Fostering Knowledge Exchange Using Group Recommendations

Abstract

The more domain knowledge individual participants of a group decision process share with each other, the higher the probability of high-quality decision outcomes. In this paper we report the results of an initial empirical study conducted on the basis of a group decision support environment. In this study, groups were confronted with recommendations with a varying degree of diversity. The higher the diversity of recommendations provided to groups, the higher was the degree of knowledge exchange.

1. INTRODUCTION

In contrast to single user recommenders [ 7 ], group recommenders determine relevant items for whole groups [ 6, 10 ]. For example, Jameson [ 5 ] introduces a prototype application that supports groups of users in the identi cation of a holiday destination. Mastho [ 9 ] introduces concepts for sequencing television items for groups of users on the basis of di erent models from social choice theory (see also [ 10 ]). O'Connor et al. [ 16 ] introduce a collaborative ltering based movie recommender system that determines recommendations for groups of users. Ninaus et al. [ 15 ] show the application of group recommendation technologies in requirements engineering scenarios where stakeholders are in charge of cooperatively developing, evaluating, and prioritizing requirements. Finally, McCarthy et al. [ 12 ] introduce a critiquing-based recommender that supports groups of users in a skiing holiday package selection process. Choicla1 is a group decision support environment which includes group recommendation technologies { this system was used as a basis for the user study presented in this paper.

Psychological aspects of group decision making play an increasingly important role in the development of (group) recommendation technologies [ 8 ]. Especially decision biases which denote suboptimal shortcuts in decision making can lead to low-quality decision outcomes. Mastho and Gatt [ 11 ] discuss approaches to the prediction of user (group member) satisfaction with recommendations { in this context, conformity and emotional contagion are mentioned as major in uence factors. Felfernig et al. [ 4, 20 ] analyze the impact of conformity in the context of group decision making and report an increasing diversity of the preferences of group members the later individual preferences are disclosed to the whole group. Chen and Pu [ 2 ] show how emotional feedback from group members can be integrated in group (music) recommendation. An outcome of their study is that emotional feedback can enhance mutual awareness of user preferences in the group. For a short overview of decision biases in recommender systems we refer to Felfernig [ 3, 21 ].

The frequency of knowledge exchange within a group can have a major impact on the quality of the decision outcome [ 14 ]. The more decision-relevant knowledge is exchanged between individual group members, the higher is the probability of discovering the hidden pro le which can be characterized as the relevant knowledge to take a good (if optimality criteria exist, also an optimal) decision [ 22 ]. A consequence for group decision environments is that decision support has to include mechanisms that pro-actively encourage knowledge exchange. One reason for increased knowledge exchange between group members is group diversity (in terms of dimensions such as demographic and educational background), i.e., the higher the degree of diversity the higher the probability of higher quality decision outcomes (measured, e.g., in terms of the degree of susceptibility to the framing e ect [ 23 ]). Schulz-Hardt et al. [ 17 ] discuss the role of dissent in group decision making: the higher the dissent in initial phases of a group decision process, the higher the probability that the group manages to share the decisionrelevant information (discover the hidden pro le).

The major focus of our empirical study was to analyze the impact of recommendation diversity on the frequency of knowledge exchange between group members. A major reason for increasing the diversity of recommendations is the 1www.choicla.com. fact that otherwise recommendations are too similar to each other and thus provide only a limited coverage of the whole item space [ 13, 18 ]. There is always a trade-o between similarity and diversity since too diverse recommendations can lead to situations were relevant items are omitted, i.e., are not recommended although relevant for the user.

In this paper we do not focus on the prediction quality of recommendation algorithms but analyze in which way recommendations can be used to increase knowledge exchange between the members of a group. In the context of group decision making it is often more important to increase the performance of the group rather than predicting decisions that will be taken by the group. In this paper we analyze three di erent basic group recommendation heuristics (min, avg, and max group distance) with regard to their impact on the communication behavior inside a group. The basis for our analysis is an empirical study that was conducted in a computer science course at our university. The results of our analysis show that recommendation diversity can trigger additional (decision-relevant) communications.

