents and assign

a task to each one. In this stage, the planner agent

also must avoid the bottlenecks distributing them on

all containers (servers) as a load balancer.

The main specialized agents, performing semantic

tasks, can be technically divided in two classes: One

Shot Behaviour agents and Dialogue agents. The

First ones runs its tasks in an autonomous way and

does not need any collaboration from any other agent.

Some of these are the tag suggestions agents, the

micro-dissemination agents, etc. Dialogue agents

need other agents to collaborate/compete and its

behaviours are mostly cyclical. When a specialized

agent ﬁnishes, it informs to its Planner and its running

is ﬁnished.

ICAART 2012 - International Conference on Agents and Artificial Intelligence

180

4 FORMAL CONCEPT ANALISYS

The convergence between Mobile Web 2.0 and Se-

mantic Web depends on the speciﬁc management of

ontologies. Ontologies and tags/folksonomies must

be conciliated in such kind of projects. A useful

bridge between these two kinds of representations

could be Formal Concept Analysis (Ganter and Wille,

1999). According to R. Wille, FCA mathematizes the

philosophical understanding of a concept as a unit of

thoughts composed of two parts: the extent and the

intent. The extent covers all objects belonging to this

concept, while the intent comprises of all common at-

tributes valid for all the objects under consideration.

It also allows the computation of concept hierarchies

from data tables. We succinctly present basic FCA

elements.

A formal context M = (O, A, I) consists of two

sets, O (objects) and A (attributes) and a relation

I ⊆ O × A. See Fig. 2 for an example of formal con-

text about live beings. The FCA main goal is the com-

putation of the concept lattice associated to the con-

text. Given X ⊆ O and Y ⊆ A it deﬁnes

:= {a ∈ A | oIa for all o ∈ X}

:= {o ∈ O | oIa for all a ∈ Y }

A (formal) concept is a pair (X, Y) such that X

= Y

and Y

= X. For example, concepts from living beings

formal context (Fig. 2, left) is depicted (right).

Using this Fig. 2, each node is a concept, and

its intension (or extension) can be formed by the

set of attributes (or objects) included along the path

to the top (or bottom). E.g. The node tagged

with the attribute Legs represents to the concept

({Legs, Mobility, NeedWater}, {Cat, Frog}).

Logical expressions in FCA are implications be-

tween attributes, a pair of sets of attributes, written as

→ Y

, which is true with respect to M. It says that

→ Y

holds in M (M |=Y

→ Y

) if for all o ∈ O, the

set {o}

respects Y

→ Y

, and it is said that Y

→ Y

is an implication of M. A set L of implications is a

(implication) basis for M if L is complete and non-

redundant. It can obtain a basis called Stem Basis

(SB) in a standard way (Ganter and Wille, 1999). A

SB for the formal context on live beings is provided

in Fig. 2 (right). It is important to remark that SB is

only an example of a implication basis. In the paper

any speciﬁc property of the SB can be used, so it can

be replaced by any basis.

Armstrong rules (Armstrong, 1974) provides im-

plicational reasoning. By deﬁning `

as the proof re-

lation by these rules, it holds that the bases are `

complete: Let L be an implication basis for M, and L

an implication. Then M |= L if and only if L `

In order to work with formal contexts,

stem basis and association rules, the Conexp

(http://sourceforge.net/projects/conexp/) software

has been selected. It is used as a library to build

the module which provides the implications (and

association rules) to the reasoning module of our

system.

Every implication has also associated some prop-

erties, e.g. support. Support is deﬁned as the number

of objects that contain all attributes from Y

and holds

the implication. Based on this property, a variant of

implicational basis is deﬁned, called Stem Kernel ba-

sis, which it will be used in 5 . A Kernel Stem basis

is the subset of implications with support greater than

zero of a Stem basis.

In case of SinNet, FCA is used as an ontologi-

cal representation of tags. Rougly speaking, ontology

class X, in SinNet’s ontology, is intended as the class

of digital objects with tags X. In fact, the ontology is

extracted from several experiments with our mobile

application and an arbitrary set of tags. Attribute ex-

ploration method (Ganter and Wille, 1999) is used to

reﬁne this ontology, as well as, to obtain a system for

supplying deﬁciencies in tagging, as seen below.

FCA is the Knowledge Engineering technology

that SinNet uses for knowledge mining and reason-

ing. FCA is a useful tool for computing and reason-

ing on folksonomies (J

aschke et al., 2008) that allows

one to ﬁnd and to work with conceptual ontologies

behind tagging. Associated SinNet tasks to FCA are

classiﬁed in two ways. On the one hand, it intensively

uses the computing of concept lattices and stem ba-

sis. On the other hand, new FCA methods have been

designed for solving SinNet speciﬁc features. These

methods have been proved to be useful for facilitat-

ing intelligent tagging reasoning and to allow the sys-

tem to delegate several tasks for the MAS allocated

into SinNet. In this way, SinNet removes several user

imperatives and minimizes their interaction with the

system, improving their knowledge organization.

