Social Semantic Web Fosters Idea Brainstorming

Abstract

Generating and identifying promising ideas represent important challenges for any Enterprise that is focused on knowledge-intensive activities. The generation of new ideas, especially high-quality creative ideas, is vital to business success. Brainstorming is a didactic method that can be exploited to sustain the development of high order skills considered fundamental to foster innovation. On the other side, brainstorming sessions produce new ideas that have to be evaluated and possibly selected. In this paper the Social Semantic Web is exploited in order to de ne an approach for brainstorming that overcomes the limitations of the existing systems supporting groups in generating ideas. The Semantic Web-based structures organize, correlate and simplify the search for user-generated contents (e.g. ideas). Meanwhile, user-generated contents are analysed in order to elicit non-asserted correlations between them that are used to enrich the aforementioned structures.

Keywords

• Social Semantic Web
• Brainstorming
• SIOC
• Knowledge Forum
• Knowledge Extraction
• Idea Generation
• Idea Selection
• Innovation

-

Generating and identifying promising ideas represent recurrent and critical challenges for any Enterprise that is focused on knowledge-intensive activities and innovation. The generation of new ideas, especially high-quality creative ideas, is vital to business success. In order to foster the idea-related processes new strategies and environments to develop High Order Thinking skills (HOT skills) have to be re-thought. Critical thinking, re ection, problem-solving, etc. are fundamental skills for maintaining and improving innovation processes [ 9 ]. The research activities on Technology Enhanced Education (TEE), and in particular on Workplace Learning, point on e-Brainstorming as a didactic method guiding a learners' group to learn by progressive argumentation and idea development. At the same time, e-Brainstorming allows developing and improving the thinking skills by exporting the identi ed promising ideas in order to further investigate them together with other groups to achieve a solid result in terms of feasibility and originality of the selected ideas. Moreover, e-Brainstorming allows to overcome the production blocking and conformity e ect in teamwork [ 5 ], by doing so it improves comparison, negotiation and decision-making processes. Some consideration have to be expressed: { The numerous existing Group Support Systems (GSSs) developed in order to assist people during the idea generation process are based on a vision known as Osborn's conjecture: if people generate more ideas, then they will produce more good ideas. Hence, these systems do not take care of the process transforming the quantity into quality with respect to the generation of ideas [ 14 ]. { The need for overcoming the limited vision of GSSs has conducted to the Bounded Ideation Theory [ 3 ] stating that an e ective brainstorming model must sustain an iterative process that involves two mains strategies: idea exchange (sharing ideas within a brainstorming group) and generation (accumulating numerous ideas) at the social level and idea expansion (building new ideas starting from existing ones) and selection (identifying of most promising ideas) at the cognitive distributed level [ 18 ]. { Despite the brainstorming literature has agreed to support the discovery of connections among di erent ideas can be signi cant to e ectively support the steps from idea generation (divergent thinking) to idea selection (convergent thinking), there exist few systems that support the automatic discovery of the aforementioned connections [ 11 ].

The present work proposes a Brainstorming Model, based on the Social Semantic Web approach, that takes care of the Bounded Ideation Theory to overcome the Osborn's conjecture. The used Semantic Web-based structures allow tool interoperability and simplify query and inference operations. On the other hand, the Brainstorming Model is based on the most common asynchronous communication/collaboration tool of the Social Web: the Discussion Forum. A languageindependent keyphrase extraction algoritm is also applied to support correlation discovery between ideas coming from di erent groups. The work is organized as follows. In the Section 2 the Brainstorming Model is de ned on the basis of the Knowledge Forum Model by extending Semantic Web-based ontologies. Furthermore, in Section 3.1 an approach, based on a keyphrase extraction algorithm, to automatically discover correlations between ideas coming from more than one brainstorming sessions is illustrated. In Section 4 some conclusion is provided. 2

