eMediation
Towards Smart Online Dispute Resolution
E. Fersini
1
, E. Messina
1
, L. Manenti
1
, G. Bagnara
1
S. El Jelali
1
and G. Arosio
2
1
DISCo, University of Milano-Bicocca,Viale Sarca 336, Milan, Italy
2
Consorzio Milano Ricerche, Viale Cozzi 53, Milan, Italy
Keywords:
e-Justice, Online Dispute Resolution, eMediation, Knowledge Management, Machine Learning.
Abstract:
In this paper, the main requirements towards the next generation of Smart Online Dispute Resolution Systems
for eMediation are presented and addressed through the definition of an advanced computational intelligence
framework. The main contributions can be distinguished with respect to the parties involved in the eMediation
process. Concerning the disputants, the main advancements are related to a smart data collection environment
to state the essence of the litigation and an intelligent retrieval of court decisions to improve the awareness of
the parties about their liability. Regarding the role of the mediator, the essential point addressed relates to an
estimation of disputant flexibility to facilitate the optimal mediation strategies.
1 INTRODUCTION
The increasing workload of civil justice courts is en-
couraging the adoption of novel litigation support
systems. A recent EU Directive
1
highlights the im-
portance of facilitating access to Alternative Dispute
Resolution (proceedings with no formal court hear-
ing), to promote the amicable settlement of disputes
by encouraging the use of ADR and ensuring a bal-
anced relationship between ADR and judicial proce-
dures. In order to understand the role that ICT could
play in settlement of conflicts, we report the num-
bers of Alternative Dispute Resolution (ADR) cases
addressed at European level. According to the 2011
report disclosed by the EU Parliament the increasing
trend in the use of ADR counts about 410.000 cases
in 2006, 473.000 in 2007 and more than 500.000 in
2008. More impressive numbers are related to the
Italian context, with a particular focus on mediation
(one of the available schemas for ADR). Accord-
ing to the statistics provided by the Italian Ministry
of Justice, about 231.500 cases have been addressed
through ADR between March 2011 and September
2013. Italy’s Central Bank has estimated 16 billion
euro loss in terms of GDP caused by the slow of civil
justice, highlighting the needs of encouraging alterna-
tive resolutions of disputes both from citizen and “jus-
tice system” points of view. The comparison of Italian
1
2008/52/EC of 21st May 2008
in and out of court civil proceedings - average trial du-
ration of 1066 days versus 65 days time limit for me-
diation - makes clear the role that ICT could play to
shift from ADR to ODR (Online Dispute Resolution),
where technology could facilitate and speed up the
resolution of disputes. Several commercial initiatives,
mainly focused on Internet-based support toolsets,
have been introduced to enable ODR, while the re-
search initiatives are mainly focused on advanced rea-
soning mechanisms to facilitate the resolution of the
disputes under a specific negotiation schema. A first
example of computational negotiation system is rep-
resented by DEUS (Zeleznikow et al., 1995) that, re-
ceiving as input specific goals and beliefs of litigants,
computes the agreement level in family law property
negotiations. More sophisticated systems are repre-
sented by Split-Up (Zeleznikow and Stranieri, 1995)
and Family-Winner (Bellucci and Zeleznikow, 2001).
Split-Up is a hybrid framework that combines rule-
based systems and neural networks to assist disputes
about properties distribution. Family-Winner, which
is a game theory-based approach for Australian fam-
ily negotiations, asks disputants to list the items in-
volved in the litigation and to assign a corresponding
relevance value. The system formulates an influence
diagrams to subsequently use game theory to deter-
mine a suitable trade-off between claims. Moreover,
the BEST-project (Uijttenbroek et al., 2008) provides
ontology-based search of law cases to allow parties
the opportunity to evaluate claims and liabilities. (For
228
Fersini E., Messina E., Manenti L., Bagnara G., El Jelali S. and Arosio G..
eMediation - Towards Smart Online Dispute Resolution.
DOI: 10.5220/0005080002280236
In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2014), pages 228-236
ISBN: 978-989-758-050-5
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
a review on commercial and research initiatives for
ODR refer to (Carneiro et al., 2014)). The results
obtained by the above mentioned investigations are
mainly related to negotiation as a form of dispute res-
olution. To the best of our knowledge, until now no
contribution has been given with respect to mediation
as an alternative schema for out of court disputes. In
this paper, the main requirements towards the next
generation of Smart ODR Systems for eMediation are
presented and addressed through the definition of a
computational framework based on advanced artifi-
cial intelligence methodologies.