Currently, SinNet runs only for reporting events

such as conferences, workshops or meetings. The

concept’s lattice that SinNet uses as its initial ontol-

ogy was obtained from an experiment with a set of

200 tagged photos. In Fig 4 part of concept lattice is

depicted. The concept which contains these attributes

is the most speciﬁc one.

5 SOLVING DEFICIENCIES

The problems described above were related to deﬁ-

cient tagging from mobile phones and for synthesiz-

ing sound metadata in platform, and they have been

CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0 BY MEANS MULTIAGENT SYSTEMS - Knowledge

Engineering Notes

181

Figure 4: Example of concept lattice within SinNet (left) and thumbnail with tags and suggested tags.

solved in SinNet by means of the agents.

The solution for the incomplete tagging prob-

lem was shown in (Aranda-Corral et al., 2009). Rec-

ommender agents are responsible for completing the

document’s tagging using a speciﬁc behaviour that it

was implemented to reason with a SB associated to its

concept lattice (computed with ConExp) To obtain the

SB, Conexp tool is used. The basis is transformed into

a production system executable by the agent. Roughly

speaking, it changes the question ”Every object that

has the attribute X

, . . . , X

has also the attribute Y?”

by ”Every document with tags X

, . . . , X

should have

the tag Y?”.

Actually we are interested in the “suggesting tags”

task, which is carried out by an agent with a delibera-

tive (rule-based) behaviour. The agent loads the rules

from the production system (from SB) created above

and initializes the rule’s engine with the facts (tags)

written by the user and read from database. It is exe-

cuted to obtain a new set of facts (tags), which will be

suggested to the users, through the SinNet web inter-

face.The soundness of the reasonig process is based

on the following result: Iet S be the SB for the context

M, A

, . . . , A

and A attributes. The following condi-

tions are equivalent:

1. S ∪ {A

, . . . A

} `

A (where `

is the deduction

associated to the production system).

2. S `

, . . . A

→ A

3. M |= {A

, . . . A

} → A.

That is, if o a document is tagged with

, . . . , A

} and the production system entails the tag

A, then {A

, . . . A

} ∪ A is a more speciﬁc attribute set

for the document. In Fig. 4 (note that tags appear

in Spanish) it shows that if a document is tagged with

{conference, speaker, talk}, agent will suggest the tag

participant”.

Dynamic creation of pages behaviour is designed

for enhancing tagging tasks from the SinNet mobile

application. An agent’s behaviour was developed that

helps users to choose the most adequate tag set to de-

scribe the appropriate concept. This method of assis-

tance consists of the creation of a sound menu pagina-

tion system with a controlled depth. SinNet empirical

studies obtained that users usually click up to 3 times

in a tagging task, with the current concept lattice. So,

we need to create a good selection for the tag menu.

When users want to annotate content, the SinNet

mobile application shows the tag-list pages to select

a tag. Only one selection per page except for the ﬁ-

nal page. The last page is a multiple selection tag-list

with, even, the option to add new tags. The paging

system aim is that, by each click that users make, they

move through the concept lattice and the concept re-

lated to the document is reﬁned.

To build the menu paging, ﬁrst a compatible span-

ning tree is built. This spanning tree holds the sub-

concept relation and the root node is related to the

most general concept (Top). Next, for each spanning

tree node a tag is selected, preserving the set com-

posed by this tag and all super-node tags, pointing to

the most general concept which contains tags and to

which the document is related. To select the tags in

each tree node, agent uses an algorithm based on the

sound selection of a concept’s own attribute (attribute

not contained by any other of its super-concepts)

The concrete menu page system obtained will be

sent to the user’s mobile phone in XML format. The

user can interrupt navigation and ﬁnalise the tag se-

lection, but the last spanning tree node visited gives

us the concept related to that document. If the last

page is reached, a multiple selection page lets us clas-

sify the document into a concept. Even it can add new

tags which causes the creation of new concepts and,

therefore, the lattice evolution.

Also it may be necessary to update old-tagged

documents. Experiments with SinNet shows a fre-

quent situation: the user decides to introduce new

ICAART 2012 - International Conference on Agents and Artificial Intelligence

182

tags for new documents. These new tags are in-

cluded in its own concept lattice, but it may be pos-

sible old documents should be re-tagged with the

new tag. The re-tagging process would be carried

out by means of the classical completion procedure

and implemented as an agent behaviour. The formal

result that supports this behaviour was presented in

(Alonso-Jim

enez et al., 2008) (deﬁnability in formal

contexts) in FCA. The new tag will appear as a sug-

gested tag in the user’s older documents preserving

old SB. Thus it completes older tagging by using the

new tag and also preserving the original Stem basis.