The remainder of this paper is organized as follows. In Section 2 we describe the design of our user study and discuss related results. In Section 3 we discuss open issues for future work. With Section 4 we conclude the paper.

USER STUDY

The task of each group (of undergraduate students) in the empirical study (N=256 participants, 12% female, 88% male) was to select their preferred exam mode for their Software Engineering course, for example, 1 theoretical assignment on Object-Relational Mapping (ORM), 1 theoretical assignment on Sequence Diagrams, and two practical assignments on State Charts (see Figure 1). The participants were informed about the fact that there is no guarantee that the articulated preferences will be taken into account in upcoming exams. Each participant was a member of exactly one group (team) that had to implement a software within the scope of the course. Alternative exam modes (di erent topics and di erent shares of theoretical and practical assignments) were modeled in Choicla (see Figure 1).

Each group had the task to use the Choicla group decision support environment to cooperatively identify a ranking for the di erent assignment types. Each group member had to de ne his/her own ranking (see Figure 1) and was not able to see the preferences of the other group members. Participants of the study were encouraged to take a look at the group recommendations (tab group preferences) which was done by 91.41% of the participants at least once. Di erent group decision heuristics were used in our study and each group was assigned to a Choicla version that implemented exactly one of these heuristics.2 Related group recommendations d di er in terms of their diversity compared to the individual user ratings (rating scale: 1-5 stars) of an alternative s determined by eval(u; s) (see Formula 1). dj

! dj ! (1) (2) (3) GDmax (s) = arg max d2f1:::5g u2Users

X jeval (u; s)

Finally, average group distance represents a value between maximum and minimum group distance (see Formula 4). 2Note that Choicla includes a set of group heuristics from social choice theory [ 10 ], GDmax and GDavg have been included for the purpose of the empirical study.

diversity(d) =

Pu2Users jeval(u; s) #U sers dj

The (low diversity) minimum group distance heuristic (GDmin) returns a rating d that represents the minimum distance to the ratings of group members (see Formula 2).

GDmin (s) = arg

min d2f1:::5g u2Users

X jeval (u; s)

The (highly diverse) maximum group distance heuristic (GDmax) returns a rating d that re ects the maximum distance to current ratings of group members (see Formula 3). GDavg (s) = GDmin (s) + GDmax (s) (4) 2

An overview of the assignment of groups to the di erent decision heuristics is depicted in Table 1.

Hypotheses. The basic assumption of hypothesis H1 is that group decision heuristics with a higher diversity lead to an increased knowledge exchange between group members. The reason for this is that recommendations can act as an anchor [ 1 ] and also have the potential to induce the feeling of dissent in the group which needs to be resolved by the group members. An increased amount of knowledge exchange can help to discover the hidden pro le of a group decision [ 14, 22 ], i.e., the amount of decision-relevant knowledge is increased. Furthermore, we assume that a higher frequency of knowledge exchange is correlated with higher time e orts per group.

Examples of knowledge exchanged within the scope of our empirical study are the following (see Table 2).3

Content-related. A student only took a look at exercises related to Object-Relational Mapping (ORM) and asks for further information regarding the topic. Another student of the same group points out that there are only a few slides with very simple and understandable rules which are also very useful in industrial contexts.

Preference-related. A student emphasizes that he/she prefers to include appointments on UML Class Diagrams compared to appointments related to the Uni ed Process.

Recommendation-related. A student does not like the group recommendation since it does not take into account his/her preferences. Furthermore, he/she articulates an urgent need to further discuss assignment topics that are acceptable for the group as a whole. For recommendationrelated comments we also evaluated the valence, i.e., how positive/negative a recommendation was perceived.