2 TOWARDS SMART ODR
SYSTEMS
The spread of mediation through a next generation of
ODR systems, is dependent on the suitable exploita-
tion of communication technologies coupled with ad-
vanced computational intelligence approaches such
as knowledge management, machine learning, infor-
mation retrieval and operational research. The re-
quirements towards an effective support to eMedia-
tion by Smart ODR Systems should take into account
the needs both disputants and mediators. Concerning
the disputants involved in a litigation, the essential re-
quirements are (1) a smart data collection functional-
ity to state the essence of the litigation and (2) an ef-
fective retrieval system of court decisions to improve
the awareness of the parties about their BATNA
2
and
WATNA
3
. Regarding the role of mediator, the main
desiderata relates to (3) the possibility of estimating
the flexibility of the parties to be guided through a set
of mediation strategies to likely settle an agreement.
Smart Data Collection. A Smart ODR System
should enable the acquisition of information about the
citizen’s case. An intuitive support to the collection
of case characteristics is compulsory for enabling any
decision process, both from disputant and mediator
points of view. If we focus on the state of the art
related to ODR, we can easily point out that claims
and requirements are typically collected by a fixed-
structure template to be filled in by parties, with no
possibility to provide argumentations by using natu-
ral language. In this context, a fundamental role is
played by those mechanisms able to acquire informa-
tion from the parties by following an intuitive data
collection process. A smart assistant able to guide the
2
Best Alternative To a Negotiation Agreement, i.e. the
best option a party has if negotiation fails
3
Worst Alternative To a Negotiation Agreement, i.e. the
worst option a party has if a resolution cannot be reached
disputants to provide the right information about their
case represents a crucial leverage to enable either “ar-
tificial” or “human” reasoning mechanism concerned
with ODR. In particular, a proper acquisition of the
case description could improve the retrieval of court
decisions related to a given case for helping the dis-
putant in better understanding his/her position: of
rights and duties, chances in a potential in-court lit-
igation, time and costs, and so on. Moreover, it could
also help the mediators for better understanding the
case, for instance by providing a summary of ques-
tions and answers that characterize the case itself. In
order to match this requirement, a smart data collec-
tion based on knowledge needs to be introduced in the
next generation of Smart ODR systems.
Improving Awareness of the Parties. From the dis-
putant point of view, once a proper acquisition of
user claims and requirements has been performed,
the first step towards a Smart ODR System is repre-
sented by the well-known BATNA and WATNA con-
cepts of principled negotiation (Fisher et al., 2011). In
fact, crucial requirements for enabling the adoption of
ODR systems are (1) the natural language argumenta-
tion of claims and requests related to the citizens case
and (2) the possibility to gain knowledge by easily
accessing and consulting former court decisions re-
lated to similar disputes. Indeed, glancing at similar
cases to understand rights and duties, relevant norms,
times and costs of potential in-court-proceedings and
prospective outcomes of the dispute may increase the
awareness of the disputant. Providing information
about the party legal positions could help to improve
the awareness about their own liability and to figure
out their chances in court proceedings (usually over-
estimated). Therefore, Smart ODR Systems should
provide a retrieval functionality able to bring the gap
between the layman case description and the court de-
cisions.
Enhancing the Flexibility of Parties. If we focus
on the mediator needs, the main requirement towards
Smart eMediation Systems relates to the evaluation
and the improvement of the flexibility of the dis-
putants involved in a conflict. Improve this flexibility
could increase the possibility to settle an agreement
when critical situations occur. The flexibility of a dis-
pute, which mainly depends on the propensity of the
parties to achieve an agreement, is a key aspect that a
mediator should evaluate after the first meeting with
the parties. When a critical situation occurs, i.e.when
the party’s flexibility is low or large asymmetries are
detected, resolution strategies should be suggested to
the mediator for improving the possibility to achieve
an agreement among parties. This could help media-
tors to enhance the flexibility of the parties by analyz-
eMediation-TowardsSmartOnlineDisputeResolution
229
ing the initial state of a mediation and detecting crit-
ical situations a-priori. The possibility of comparing
different what-if scenarios could help in the planning
of further meeting and in deciding what strategies to
apply in a particular scenario. Moreover, the system
could also be used as learning environment in which
hypothesize and compare different real situations or
toy examples, analyze flexibility evaluation and com-
pare suggested strategies.