Lastly, Users’s Knowledge Conciliation aims to

exploit an important beneﬁt of the Web 2.0, the

sharing of information and knowledge among users.

A potential threat is that semantic techniques are

adapted to each user. Over time, user’s knowledge

can vary a great deal, and this difference could cre-

ate knowledge incompatibility issues. This is because

ontologies evolve with the each user, and not in a

joint manner. In order to navigate through the set

of tags and documents from different users, SinNet

has delegated to agents to make these different con-

ceptualizations compatible. The agent-based concili-

ation algorithm was presented in (Aranda-Corral and

Borrego-D

ıaz, 2010). This algorithm’s foundations

are on the basis that conceptual structure associated to

tags gives more information on the user’s tagging (as

in (J

aschke et al., 2008)). Once ﬁnished the process,

the agents will get a common context. So they can

extract new concepts and suggestions from a common

context, and therefore, ashared conceptualization.

6 CONCLUSIONS

SinNet is a MW2.0 platform that uses the MMS/SMS

channel for social relationship (WWW platform is

not the user main reference), in its pure mobile ver-

sion, and thus it may avoid the use of Mobile In-

ternet. Most important semantic/reasoning activities

carry out from user’s mobile devices to a MAS. Cog-

nitive tasks performance by agents represented an ef-

fort of agent’s behaviour design.

As MW2.0 project, SinNet seems similar to

SeeMeTV (launched by the mobile network oper-

ator Hutchison 3G UK) using MMS channel for

sharing content. SinNet also shares some fea-

tures with the mobile extension of German video-

sharing platform Clipﬁsh (http://www.clipﬁsh.com)

called ”Handy Clipﬁsh”, as a stand-alone mobile por-

tal. However, SinNet’s basic idea differs from both

ones. Authors do not know of any MW2.0 platforms

based on MMS/SMS channel and semantic features

implemented. SinNet’s tag reasoning is unique. To

the best of our knowledge, no other MW2.0, inten-

sively using FCA tools, exists.

ACKNOWLEDGEMENTS

Supported by TIN2009-09492 Spanish Ministry of

Science and Innovation project, co-ﬁnanced with

FEDER founds. Technological infrastructure was

provided by Minerva project co-ﬁnanced by Junta de

Andaluc

ıa and Vodafone.

REFERENCES

Alonso-Jim

enez, J. A., Aranda-Corral, G. A., Borrego-

ıaz, J., Fern

andez-Lebr

on, M. M., and Hidalgo-

Doblado, M. J. (2008). Extending attribute explo-

ration by means of boolean derivatives. In Proc.

6th. Conf. Concept Lattices and Their Applications

(CLA2008), pages 121–132.

Aranda-Corral, G. A. and Borrego-D

ıaz, J. (2010). Rec-

onciling knowledge in social tagging web services.

In Proc. 5th Int. Conf. on Hybrid AI Systems (HAIS

2010), volume 6077 of Lecture Notes in Computer

Science, pages 383–390.

Aranda-Corral, G. A., Borrego-D

ıaz, J., and G

omez-Mar

ın,

F. (2009). Toward semantic mobile web 2.0 through

multiagent systems. In Proc. 3th KES International

Symposium on Agent and Multi-Agent Systems, Lec-

ture Notes in Computer Science, pages 400–409.

Armstrong, W. W. (1974). Dependency structures of data

base relationships. In IFIP Congress, pages 580–583.

Damme, C. V., Hepp, M., and Siorpaes, K. (2007). Folk-

sontology: An integrated approach for turning folk-

sonomies into ontologies. In Proc. ESWC Workshop

“Bridging the Gap between Semantic Web and Web

2.0” (SemNet 2007), pages 57–70.

Ganter, B. and Wille, R. (1999). Formal Concept Analysis:

Mathematical Foundations. Springer.

Gruber, T. (2007). Ontology of folksonomy: A mash-up of

apples and oranges. Int’l Journal on Semantic Web &

Information Systems, 3(2).

aschke, R., Hotho, A., Schmitz, C., Ganter, B., and Stum,

G. (2008). Discovering shared conceptualizations in

folksonomies. Journal Web Semant., 6:38–53.

Kim, H. L., Scerri, S., Breslin, J. G., Decker, S., and Kim,

H. G. (2008). The state of the art in tag ontologies:

a semantic model for tagging and folksonomies. In

Proc. Int. Conf. Dublin Core and Metadata Applica-

tions, pages 128–137.

Martignoni, R. and Stanoevska-Slabeva, K. (2007). Mobile

web 2.0. In Proc. 20th Bled eConference, pages 1–18.

CONCEPTUAL-BASED REASONING IN MOBILE WEB 2.0 BY MEANS MULTIAGENT SYSTEMS - Knowledge

Engineering Notes

183