Professional Documents
Culture Documents
Abstract Organizations are facing the problem of having more projects than
resources to implement them. In this paper, we present a dialogue interaction
module of a framework for a Decision Support System (DSS) to aid in the selection
of public project portfolios. The Interaction module of this DSS is based on mul-
tiple argumentation schemes and dialogue games that not only allow the system to
generate and justify a recommendation. This module is also able to obtain new
information during the dialogue that allows changing the recommendation
according to the Decision Maker’s preferences. Researchers have commonly
addressed the public portfolio selection problem with multicriteria algorithms.
However, in the real life the final selection of the solution depends on the decision
maker (DM). We modeled the reasoning of DM by a Dialogue Corpus. This corpus
is a database, supported by an argument tree that validates the system’s recom-
mendations with the preferences of the DM.
1 Introduction
The decision support systems (DSS) are important tools because they allow users to
have a support on why they reached a decision. According to [1] a DSS is defined
as “an interactive computer-based system or subsystem intended to help decision
makers use communications technologies, data, documents, knowledge and/or
models to identify and solve problems, complete decision process tasks, and make
decisions”. It is a general term for any computer application that improves the
ability of a person or group of persons to make decisions. A DSS helps to retrieve,
summarize and analyze data relevant to the decision.
A Dialogue Interaction Module for a Decision Support System … 743
A DSS for public project selection have different subsystems that work together
to support a decision maker, this article focuses mainly on the subsystem to gen-
erate recommendations in an interactive way.
The definition of the term “interactive”, taken from the Oxford Dictionary is:
“permit a flow of two-way information between a computer and a computer user.”
Furthermore, in the context of computer science, it is defined as the interaction with
a human user, usually in a conversational way to obtain data or commands and
provide immediate results or updates [2].
In this section, the proposed methodology to obtain explanations in public
project portfolios is detailed. Figure 1 shows the complete methodology, ranging
from obtaining information to the calculus of reduced decision rules. An overview
of the process is presented below:
1. First, we need to obtain the preferential information of the organization, these
preferences are necessary for the selection of the adequate multicriteria
algorithm.
2. With the correct selection of the multicriteria algorithm, using the preferential
information, we obtain the set of portfolios to be presented, which is reduced to
a subset in the region of interest of the DM.
Fig. 1 Proposed methodology for recommendation of project portfolio with reduced decision
rules
744 L. Cruz-Reyes et al.
3. In the recommendation subsystem there are stored the possible portfolios for
recommendation to the DM, the selection of one of these portfolios is used to
generate reducts.
4. A transformation of the public portfolio instance to a decision table is made. The
solution generated from the multicriteria method is added at the end of the
original instance to act as a decision attribute, the original attributes will be the
condition attributes of the decision table.
5. This tabular representation is necessary to function with the rough set approach.
The reduct calculation is performed starting with the decision table with the
purpose of generation a reduced decision table using the reducts generated with
the genetic algorithm.
6. By means of an exact method, we work with the newly generated reduced
decision table to get the attributes of the rows decrement (value reduction). The
number of attributes is diminished, rule by rule (row by row), reducing the
quantity of clauses on each rule.
7. Using a small subset of rules on the projects in the portfolio (which projects are
supported and which ones are not), we can summarize the information about the
construction of the portfolio and verify how far is from the organization policy.
This decision is a critical problem the DM is faced with.
8. The process of recommending a justified portfolio can end when such a rec-
ommendation is presented to the DM. At this stage, he can make an intro-
spection when he sees the summary of the characteristics of the projects
conforming the recommended portfolio, and—based on such information-there
is a possibility that the DM is not satisfied with this portfolio. The DM can
update his preferences and make a revaluation of the attributes values of some
projects. The option to perform this updates is a work in progress, this process
must be repeated if the DM changes his preferences, an—hereby—a new jus-
tification (on how the new portfolio is constructed) is obtained.
9. The updated preferential information can be used to restart the process to re-run
the multicriteria algorithm.
The main contribution of this methodology is the reduction of attributes and the
simplification of decision rules by means of a hybrid algorithm, which consist of
two stages:
1. A genetic algorithm to generate reducts, and
2. An exhaustive calculation of the decision rules with an exact method, using the
reduced decision table generated by the genetic algorithm.
The argumentation theory comes into play when the final recommendation is
presented, within a dialog manager the DM and the framework can enter in an
interaction to justify the solution presented to him, if this solution satisfies his base
of preferences.