These requirements have converged in the eJRM
project, acronym of electronic Justice Relationship
Management , which represents an ongoing Italian
initiative aimed at improving the awareness of citi-
zens to personally evaluate the outcome of a poten-
tial litigation and to be guided to a non-conflict set-
tlement. In particular, the above mentioned require-
ments have been addressed to allow a radical im-
provement of two major processes:
• eMediation: online management of activities re-
lated to the mediation process
• Self-Litigation: capability of a citizen to au-
tonomously classify, formalize and understand the
potential outcome of a dispute
In the following sections, the main contributions to-
wards a Smart ODR for eMediation are detailed.
3 ONTOLOGY-DRIVEN DATA
ACQUISITION
An interactive and self-administered interviewing
system has been designed in order to collect use-
ful information for enabling either eMediation or
Self Litigation processes (Arosio et al., 2013). Dis-
putants respond to a sequence of questions on their
specific case: the system selects pertinent questions
depending on the disputants’s individual responses.
Upon completion, a summary is presented to the par-
ties. The interview system, based on specific do-
main knowledge, helps both disputants and mediator
to save time, offers a database for retrieval function-
alities, and provides only relevant information related
to the conflict. The Ontology-driven Data Acquisi-
tion system (ODA) is composed of two main parts:
(1) an ontological structure aimed at modeling the ju-
ridical knowledge related to a specific application do-
main and (2) a logical engine targeted at exploring the
ontological structure in order to provide questions and
collect responses to/from the disputant.
3.1 Ontological Structure
The ontological structure underlying ODA formalizes
the concepts to be acquired from the user by present-
ing a set of interrelated questions. Basically, con-
cepts and relationships have been designed in order to
represent a smart question-answering flow, by mod-
eling yes/no questions, multiple-choice questions as
well as the potential correspondence between answers
and violation of specific norms. Questions, Responses
and Norms (applicable rules and violated norms)
have been modeled as schema concepts (owl:Class),
while the individuals - that can belong to only one of
the three concepts - are acquired via questions pre-
sented to the user. Whenever a given norm is vio-
lated, the system presents the text of the correspond-
ing norm for consultation purposes.
The relationships (predicates) designed in the
ontological structure link couples of concepts in
order to represent the whole question flow pre-
sented to the user. The considered predicates
(owl:objectProperty) are listed below:
• assume: relationship that links two concepts such
that the domain concept is verified if and only if
the co-domain concept has been already acquired
during the interview;
• assumeAND: relationship that links a concept to
multiple concepts such that the domain concept is
verified if and only if all the co-domain concepts
have been already acquired during the interview;
• assumeEXCL: relationship that links a concept to
multiple concepts such that the domain concept is
verified if and only if only one of the co-domain
concepts has been previously acquired;
• assumeOR: a relationship that links a multiple-
choice question concept with all its own answer
concepts;
Between questions and norms, two kinds of relation-
ships have been modeled, a violatedWhen relation-
ship that links a norm to a question that could lead to
its violation and a verifiedWhen relationship linking a
norm to a question that could lead to its compliance.
The modeling of knowledge to be included in the on-
tological structure relates to specific domains, but it
does not require a tedious and often unfeasible top-
down approaches for representing the whole domain
considered. The modelling criteria are indeed generic
and can be in principle applied to any judicial domain.
In our investigation, locations, taxes and tributes, and
civil liability related to the use of motor vehicles have
been modeled as case studies.