Extending SIOC for Brainstorming

In this section a Brainstorming Model is de ned. The approach proposed in the present paper is to exploit the Knowledge Forum in order to provide a suitable brainstorming environment. Morever, the de ned Brainstorming Model will be described by extending SIOC (Semantically-Interlinked Online Communities) [ 2 ]. SIOC is an attempt to link online community sites, to use Semantic Web technologies to describe the information that communities have about their structure and contents, and to nd related information and new connections between content items and other community objects. SIOC is based around the use of machine-readable information provided by these sites. The adoption of SIOC provides the following bene ts: { fostering interoperability among di erent tools (also of di erent typologies like wikis, blogs, instant messaging, etc.); { simplifying the link with external data sets, vocabularies, thesauri, folksonomies and with other Semantic Web-based schemes; { improving and making cheaper the reuse of user-generated content; { providing a semantic layer to be queried and inferred by using standard languages (SPARQL4, OWL/OWL2[ 10 ]) and reasoners. 2.1

Brainstorming Model De nition

The brainstorming is a problem-solving technique de ned by Osborn [ 12 ] based on a group discussion led by a moderator. The purpose of a brainstorming session is to make possible the growth of the biggest possible number of ideas about a speci c issue. The brainstorming technique is also considered a relevant didactic method. In fact, it can be also classi ed as an argumentative practice [ 1 ]. A strong point of brainstorming is the ability to use the suggestions provided by all participants in the group, so that an idea proposed by a group member can suggest to another a new idea, perhaps more appropriate to reach the best solution. The focus, in the rst phase is to produce the greatest number of ideas, which is initially more important than their quality, especially because the greater the number of ideas, the greater the likelyhood of nding some useful. In a second step, which is the more challenging phase of a brainstorming session, ideas should be evaluated, in relation to their e ectiveness, selected and developed further. In the proposed approach, a brainstorming session pre gures the presence of a moderator while the other participants have no speci c roles. The topic of discussion has to be not completely de ned in order to unleash the power of idea generation, the ideas have to be freely expressed in the initial phase given that quantity is more important than quality at this stage. So, according to our model the brainstorming session consists of three di erent phases: { Activation. In this phase the issue, on which the discussion has to take place, is presented and the participants have the possibility to socialize. { Production. In this phase the moderator asks participants to speak freely on the subject, urges them to be active, asks questions, rewords questions. The participants freely express ideas, thoughts, opinions. Ideas are not subject to criticism during the meeting, in fact the adverse judgement of ideas must be withheld until later (deferring judgement [ 18 ]).

4 http://www.w3.org/TR/rdf-sparql-query/

{ Synthesis. The moderator summarizes the generated ideas, uses various criteria to stimulate participants to assess and select the best ideas. At this stage combinations and improvements of ideas are seeked. In addition, participants should suggest how the ideas of others can be turned into better ideas or how two ideas can be merged into new ones.

In order to de ne a digital environment able to support Brainstorming sessions as we have de ned them above, the Knowledge Forum [ 4 ] can be exploited to support the creation and the continuous improvement of knowledge. To facilitate discussion, and therefore the transparency of the communicative intention of each author, the Knowledge Forum provides some prede ned linguistic structures called sca olds, through which it is possible to identify a set of descriptors of thought (thinking types ), e.g. my theory, need to understand and so on.

In our model the use of three di erent sca olds is proposed in order to sustain the main phases of a brainstorming session: Idea Generation, Knowledge Construction and Revision Circle. The rst one covers the Activation and the Production phases of the Brainstorming session. While, the second one and the third one cover the Synthesis phase. Figure 1 shows the list of the Thinking Types for each considered sca old. The SIOC initiative aims to enable the integration of online-community information. For instance, users create posts (sioc:Post) organized in forums (sioc:Forum), which are hosted on sites (sioc:Site). These concepts are subclasses of higher-level concepts that were added to SIOC: sioc:Item, sioc: Container and (sioc:Space. The sioc:has reply property links reply posts to the content to which they are replying, the sioc:has creator property links user-generated content to its authors, and the sioc:topic property points to a resource describing the topic of content items. The SIOC Type module introduces new sub-classes for describing di erent kinds of Social Web objects in SIOC. In addition, the module points to existing ontologies suitable for describing details on these objects. For instance, a sioc t:ReviewArea might contain reviews asserted by using Review RDF 5 that is a domain speci c vocabulary used to describe the main properties of a review. The most important classes are rev:Review, rev:Feedback and rev:Comment, while the important properties are createdOn, hasReview,rating and reviewer. The link between an instance of a sioc:Post and a review (an instance of the rev:Review class) is realized by the property rev:hasReview (rdfs:Resource as range and rev:Review as domain). The ReviewRDF scheme is important for the BrainSIOC in order to handle ratings on ideas during the last phase of a brainstorming session (i.e. Synthesis) when the most promising ideas are evaluated, selected and packaged (described more formally).