A Dialogue Interaction Module for a Decision Support System … 745
rules, when given a set of premises a conclusion can be reached [6]. However,
because of the defeasible nature of the arguments, said schemes are not strictly
deductive. The schemes allow the arguments to be represented within a certain
context and take in consideration that the presented reasoning can be modified in
the light of new proof or rule exceptions.
The argumentation schemes are composed of the following elements:
• Premises: Arguments that works as either support or opposition of the
conclusion.
• Conclusion: Statement that is reached from analyzing the premises.
• Critical Questions: Attacks or challenges that weak the argument within the
scheme if they are not answered.
Another important point to consider within the argumentation theory is how the
statements are going to be evaluated, using a proof standard, allows comparing the
arguments made in favor and against a certain statement to evaluate if it holds true
or not [7]. However, every proof standard works in a different way, which means
that while some might consider a statement true others might not; it is important to
be careful when choosing which proof standard to use when evaluating a statement.
The dialogue games (or dialogue systems) essentially define the principle of con-
sistent dialogue and conditions under which a statement made by an individual is
adequate. There are several formal dialogues, taking into account various infor-
mation such as the participants, the communication language, roles of participants,
the aim of the dialogue, etc. These types of dialogues system are generally sustained
by a set of rules:
• Locution Rules (speech acts, movements). The rules indicate which expressions
are allowed. Generally, legal phrases enable the participants to affirm proposi-
tions, allowing others to question or challenge the above statements, and allow
those affirmations that can be claimed, questioned or challenged to justify the
statement. The justifications may involve submitting a proof of the proposition
or an argument for it.
• Commitments Rules. Rules defining the effects of movements in the “commit-
ments”; associated with each player is a compromise to maintain the statements
that the players have made and the challenges they have issued; so there are
rules that define how the commitments are updated.
• Dialogue Rules. Rules to regulate movements. It specifies, for example, all the
acts of speech allowed in a dialogue and types of responses allowed at a certain
state. Various dialogue protocols can be found in the literature, especially for
persuasion [8] and negotiation [9, 10]
A Dialogue Interaction Module for a Decision Support System … 747
• Termination rules. The rules that define the circumstances in which the dialogue
ends.
• Acceptability. In a process of argumentation, it is important to define (or eval-
uate) the status of arguments based on all the ways in which they interact. Thus,
the best or acceptable arguments must be identified at the end of the process of
argumentation.
Most arguments systems are based on the notion of acceptability as identified by
Dung [11], he has proposed an abstract framework for the argument which only
focuses on the definition of the status of the arguments. In this framework, the
acceptability of an argument depends on its membership to some sets, called
acceptable sets or extensions. In other words, the acceptability of the arguments is
defined without considering the internal structure of the arguments.
In this section the modular diagram shown in Fig. 3 will be described in detail.
Load Instance: The system reads a file which provides the system with the
necessary information to generate a recommendation and start a dialogue game
between it and the user, the data that this module requires to permit the system to
continue its process are the following:
• Number of alternatives
• Number of criteria
• Weight of each criterion
• Lexicographic order
• Performance table
• Number of solutions
• Solution matrix
Configuration: The system analyses the information obtained from the previous
process to determinate the initial configuration of elements such as locution rules,
the state transition diagram and which proof standard to use. This will allow the
system to generate an initial recommendation, that lets both it and the user start a
dialogue game.
Dialogue: In this process, the user and the system initialize an exchange of
arguments, supported by the argumentation schemes, in which the system will look
to convince the user to accept the recommendation established by it. However, the
user is also capable of rejecting the current recommendation or to manipulate the
initial configuration to force the system to generate a new recommendation, all
according to the user’s preferences.
Acceptation/Rejection of the Final Recommendation: The user can conclude if
the recommendation received is satisfactory or not to his needs, even if it is the best
possible solution available. This rejection option is established in consideration of
the human factor (the user) that will be in contact with the system. Since he doesn’t
follow a strict set of rules, the user could simply ignore the dialogue following the
initial recommendation and reject it as soon as it is shown, instead of searching for a
better solution.
Based on the work of Walton et al. [12], and Ouerdane [4], some argumentation
schemes were found to be useful for the system for its interaction process with the
user, the schemes presented in this section are formally described in [12].
The system requires defining a proof standard to create an initial recommen-
dation. It also needs to be capable of changing the proof standard on use based on
the information obtained in the interaction with the user. The abductive reasoning is
a process that allows the system to select from the set of proof standards the one that
is closer to the active properties. Said properties are defined both in the configu-
ration process and in the dialogue game. The argumentation scheme for the
abductive reasoning is shown in Table 1.