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
230
3.2 Logical Engine
In order to explore the ontological structure as defined
in the previous subsection, we have defined a logi-
cal engine able to provide a context-sensitive adaptive
questionnaire. The logical engine has two main goals:
(1) to explore the ontology in order to gather con-
cepts to be characterized by the disputant and (2) to
show to the user the question related to the given con-
cept and to acquire his/her response. Concepts mod-
eled within the ontological structure represent rele-
vant information to be acquired in order to charac-
terize a disputant’s case. Each concept is therefore
directly connected with the corresponding question to
be presented to the user. In order to provide a mech-
anism able to firstly explore the ontological structure
and consequently manage questions/answers, we de-
signed a logical engine based on the “Last State-Next
State” model (LSNS) (Bouamrane et al., 2008). Ac-
cording to this model, a given concept to be acquired
(what question is currently processed) could lead to
several potential subsequent concepts to be explored
(the next question given the user’s response). The pro-
posed logical engine implements a short-term mem-
ory approach based on predicate priorities. This al-
lows us to provide questions related to a given con-
cept only if prerequisite concepts have been previ-
ously verified (according to their ontological proper-
ties). This logical engine makes it possible to dis-
regard long-term dependencies among the concepts
underlying the ontology questionnaire and to be inde-
pendent on the juridical domain modeled. The system
makes also available the consultation of the norms
(both applicable and violated) according to the dis-
putant’s responses, enabling therefore a better under-
standing of rights and duties. The main advantage of
the proposed systems, both for disputants and media-
tors, is the reduction of time in respect of the a simple
conditional branching questionnaire. Moreover, ODA
system helps disputants to autonomously classify, for-
malize and understand the potential outcome of a lit-
igation, as well as mediator to save time and to avoid
biases and errors when collecting the interview’s re-
sponses.
4 MACHINE LEARNING BASED
RETRIEVAL
In order to improve the awareness of the parties in-
volved in a eMediation, a retrieval functionality needs
to be designed to support the disputants in glanc-
ing court decisions that could potentially be related
to their own case description. The main issues that
a retrieval functionality should address relates to the
nature of the language, i.e. formal and verbose for
court decisions and informal and concise for the con-
flict description provided by the disputants. In order
to deal with this issue, a Machine Learning based Re-
trieval has been defined. The main functionalities of
this component are:
1. Indexing: is the first step for processing the court
decisions documents in order to create a struc-
tured representation as bag-of-words. First of all,
stop-words, digital numbers and separator charac-
ters are removed. Therefore, a stemming is ap-
plied to the resulting terms in order to reduce the
size of the terms dictionary
4
and thus to prevent a
potential over-fitting phenomenon.
2. Core Mining: the goal of this functionality is
to associate a main topic (e.g., Damage, Fam-
ily, Failure of business, Divorce, etc...) to a dis-
putant description. First, court decisions are rep-
resented according to the bag-of-words models
(Salton et al., 1975) and the TF-IDF scoring tech-
nique (Salton and Buckley, 1988). Considering
that the bag-of-words representation is character-
ized by a huge number of attributes (terms) that
could affect the prediction of the topic related to
a given dispute description, a dimensionality re-
duction has been enclosed. In particular, Prin-
cipal Component Analysis has been adopted to
reduce the dimension of the bag-of-words repre-
sentation. Generally, 98% of terms are reduced
without losing the relevant information, that al-
lows to build a classification machines in an ef-
ficient manner. Once a proper representation has
been derived for court decision documents, sev-
eral machine learning classifiers (Support Vector
Machine, Naive Bayes and Tree Decision) have
been trained to derive a model able to pr capture
the topic related to a dispute description.
3. Query Processing: Before providing the dispute
description to the Core Mining functionality, the
corresponding text has been tokenized and prepro-
cessed. The output is a vector of terms that has the
same dimension of the term dictionary (the terms
extracted from the user query that do not match
the terms dictionary are ignored).
4. Ranking: It computes the similarity between the
disputant case description and the court decision
documents corresponding the topic predicted by
the Core Mining functionality. Then, it ranks the
similarity scores in a descendant order to finally
4
The terms dictionary is the list of terms shared among
court decisions
eMediation-TowardsSmartOnlineDisputeResolution
231
provide the top ten court decisions to the dis-
putant. In order to address the issue related to
the language difference between the court deci-
sion documents and the disputant query, two main
contributions have been provided:
• Term Evaluation and Selection (TES), to better
identify the terms that characterize a given topic
(e.g. vehicle for the topic Car Accident)
• Zipf-based Similarity Measure (ZSM), to bet-
ter capture the similarity of the natural lan-
guage used both in court decisions and dis-
putant query.
TES is based on two criterions for detecting the
Relevant Terms (RTs) from the court decision
document collection: the percentage of term fre-
quency per class (ptfc) and the term presence per
each class. If a term is omni-present in a class
(high value of pftc), the term will be a RT, oth-
erwise it will be disregarded. RTs are then used
to train classifiers and to predict the most prob-
able topic of a given disputant case description.