SIOC can be used in combination with other Semantic Web-based schemes. First of all, SCOT (Social Semantic Cloud of Tags) [ 7 ] can be used to model tagging operations. SCOT aims to describe the structure and the semantics of tagging data and to o er social interoperability for sharing and reusing tag data and representing social relations amongst individuals across di erent sources. The scot:Tag class is used to manage tags. SCOT also enables the modeling of some aspects regarding who uses a speci c tag. In fact, the property scot:usedBy links a tag to a speci c user. An instance of sioc:Post can be tagged by using the scot:hasTag property, or conversely by using the scot:tagOf property with domain scot:Tag and range sioc:Item (a subclass of sioc:Item). SCOT can be also integrated with the MOAT (Meaning Of A Tag) 6 ontology that provides a mechanism to enrich data regarding tags by considering their meaning. Tagging ontologies are particularly useful in the context of BrainSIOC because they improve ndability of ideas across brainstorming sessions. Moreover tagging ontologies allow to simply correlate ideas with any kind of user-generated content. The SIOC ontology follows this practice by reusing the FOAF vocabulary 7 to describe person-centric data. A person (described by foaf:Person) will usually have a number of online accounts (sioc:UserAccount that is a sub-class of foaf:OnlineAccount) on di erent online-community sites. FOAF allows to model a social network where persons' pro les are linked together by using the foaf:knows property between two instances of foaf:Person class. In the end, SKOS (Simple Knowledge Organization System)8 is a Semantic Web

5 http://vocab.org/review/terms.html

6 http://moat-project.org/

7 http://www.foaf-project.org/

8 http://www.w3.org/TR/ skos- primer/

scheme used to build taxonomies and controlled vocabularies. For the aim of this work, SKOS will be used to model a controlled vocabulary of contexts of interest in a given organization using skos:narrower and skos:broader properties to relate instances of skos:Concept. SKOS can be used in order to construct controlled vocabularies and taxonomies for topics in SIOC to be linked to instances of sioc:Post or sioc:Item by means of the sioc:topic property. SKOS can improve knowledge sharing and correlation processes across di erent collaboration/communication sessions and tools. By linking FOAF, SIOC, SCOT/MOAT and SKOS it is possible to enrich a person's (a worker in the Enterprise context) pro le with the generated ideas, the used tags, etc. in order to foster people search operations. 2.3

The BrainSIOC ontology

The BrainSIOC ontology extends the SIOC ontology to support the brainstorming sessions described in Section 2.1 and sca olds and thinking types illustrated in Figure 1. In order to de ne the aforementioned extension, several schemes have been considered. In particular, the attention has been focused on Argumentative Discussion schemes [ 17 ]. Among the others, IBIS OWL and DILIGENT are relevant for the aims of this work. The IBIS OWL Model is a RDF representation of IBIS, providing URIs for terms regarding argumentations. DILIGENT is primarily a methodology for engineering an ontology; the acronym comes from certain letters in the phrase DIstributed, Looselycontrolled and evolvInG. Other interesting works are Idea Ontology [ 15 ] and SWAN/SIOC [ 17 ]. The rst one introduces an ontology to represent ideas. This ontology provides a common language to foster interoperability between tools and to support the idea life cycle. Through the use of this ontology additional bene ts like semantic reasoning and automatic analysis become available. With respect to the aforementioned work, BrainSIOC does not cover the whole idea life cycle management but it proposes a model to represent and support the activities in the context of brainstorming sessions by exploiting a modelling approach similar to those presented in [ 15 ]. The second one is a domain-dependent scheme modelling scienti c discourses using Semantic Web-based approaches.