The assumption of hypothesis H2 is that a higher degree of knowledge exchange increases the exibility of group members to change their initial preferences. Due to the fact that more decision-relevant knowledge is exchanged between group members, the amount of global decisionrelevant knowledge is increased which improves the individual capabilities of taking into account additional decision alternatives. Increased knowledge exchange between group members helps to overcome a discussion bias (group discussions tend to be dominated by information group members already knew before the discussion [ 19 ]).

Hypothesis H1 can be con rmed, i.e., the amount of decision relevant knowledge exchanged between group members increases with the diversity degree of the used group recommendation heuristic. The higher the diversity, the higher

the types content-related, recommendation-related was 3The categorization preference-related, performed manually. the number of decision-relevant comments given within the scope of the decision process (see Table 2). Furthermore, also the overall time investments increase with the diversity of the decision heuristic (see Table 3).

Summarizing, the higher the diversity of the used decision heuristic, the higher the frequency of knowledge exchange between group members. Consequently, recommendations in the context of group decision support can also be exploited to adapt a user's group decision behavior which can lead to higher quality decision outcomes. Diverse recommendations can help to detect hidden pro les [ 17, 19 ] which represent an amount of global decision-relevant knowledge needed to take good (or even optimal) decisions. Online group decision support environments have to be aware of this fact and should also take into account diversity in group recommendations. 3.

FUTURE WORK

Major issues for future work are the following. Our study is limited in the sense of having investigated a set of basic heuristics (diversity measures) (min, avg, and max group distance). In our future research we will investigate further decision heuristics (see, e.g., [ 10 ]) with regard to their capability to increase the frequency of knowledge exchange and to increase decision quality. We will also focus on a more ne-grained analysis of potential optimal degrees of diversity that help to maximize knowledge exchange while decreasing the perceived quality of recommendations as little as possible. The average diversity (Formula 1) of recommendations avg max determined by the three di erent heuristics is depicted in Table 5. We want to emphasize that the satisfaction with group recommendations signi cantly decreases if the degree of diversity is too high { Table 6 summarizes user feedback regarding the perceived satisfaction with the group recommendations.

CONCLUSIONS

In this paper we presented the results of an initial empirical study that focused on possibilities of increasing the amount of knowledge exchange in group decision scenarios. In this context, we showed that the diversity of recommendations can have a signi cant impact on the frequency of knowledge exchange { the higher the diversity of group recommendations, the higher the corresponding number of comments included in the group decision process. The results presented in this paper are a rst step towards the application of recommendation technologies to foster knowledge exchange in group decision making.

References

[1] Adomavicius , G. , Bockstedt , J. , Curley , S. , and Zhang, J. Recommender systems, consumer preferences, and anchoring e ects . In Decisions@RecSys11 Workshop (Chicago, IL, USA, 2011 ), 35 { 42 .

[2] Chen , Y. , and Pu , P. Cofeel: Emotional social interface in group recommender systems . In RecSys'12 Workshop on Interfaces for Recommender Systems (Dublin, Ireland, 2012 ), 48 { 55 .

[3] Felfernig , A. Biases in decision making . In International Workshop on Decision Making and Recommender Systems 2014 , vol. 1278 of CEUR Proceedings ( 2014 ), 32 { 37 .

[4] Felfernig , A. , Zehentner , C. , Ninaus , G. , Grabner , H. , Maaleij , W. , Pagano , D. , Weninger , L. , and Reinfrank , F. Group decision support for requirements negotiation . In Advances in User Modeling , vol. 7138 of LNCS, Springer ( 2012 ), 105 { 116 .

[5] Jameson , A. More than the sum of its members: Challenges for group recommender systems . In Intl. Working Conf. on Adv. Visual Interf . ( 2004 ), 48 { 54 .

[6] Jameson , A. , and Smyth , B. Recommendation to Groups. In The Adaptive Web: Methods and Strategies of Web Personalization , P. Brusilovsky , A. Kobsa , and W. Neijdl, Eds. Springer, 2007 , ch. 20 , 596 { 627 .

[7] Jannach , D. , Zanker , M. , Felfernig , A. , and Friedrich , G. Recommender Systems: An Introduction . Cambridge University Press, 2010 .