Once a topic has been associated to the disputant
case description, the most relevant court decisions
are determined according to the ZSM. To this pur-
pose, court decisions and dispute description have
been represented by using power law distributions
that reflect the Zipf’s Law underlying the natural
language. Once the Zipf distributions have been
derived, ZSM can applied: term frequencies that
are close to the mean of the distributions (i.e. the
most important terms) contribute more in the sim-
ilarity computation, while either rare or common
terms have a reduced impact. ZSM allows us to
tackle the retrieval problem for eMediation pro-
cesses: ZMS is able to addresses different kind of
languages, i.e. lengthy for court decisions while
condensed for the conflict description provided by
the disputants.
5 FLEXIBILITY-BASED
RESOLUTION ASSISTANT TO
eMEdiations
In the context of eMediation, at our knowl-
edge, Flexibility-based Resolution Assistant to
eMEdiations (FRAME) represents the first attempt to
quantitatively model qualitative aspects that media-
tors usually consider in the management of a medi-
ation: during the first fact-finding meeting, mediators
evaluate the initial state of the mediation, in order to
understand how to conduct the further meetings with
the disputants with the scope to solve the conflict. The
evaluation of the dispute and the decisions about the
actions to be taken to likely achieve an agreement are
performed by mediators in an implicit way. This pro-
cess is mainly related to the experience of the medi-
ator itself. Starting from this consideration, the main
scope of FRAME is to model elements considered by
mediators in evaluating and conducting the mediation
process: in order to achieve this aim, and to under-
stand which aspects are actually relevant, different in-
terviews with a domain expert (i.e. a mediator with
more than 20 years of experience in the mediation
field) were conducted in order to acquire all the rel-
evant information in this context. These knowledge
acquisition sessions produce as output a representa-
tion of the mediator mental process in terms of fac-
tors that characterize the mediation and determine its
possibility to reach a positive solution, and a set of
strategies/mechanisms that the experts can apply in
order to solve critical situations.
In order to match the requirements introduced in
section 2 and to help mediators in the management
of their disputes, we have developed FRAME taking
into account all the information acquired during the
interviews with the domain expert. FRAME is a web-
based Java application, composed of two parts: the
first one relates to the evaluation of a mediation in
terms of flexibility values associated to the disputants
involved, and the second one relates to the sugges-
tion of strategies/mechanisms to be applied to likely
achieve an agreement among parties. In the next sec-
tions, the two components will be discussed.
5.1 Flexibility Evaluation
A value of flexibility, which represents the disputant’s
positive aptitude to achieve an agreement, can be as-
sociated to each litigant
5
. According to the knowl-
edge acquired from the domain expert, a set of rel-
evant factors that could be observed in the case of
Italian mediation have been identified. These factors
have been used to create a fixed-structure question-
naire to be compiled by the mediator after the first
meeting with the disputants, where each multiple-
choice question is modeled to acquire the value of
a given factor as positive or negative. We define F
as the set of factors that influences the resolution of
a mediation. We identify three categories of factors,
such that F = Fm ∪ Fd ∪ F p:
• F
m
= { f
m
1
, . . . , f
m
α
}, set of factors that character-
ize the mediation in terms of the full process, such
as the kind of mediation, the type of mediation
5
An impasse or a critical situation are mainly due to low
levels of flexibility and/or large asymmetry between the dis-
putant flexibility.
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
232
opening demand, the presence of an expert in the
dispute domain, and so on;
• F
d
= { f
d
1
, . . . , f
d
β
}, set of factors that characterize
the dispute, such as the economic value, the level
of complexity, the overall level of conflict and so
on;
• F
p
= { f
p
1
, . . . , f
p
γ
} , set of factors that character-
ize the parties involved in the mediation, such as
their attitude toward the mediation process, their
behavior with respect to the mediator and to the
other parties, their necessity to close the dispute
and so on.