First of all, the BrainSIOC ontology considers two roles for the brainstorming activity, i.e. the generic participant and the moderator. In order to model the rst one we need to de ne the bsioc:Participant class as a subclass of sioc:Role. While the class bsioc:Moderator is de ned by subclassing bsioc:Moderator. An instance of sioc:UserAccount is linked to a speci c role by using the sioc:funcion of property (its inverse is sioc:has function). The link between a moderator and a speci c container (e.g. a forum) can be also asserted by using the sioc:has moderator property with domain sioc:Forum and range sioc:UserAccount. Furthermore a brainstorming session is modelled by subclassing the sioc:Forum class and de ning the bsioc:Brainstorming in order to reuse all the properties de ned for sioc:Forum. Figure 2 provides the list of the other classes de ned in the BrainSIOC ontology (bsioc namespace). In particular, there are correspondences between BrainSIOC classes and both IBIS

Fig. 2. Classes of the BrainSIOC ontology.

OWL and DILIGENT: sioc t:Question is related to IBIS ibis:Question, bsioc:Evaluation is related to DILIGENT Evaluation, bsioc:Example is related to DILIGENT Example, bsioc:Decision is related to IBIS ibis:Decision, bsioc:Idea is related to IBIS Idea. Furthermore, we need to de ne new properties to be added to the BrainSIOC ontology. In SIOC, there exist several properties that are useful to link instances of sioc:Item (and hence of sioc:Post) to each other. In particular, the has reply property is used to relate two items, while the sioc:reply of property is its inverse. Both the aforementioned properties are de ned as sub-properties of sioc:related to that is adopted in the BrainSIOC. Another useful property is sioc:next version that can be used to link two di erent versions of the same item. In the end, the sioc:content property (with domain sioc:Item and range rdfs:Literal) is used to store the text representing ideas, questions, answers and so on. Figure 3 illustrates an instance of the BrainSIOC ontology that shows the generation of some ideas in response to a proposed issue. The example illustrates how to the brainstorming takes place across several threads and how ideas evolve step by step until becoming a packaged idea or aborting. 3

Knowledge Discovery in Brainstorming Sessions

In this section, two knowledge discovery modalities in brainstorming sessions are described. The rst one deals with discovering correlated ideas across brainstorming sessions. The second one concerns with the capability of BrainSIOC, being based on the Semantic Web stack, to provide high interoperability among people and applications while accessing, retrieving and sharing knowledge in standard way. Figure 4 shows both the modalities also explained in 3.1 and 3.2.

In Fig. 4, four brainstorming sessions are considered. For each session there is a group of participants taking part in the brainstorming. The sessions are disjoint except for the Knowledge Construction phase where correlations among ideas are discovered (see section 3.1) in order to unlock the independent sessions by providing external stimuli represented by similar ideas coming from other sessions. 3.1