[8] Mandl , M. , Felfernig , A. , Teppan , E. , and Schubert , M. Consumer Decision Making in Knowledge-based Recommendation . Journal of Intelligent Information Systems (JIIS) 37 , 1 ( 2010 ), 1 { 22 .

[9] Mastho , J. Group modeling: Selecting a sequence of television items to suit a group of viewers . UMUAI 14 , 1 ( 2004 ), 37 { 85 .

[10] Mastho , J. Group Recommender Systems: Combining Individual Models. Recommender Systems Handbook ( 2011 ), 677 { 702 .

[11] Mastho , J. , and Gatt , A. In Pursuit of Satisfaction and the Prevention of Embarrassment: A ective State in Group Recommender Systems . User Modeling and User-Adapted Interaction 16 , 3 { 4 ( 2006 ), 281 { 319 .

[12] McCarthy , K. , Salamo , M. , Coyle , L. , McGinty , L. , Smyth , B. , and Nixon , P. Group Recommender Systems: A Critiquing based Approach . In IUI'06 , ACM ( 2006 ), 267 { 269 .

[13] McGinty , L. , and Smyth , B. On the Role of Diversity in Conversational Recommender Systems . In 5th Intl. Conf. on Case-based Reasoning ( 2003 ), 276 { 290 .

[14] Mojzisch , A. , and Schulz-Hardt , S. Knowing Other's Preferences Degrades the Quality of Group Decisions . Jrnl. of Pers. & Social Psy . 98 , 5 ( 2010 ), 794 { 808 .

[15] Ninaus , G. , Felfernig , A. , Stettinger , M. , Reiterer , S. , Leitner , G. , Weninger , L. , and Schanil , W. Intelligent techniques for software requirements engineering . In Europ. Conf. on AI , Prestigious Applications of Intelligent Systems (PAIS) ( 2014 ), 1161 { 1166 .

[16] O 'Connor , M. , Cosley , D. , Konstan , J. , and Riedl , J. PolyLens: A Recommender System for Groups of Users . In Europ. Conf. on Computer-Supported Cooperative Work , ACM ( 2001 ), 199 { 218 .

[17] Schulz-Hardt , S. , Brodbeck , F. , Mojzisch , A. , Kerschreiter , R. , and Frey , D. Group Decision Making in Hidden Pro le Situations: Dissent as a Facilitator of Decision Quality . Journal of Personality & Social Psychology 91 , 6 ( 2006 ), 1080 { 1093 .

[18] Smyth , B. , and McClave , P. Similarity vs . diversity. In 4th Intl. Conf. on Case-based Reasoning (ICCBR'01) , Springer (London, UK, 2001 ), 347 { 361 .

[19] Stasser , G. , and Titus , W. Pooling of Unshared Information in Group Decision Making: Biased Information Sharing During Discussion . Jrnl. of Pers. and Social Psy . 48 , 6 ( 1985 ), 1467 { 1478 .

[20] Stettinger , M. , Felfernig , A. , Leitner , G. , and Reiterer , S. Counteracting Anchoring E ects in Group Decision Making . In UMAP 2015 , vol. 9146 of LNCS (Dublin, Ireland, 2015 ), 118 { 130 .

[21] Stettinger , M. , Felfernig , A. , Leitner , G. , Reiterer , S. , and Jeran , M. Counteracting Serial Position E ects in the CHOICLA Group Decision Support Environment. In ACM IUI2015 (Atlanta, GA , USA, 2015 ), 148 { 157 .

[22] Wittenbaum , G. , Hollingshead , A. , and Botero , I. From Cooperative to Motivated Information Sharing in Groups: Moving Beyond the Hidden Pro le Paradigm . Comm. Monographs 71 , 3 ( 2004 ), 286 { 310 .

[23] Yaniv , I. Group Diversity and Decision Quality: Ampli cation and Attenuation of the Framing E ect . International Journal of Forecasting 27 ( 2011 ), 41 { 49 .