Inspired by the work of Druckman (Druckman
et al., 2002; Druckman et al., 2004), the overall flexi-
bility associated to a disputant h has been defined as:
Flex(h) =
|F
m
|
∑
i=1
f
m
i
w
i
v
i,h
+
|F
p
|
∑
j=1
f
p
j
w
j
v
j,h
+
|F
p
|
∑
k=1
f p
k
w
k
v
k,h
where w
ζ
∈ [0, 1] denotes the relevance weight of each
factor f ∈ F and v
ζ,h
represents an estimate provided
by the mediator to a given factor with respect to the
given dispute. The variable v
ζ,h
takes values into the
continuous interval [1, 4], where v
ζ,h
∈ [1 − 2] means
a negative evaluation of the factor while v
ζ,h
∈ [3 − 4]
a positive one. For seek of simplicity, let’s assume a
mediation with two parties. In this case, the question-
naire is composed of a total of 54 questions (the val-
ues between brackets represent the estimate v
ζ,h
pro-
vided by the mediator to a given factor with respect to
the given dispute):
• Mediation factors [F
m
] (8 questions):
◦ [M1] type of mediation: voluntary (4) - stated
by contract (3) - court ordered (2) - compulsory
(1);
◦ [M2] type of instance: joint (4) - ordinary (2);
◦ [M3] information of the mediation instance:
complete (4) - incomplete (1);
◦ [M4] information to the disputants: complete
(4) - incomplete (1);
◦ [M5] disclosure: yes (4) - no(1);
◦ [M6] expert involved: yes (4) - no (1);
◦ [M7] confidentiality: reserved (4) - public (1);
◦ [M8] relationship among disputants: amicable
(4) - good natured (3) - neutral (2) - hostile (1);
• Dispute factors [F
d
] (6 questions):
◦ [D1] level of divergence among parties: no con-
flict (4) - weakly conflictual (3) - slightly con-
flictual (2) - strongly conflictual (1);
◦ [D2] third parties involved: yes (4) - no (1);
◦ [D3] economic value of the dispute: up to 50k
euro (4) - from 50k to 100k euro (3) - from 100k
to 250k euro - above 250k euro;
◦ [D4] complexity of the dispute: easy (4) -
weakly complex (3) - slightly complex (2) -
strongly complex (1);
◦ [D5] number of elements that can leverage the
parties to settle an agreement: high number (4)
- medium number (3) - sufficient number (2) -
scarce number (1);
◦ [D6] experience required to the mediator: low
(4) - medium (3) - high (2) - very high (1);
• Parties factors [F
p
] (20 questions):
◦ [P1] mediation professional body: selected by
the disputant (4) - non selected by the disputant
(1) - no mediation instance (4);
◦ [P2] joint judicial instance: yes (4) - no (1);
◦ [P3] time needs to close the dispute: high (4) -
medium (3) - low (2) - no time limits (1);
◦ [P4] costs needs to close the dispute: high (4) -
medium (3) - low (2) - no time limits (1);
◦ [P5] type of disputant: natural person (4) - legal
representative (1);
◦ [P6] type of lawyers: lawyer and mediator (4) -
lawyer (1);
◦ [P7] lawyer predisposition: collaborative (4) -
positive (3) - indifferent (2) - uncooperative (1);
◦ [P8] disputant objective vision: high (4) -
medium (3) - low (2) - no objective vision (1);
◦ [P9] understanding of the other disputant posi-
tion: total (4) - partial (3) - sufficient (2) - no
understanding (1);
◦ [P10] previous relationships among parties:
previous relationships (4) - no former relation-
ships (1);
◦ [P11] possibility of future relationships among
disputants: high (4) - medium (3) - low (2) - no
possibility (1);
◦ [P12] socio-cultural aspect: equal (4) - higher
than the other (2) - lower than the other (2) -
very high/low than the other (1);
◦ [P13] level of education: equal (4) - higher than
the other (2) - lower than the other (2) - very
high/low than the other (1);
◦ [P14] level of psychological intimidation: no
intimidation (4) - weakly intimidated (3) -
slightly intimidated (2) - strongly intimidated
(1);
◦ [P15] emotional involvement: no involvement
(4) - weakly involved (3) - slightly involved (2)
- strongly involved (1);
eMediation-TowardsSmartOnlineDisputeResolution
233
◦ [P16] conflict of disputant: high (4) - medium
(3) - low (2) - no conflict (1);
◦ [P17] linguistic register: clear (4) - vague (1);
◦ [P18] linguistic/gesture: friendly (4) - aggres-
sive (1);
◦ [P19] listening availability: high (4) - medium
(3) - low (2) - no availability (1);
◦ [P20] behavior during the meeting: friendly (4)
- sully (1);
Once the questionnaire has been filled in by the me-
diator, the partial flexibility for each factor set (i.e.