Discovery of correlations among Ideas

In order to satisfy the requirement described in Section 1 regarding the need for correlating ideas, an approach to discover similar ideas across multiple brainstorming sessions (and to suggest these correlations to the participants) is proposed. During the Knowledge Construction phase, for a given idea A (an instance of the bsioc:Idea class), the literal associated with the sioc:content is compared with other ideas coming from other brainstorming sessions. The ideas A1, A2, ... , An more similar to A are suggested to the participants of the brainstorming sessions where A is emerged (in Figure 4, X1 and X2 are similar, so they are respectively suggested to sessions 4 and 3). The proposed approach is based on the application of the DegExt algorithm to build a graph representation of a single idea. In order to calculate the similarity, a distance measure that computes the distance between graphs is exploited. A threshold passing value must be considered in order to select only the most similar idea couples. Furthermore, we suggest to rank the idea couples that pass the threshold using a measure of diversity between the two idea proposers. The bigger the diversity value, the greater the rank value. This approach is supported by scienti c and methodological approaches concerning the team building approaches. In particular, in [ 13 ] and [ 6 ] it is emphasized that highly heterogeneous workgroups (diversity of competencies, skills, knowledge, culture, etc.) are more performant and e ective with respect to the idea generation tasks. The diversity measure can be calculated by using the FOAF pro les of the idea proposers and applying some distance measure. The correlations, that are automatically elicited and accepted by participants after a discussion, can be asserted through the use of the new re exive property bsioc:correlated to that is de ned by subclassing the sioc:related to property. DegExt [ 8 ] is an unsupervised, graph-based, crosslingual word and keyphrase extractor. DegExt uses graph representation based on the simple graph-based syntactic representation of text, which enhances the traditional vector-space model by taking into account some structural content features. The simple graph representation provides unlabeled edges representing order-relationship between the words represented by nodes. The stemming and stopword removal operations of basic text preprocessing are executed before constructing the graph. A single vertex is created for each distinct word, even if the word appears more than once in the text. Thus, each vertex label in the graph is unique. Edges represent order-relationships between two terms: there is a directed edge from A to B if an A term immediately precedes a B term in any sentence of the document. The syntactic graph-based representations were shown by Schenker et al. [ 16 ] to perform better than the classical vector-space model on several clustering and classi cation tasks. The most connected nodes in a document graph are assumed by DegExt to represent the keywords. When document representation is complete, every node is ranked by the extent of its connectedness with the other nodes, and the top ranked nodes are then extracted. Intuitively, the most connected nodes represent the most salient words. According to the above representation, words that appear in many sentences that diverge contextually will be represented by strongly connected nodes. DegExt is convenient for the aim of our work because it is relatively cheap in terms of processing time (linear computational complexity) and memory resources while providing nearly the best results for the two above text mining tasks and it does not require training. In order to exploit the result of the DegExt algorithm a distance measure between graphs has to be adopted. In particular, the measure proposed in [ 16 ] is considered: distMCS(G1; G2) = 1

mcs(G1; G2) max(jG1j ; jG2j) (1) where G1 and G2 are graphs representing ideas (constructed by using DegExt algorithm applied on the sioc:content property of instances of the bsioc:Idea class), mcs(G1; G2) is their maximum common subgraph, max(:::) is the standard numerical maximum operation, and j:::j denotes the size of the graph that can be taken as the number of nodes and edges contained in the graph. The computation of mcs can be accomplished in polynomial time due to the existence of unique node labels in the considered application. The proposed method provides more accuracy with respect to traditional methods based on numerical feature vectors because it considers the order in which terms appear, where in the document the terms appear, how close the terms are to each other, etc. 3.2

Querying on BrainSIOC

In order to demonstrate the e ectiveness of the Semantic Web stack to model, represent and integrate data, a simple SPARQL query able to nd, across all brainstorming sessions, all packaged ideas annotated with the tag "Social Web"is listed here. select ?title, ?content, ?topic where { } ?s a bsioc:PackagedIdea. optional { ?s dc:title ?title }. ?s sioc:content ?content . optional { ?s sioc:topic ?topic .

?topic rdf:type skos:Concept .

?topic skos:prefLabel "Social Web" }

In particular, the above query foresees the use of the Dublin Core9 property namely dc:title and the use of SKOS to de ne a shared (across all brainstorming sessions) controlled vocabulary in order to tag the posts. Moreover, this simple query envisages the capability of BrainSIOC to enable the integration of brainstorming sessions with collaborative working and learning scenarios in order to foster and improve knowledge maturing and knowledge sharing processes within the Organizations. 4

Conclusions and Future Works

This work proposes an approach consisting in i ) a novel Brainstorming Model implemented by extending the SIOC ontology and de ning BrainSIOC, ii ) a technique based on the application of the DegExt algorithm to automatically discover correlations among ideas across multiple brainstorming sessions. The approach will be experimented and exploited in the ARISTOTELE project (which also foresees the development of a tool implementing the BrainSIOC) by also considering the competencies that may be developed by the participants during brainstorming sessions.

Acknowledgement

This research is partially supported by the EC under the Project ARISTOTELE "Personalised Learning & Collaborative Working Environments Fostering Social Creativity and Innovations Inside the Organisations", VII FP, Theme ICT2009.4.2 (Technology-Enhanced Learning), Grant Agreement n. 257886.