After choosing a proof standard, the system must generate a recommendation.
Two argumentation schemes have been identified that put the system as a capable
entity for this action. The argument from position to know (see Table 2) and the
argument from an expert opinion (see Table 3).
The argument from position to know is utilized during the initial recommen-
dation and a few states after that, as the system doesn’t have enough information yet
to be considered an expert for the instance that is being analyzed. The argument
A Dialogue Interaction Module for a Decision Support System … 749
from an expert opinion defines the system not only as capable of making a rec-
ommendation of high quality, but also confirms it as an expert of the current
instance as it has obtained enough information.
Several argumentation schemes were found to be useful for the system during
the interaction process that could be used to establish a coherent dialogue with the
user. These schemes allow the system to defend its recommendation or obtain new
information. The use of each of these schemes will be explained below.
750 L. Cruz-Reyes et al.
A dialogue corpus will be located in the dialogue module, it will permit the
characterization of the decision maker, allowing emulating him to make experi-
ments relevant to decision making.
In Fig. 4 the each of the modules for the interaction module can be seen in
greater detail; in the configuration module the premises, locution rules, state tran-
sition diagram and the proof standards are set.
Furthermore, in the dialog module the argumentation schemes and the corpus of
argumentative dialogue interact each other; criteria may be updated and the proofs
standard, according to different conditions that may occur in the dialogue. This
process is performed with the acceptance or rejection of the recommendation.
The architecture shown in Fig. 4 is based on the work of Querdane [4], she
presents a theoretical architecture for a recommendation system, but the imple-
mentation was not realized. The architecture of the recommendation system of this
work will change according to the context of this particular problem.
We present in Fig. 5, the architecture that show the interaction between the
different elements that are involved in the corpus dialogue; the DM preferences will
752 L. Cruz-Reyes et al.
Fig. 4 Second layer of the modular diagram for the interaction module
play a very important role in the final decision; there are other elements that interact
such as PDA, arguments tree, the dialogue corpus, reference sets, work and so on,
they each other in the recommendation system, in order to help in the decision
making.
The proposal in this article covers the generation of argument trees, we are
building this generation based in the Araucaria software [13], and this software has
the distinction of generating an arguments tree with premises and conclusions
The recommendation system will generate an initial solution, this solution will
be verified against the preferences of the DM (these will be based on a PDA, linked
to a reference set), and the arguments corpus, which is a repository of dialogues
that characterize a DM, in order to construct an argument tree.
The recommended solution will be evaluated by the preference and argument
tree to provide an argument as conclusion, which will be determined by the actions
established in the state transition diagram (the user); from here, the interaction will
continue in the dialogue established between the system and the user.
A proposal for the argumentation corpus is shown on Fig. 6, it will be filled with
argumentative texts, then, it generates different argument trees [14]. These argu-
ment trees will characterize the DM, and interact with techniques to extract
information in the corpus, we can obtain a conclusion and generate new data to
construct arguments in the corpus.
Figure 7 shows a resume of the different tasks of the interaction process of the
expert and the simulated DM; the expert will generate a recommendation (solution),
if this solution matches the one with the arguments tree, to conclude if the solution
754 L. Cruz-Reyes et al.
is within the data in the corpus, with that comparison, the artificial DM will gen-
erate an answer according to the state transition diagram and continue with the
interaction.
transition diagram. Finally, the text area located in the central part of the application
shows the dialog between the DM and the Expert.
The architecture introduced has been implemented in the Java programming lan-
guage. The current development has been focused on the initial configuration setup,
recommendation process, proof standard selection and acceptation or rejection of
the final recommendation. We have also worked on the update modules, for
properties and criterion values.
Two of the four main modules have been completed (Load instance and
acceptation/rejection of the final recommendation). However, the remaining mod-
ules (Configuration and Dialogue) are the most code-heavy processes of the system
and require more time to be developed.
Although some of the argumentation schemes and part of the interaction have
been already developed, it is necessary to keep working on those sections to have a
full working system, capable of generating high quality recommendations.
The argumentative dialogue corpus is in the final design, when completed, it will
be incorporated to interact with the argumentation schemes and continue with the
experimentation.
6 Conclusions
Acknowledgments We express our gratitude to CONACYT for partially financing this work.
756 L. Cruz-Reyes et al.
References