F
m
, F
d
and F
p
) is computed and the final flexibil-
ity Flex(h) is estimated for each disputant involved
in the conflict. Flexibility values are then displayed
to the mediator and graphically shown in a Cartesian
diagram displayed as a grid (see Figure 1). Cells C
and E represent agreement among parties, in particu-
lar joint maximum flexibility in cell C and joint mod-
erate flexibility in E. Asymmetrical flexibility can be
identified in cells A and I. The remaining cells rep-
resent no agreement among parties, with cell G in-
dicating joint intransigence or no movement. Figure
1 shows an example of the flexibility graphical rep-
resentation, in which a large asymmetry between the
two parties can be easily detected: party 1 (blue line)
and party 2 (green line) have a flexibility value equal
to 37.1% and 65.28%, respectively. The intersection,
represented as a red point, provides a first qualitative
evaluation of the state of the mediation.
Figure 1: Graphical representation of party 1 (blue line) and
party 2 (green line) flexibility: the red point indicates the
intersection between the two flexibility lines and provides a
qualitative evaluation of the state of the mediation.
In the example above the state of the mediation (red
point) denotes a potential agreement among parties
(cell E) but with a quite large asymmetry (red point
far from the bisector). The best case appears when
the intersection of flexibility lines lies on the top-
right quadrant (cell C) and near to the bisector axis
(i.e. the two values of flexibility are enough high and
there is no asymmetry). Differently, when the inter-
section lies in the other quadrants and/or far from the
bisector axis, the mediation can present critical situ-
ations of impasse or low levels of flexibility for one
or for both parties. The evaluation of flexibility pro-
vided by FRAME is a first contribution towards Smart
ODR Systems for eMediation, which not only helps
the mediators to understand critical situations but also
to clearly state the factors (mediation, conflict and
parties) that partially affect the settlement of agree-
ments. After the flexibility evaluation, the mediator
can choose to start an analysis of the initial situation,
in order to automatically identify critical situations
and obtain suggestions related to which strategies and
mechanisms can be used in case of low values of flex-
ibility or asymmetries.
5.2 Strategy Suggestion
An analysis on the state of the art of the mediation
management (Danovi and Ferraris, 2013; Bush and
Folger, 2004) has revealed that some mechanisms and
strategies exist and can be applied to improve the flex-
ibility of the disputants and to solve asymmetries. To
this purpose, a set of strategies has been identified as
potential leverage that could have positive impact on
some factors and therefore improve the overall flexi-
bility. More in detail, we have defined a set of strate-
gies S = {s
1
, . . . , s
r
} where each s
l
has a cost c
l
∈ [0, 1]
and a probability g
l
∈ [0, 1] to solve a given subset of
factors. Each strategy can be applied to improve the
value of some factors f ∈ F when they are evaluated
as negative by the mediator answering to the question-
naire (i.e., when factors takes values among 1 and 2
in the evaluation process). The strategies identified by
the domain expert are reported in the following:
• communication to improve and manage the com-
munication between the disputants;
• focus on mediation principles to recall the dis-
putants the mediation purposes and benefits;
• containment to smooth the contraposition be-
tween disputants;
• active listening to bring a disputant towards the
understanding of the other litigant empathy;
• empowerment to make disputants more confident
in rights and duties;
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
234
• improve the mediator authoritativeness to make
disputants more confident in the mediator;
• entailment of lawyers to involve more lawyers and
therefore parties;
• search of the substantial leader to improve his/her
involvement into the mediation;
• search of information to discover additional doc-
uments and objective evaluations;
• interests finding to discover emotional aspects that
can be used to understand and solve the dispute;
• brainstorming free discussion about the dispute
that could enable novel and original solutions;
• involvement of experts to collect additional infor-
mation and evaluation about the dispute;
• private meetings with disputant and lawyer to dis-
cover additional (private) information;
• reality test to allow parties to understand real and
objective needs.
FRAME performs a two-phase analysis to find the set
of strategies to be proposed to the mediator to improve
the chance to find an agreement: the first step consists
in the analysis of the collected answers, in order to
identify the factors that have been indicated as nega-
tive by the mediator. A set of the potentially applica-
ble strategies is then inferred by means of an expert
system based on JESS rule-based engine
6
. After the
identification of this preliminary set, a refinement is
produced by solving an integer linear programming
problem in order to find the optimal set of strategies
to be applied. The goal is to propose to the mediator
only the strategies that (i) minimize the effort of the
mediator and that (ii) maximize the value of flexibility
of the party (under the constraint to limit the asymme-
try between the two parties). The identification of the
optimal set of strategies to be applied considers two
main aspects: (1) the cost c
l
∈ [0, 1] of each strategy,
i.e. the difficulty for a mediator to apply a given tech-
nique, and the corresponding probability g
l
∈ [0, 1],
i.e. how likely a strategy can solve the related fac-
tors. Once the optimal set of strategies has been de-
termined, it is provided to the mediator in order to
improve the chances to achieve an agreement.