9 http://dublincore.org

References

1. Bonaiuti , G. , Calvani , A. , Ranieri , M. : Fondamenti di didattica. Teoria e prassi dei dispositivi formativi . Carocci ( 2007 )

2. Breslin , J. , Bojars , U. : SIOC Project Homepage ( 2008 ), http://sioc-project. org/

3. Briggs , R. , Reinig , B. : Bounded Ideation Theory: A New Model of the Relationship Between Ideaquantity and Idea-quality during Ideation . In: 2007 40th Annual Hawaii International Conference on System Sciences HICSS07 . pp. 16 { 16 . Ieee ( 2007 )

4. Chen , B. , Chuy , M. , Resendes , M. , Scardamalia , M. : "Big Ideas Tool" as a New Feature of Knowledge Forum . In: 2010 Knowledge Building Summer Institute . Toronto, Canada ( 2010 )

5. Girotra , K. , Terwiesch , C. , Ulrich , K.T.: Idea Generation and the Quality of the Best Idea . Manage. Sci . 56 ( 4 ), 591 {605 (Apr 2010 )

6. Kerr , D.S. , Murthy , U.S.: Divergent and Convergent Idea Generation in Teams: A Comparison of Computer-Mediated and Face-to-Face Communication . Group Decision and Negotiation 13 ( 4 ), 381 { 399 ( 2004 )

7. Kim , H.L. , Yang , S.K. , Song , S.J. , Breslin , J.G. , Kim , H.G. : Tag Mediated Society with SCOT Ontology . Business ( 2007 )

8. Litvak , M. , Last , M. , Aizenman , H. , Gobits , I. , Kandel , A. : DegExtA LanguageIndependent Graph-Based Keyphrase Extractor . Advances in Intelligent and Soft Computing 86 , 121 { 130 ( 2011 )

9. Miri , B. , David , B.C. , Uri , Z. : Purposely Teaching for the Promotion of Higherorder Thinking Skills: A Case of Critical Thinking . Research in Science Education 37 ( 4 ), 353 { 369 ( 2007 )

10. Motik , B. : OWL 2 Web Ontology Language Document Overview ( 2009 ), http: //www.w3.org/TR/2009/REC-owl2 - overview-20091027/

11. Nijstad , B.A. , Stroebe , W. : How the Group A ects the Mind: A Cognitive Model of Idea Generation in Groups . Personality and Social Psychology Review 10 ( 3 ), 186 { 213 ( 2006 )

12. Osborn , A.F. : Applied Imagination: Principles and Procedures of Creative Problem-Solving 3rd Edition. Creative Education Foundation ( 1993 )

13. Pissarra , J. , Jesuino , J.C. : Idea generation through computer-mediated communication: The e ects of anonymity . Journal of Managerial Psychology 20 ( 3 /4), 275 { 291 ( 2005 )

14. Reinig , B.A. , Briggs , R.O. : On The Relationship Between Idea-Quantity and IdeaQuality During Ideation . Group Decision and Negotiation 17 ( 5 ), 403 { 420 ( 2008 )

15. Riedl , C. , May , N. , Finzen , J. , Stathel , S. , Kaufman , V., Krcmar , H. : An Idea Ontology for Innovation Management . International Journal on Semantic Web and Information Systems 5 ( 4 ), 1 { 18 ( 2009 )

16. Schenker , A. , Last , M. , Bunke , H. , Kandel , A. : Classi cation of Web Documents Using a Graph Model . In: Seventh International Conference on Document Analysis and Recognition 2003 Proceedings . pp. 240 { 244 . No. Icdar, IEEE Comput. Soc ( 2003 )

17. Schneider , J. , Passant , A. , Groza , T. , Breslin , J.G. : Argumentation 3.0: how Semantic Web technologies can improve argumentation modeling in Web 2 .0 environments pp. 439 { 446 ( 2010 )

18. Yuan , S.T. , Chen , Y.C. : Semantic Ideation Learning for Agent-Based EBrainstorming . IEEE Transactions on Knowledge and Data Engineering 20 ( 2 ), 261 { 275 ( 2008 )