6 CONCLUSIONS
The diffusion of both ADR and ODR has become a
significant factor in instilling confidence in the legal
framework as a whole, supporting and promoting the
6
http://herzberg.ca.sandia.gov/
rule of law. In this paper, the main requirements to-
wards Smart Online Dispute Resolution for eMedia-
tion has been discussed. As a consequence, a com-
putational intelligent system has been proposed to ad-
dress (1) a smart data collection to state the essence of
the litigation, (2) an intelligent retrieval of court deci-
sions to improve the awareness of the parties about
their liability and (3) an estimation of disputant flexi-
bility for the subsequent identification of optimal me-
diation strategies to achieve an agreement among par-
ties. This work represents a first initiative towards
smart mediation that could be used as roadmap for
the next generation of intelligent eMediation systems.
ACKNOWLEDGEMENTS
This work has been partially supported by the eJRM
project (ref. PON01 01286).
REFERENCES
Arosio, G., Bagnara, G., Capuano, N., Fersini, E., and Toti,
D. (2013). Ontology-driven Data Acquisition: Intel-
ligent Support to Legal ODR Systems. Frontiers in
Artificial Intelligence and Applications, pages 25–28.
IOS Press.
Bellucci, E. and Zeleznikow, J. (2001). Family winner: A
computerised negotiation support system which ad-
vises upon australian family law. ISDSS2001, pages
74–85.
Bouamrane, M.-M., Rector, A., and Hurrell, M. (2008).
Ontology-driven adaptive medical information collec-
tion system. In Foundations of Intelligent Systems,
pages 574–584. Springer.
Bush, R. A. B. and Folger, J. P. (2004). The promise of
mediation: The transformative approach to conflict.
John Wiley & Sons.
Carneiro, D., Novais, P., Andrade, F., Zeleznikow, J., and
Neves, J. (2014). Online dispute resolution: an arti-
ficial intelligence perspective. Artificial Intelligence
Review, 41(2):211–240.
Danovi, F. and Ferraris, F. (2013). La cultura della medi-
azione e la mediazione come cultura. Giuffr
`
e Editore.
Druckman, D., Druckman, J. N., and Arai, T. (2004). e-
mediation: evaluating the impacts of an electronic me-
diator on negotiating behavior. Group Decision and
Negotiation, 13(6):481–511.
Druckman, D., Ramberg, B., and Harris, R. (2002).
Computer-assisted international negotiation: a tool for
research and practice. Group Decision and negotia-
tion, 11(3):231–256.
Fisher, R., Ury, W. L., and Patton, B. (2011). Getting to yes:
Negotiating agreement without giving in. Penguin.
eMediation-TowardsSmartOnlineDisputeResolution
235
Salton, G. and Buckley, C. (1988). Term-weighting ap-
proaches in automatic text retrieval. Inf. Process.
Manage., 24(5):513–523.
Salton, G., Wong, A., and Yang, C. S. (1975). A vector
space model for automatic indexing. Communications
of the ACM, 18(11):613–620.
Uijttenbroek, E. M., Lodder, A. R., Klein, M. C., Wilde-
boer, G. R., Van Steenbergen, W., Sie, R. L., Huygen,
P. E., and Van Harmelen, F. (2008). Retrieval of case
law to provide layman with information about liabil-
ity: Preliminary results of the best-project. In Com-
putable models of the law, pages 291–311. Springer.
Zeleznikow, J., Meersman, R., Hunter, D., and van
Helvoort, E. (1995). Computer tools for aiding legal
negotiation. In Proc. of the 6th Australasian Confer-
ence on InformationSystems.
Zeleznikow, J. and Stranieri, A. (1995). The split-up sys-
tem: integrating neural networks and rule-based rea-
soning in the legal domain. In ICAIL, pages 185–194.
KMIS2014-InternationalConferenceonKnowledgeManagementandInformationSharing
236
