<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta />
    <article-meta>
      <title-group>
        <article-title>Causes for Changing Profiles (Preliminary Report) ⋆</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <string-name>Eduardo Fermé</string-name>
          <xref ref-type="aff" rid="aff1">1</xref>
          <xref ref-type="aff" rid="aff2">2</xref>
        </contrib>
        <contrib contrib-type="author">
          <string-name>Marco Garapa</string-name>
          <xref ref-type="aff" rid="aff0">0</xref>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <contrib contrib-type="author">
          <string-name>Maurício D. L. Reis</string-name>
          <xref ref-type="aff" rid="aff0">0</xref>
          <xref ref-type="aff" rid="aff1">1</xref>
        </contrib>
        <aff id="aff0">
          <label>0</label>
          <institution>CIMA - Centro de Investigação em Matemática e Aplicações</institution>
        </aff>
        <aff id="aff1">
          <label>1</label>
          <institution>Faculdade de Ciências Exatas e da Engenharia, Universidade da Madeira</institution>
        </aff>
        <aff id="aff2">
          <label>2</label>
          <institution>NOVA Laboratory for Computer Science and Informatics, NOVA LINCS</institution>
        </aff>
      </contrib-group>
      <fpage>6</fpage>
      <lpage>15</lpage>
      <abstract>
        <p>User profiles are an essential Knowledge Representation tool in several areas of information technology. In a recent paper, Fermé et al. presented a formal framework for representing user profiles and profile revision operators defined through a Knowledge-Driven perspective. In this paper, we analyse the possibility of going from one given user profile to another by means of a profile revision operator. More precisely, given two profiles  and  we present some conditions which ensure that there is a profile revision operator ⊙ on  and a sentence  such that  ⊙  = . Furthermore, considering a fixed operator ⊙ , we characterize the change formulas  which are such that  ⊙  = , by identifying upper and lower bounds for their sets of models. Analogous results are obtained for the case of a “system of equations"  ⊙   =  for every  ∈ {1, . . . , }. Furthermore, a similar study is carried out considering profile revision operators defined on sets of profiles (which take sets of profiles to sets profiles rather than a single profile to a single profile).</p>
      </abstract>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>1. Introduction</title>
      <p>
        The study of user profiles and their dynamics over time
has gained increasing attention in the field of information
technology. User profiles are widely used in various areas,
including recommendation systems [
        <xref ref-type="bibr" rid="ref1 ref2">1, 2</xref>
        ], adaptable user
interfaces [
        <xref ref-type="bibr" rid="ref3">3</xref>
        ], personalized systems [
        <xref ref-type="bibr" rid="ref4">4</xref>
        ], and cognitive or
physical rehabilitation systems [
        <xref ref-type="bibr" rid="ref5">5</xref>
        ]. In [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ], a formalization
of the creation, representation, and dynamics of profiles
from a Knowledge-Driven perspective is proposed. The
paper introduces a formal profile representation structure,
using a formal language that enables clear representation
of user profiles and their attributes, as well as the
properties of different types of profile operations. The paper also
presents several dynamic operators for profiles, including
two profile revision operators, which are described below:
(a) From one single profile to one single profile: in
this case, the model represents the evolution of
a single profile. This evolution is caused by an
external stimulus, for example an interaction with
a system, a training program, etc.
(b) From a set of profiles to a set of profiles: these
operators model the changes produced by an input in
a collection of profiles and return the new profiles
resulting from this process. This type of operators
is particularly useful in two contexts: (i) to
capture the change produced by a single event in a
population (set of profiles) (ii) to analyze changes
when a single individual is represented by a set of
possible profiles due to lack of information. For
example, if we only know that a person’s age is
between 18 and 20 years old, we can represent
that person using a set of three profiles: one with
age 18, one with age 19, and one with age 20.
      </p>
      <p>
        In this paper, we expand the work presented in [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ] by
investigating the following aspects related to the operators
discussed above: (i) the possibility of always transitioning
from one given user profile to another; (ii) the bounds of
the models of a change formula (in terms of set inclusion)
that causes a change from a given set of user profiles
into another; (iii) the existence of a change formula that
is solution of a given “system of equations" of the form
Γ  ⊙   = Φ , where Γ  and Φ  are the initial and final
sets of profiles of an agent, and  is the change formula,
and, if such a solution exists, we determine the bounds for
the set of models of that solution. Solving a system of this
type could be valuable, for instance, in scenarios where
one aims to train a group of agents (each represented by a
profile or a set of profiles) in order to alter them towards
achieving a specific objective.
      </p>
      <p>The rest of the paper is organized as follows: In
Section 2, we introduce the notations and recall the main
background concepts that will be needed throughout this
article. In Section 3 we conduct the study mentioned
above. In Section 4 we briefly mention other works
related to the one presented in this paper. In Section 5 we
summarize the main results obtained and point out some
topics for future research.</p>
    </sec>
    <sec id="sec-2">
      <title>2. Background</title>
      <p>
        Definition 1. [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ] Let L =≪ 1, 2, ...,  ≫ be a tuple
of labels. For each  ∈ {1, . . . , } let  be a finite set
In this section, we present some concepts and notations associated with label , that we will designate by the
that will be used throughout this article. domain of . A profile , associated with L, denoted by
L (or simply by  if the tuple of labels is clear from
2.1. Formal Preliminaries the context), is an element of 1 × 2 × ... × . The
set of all profiles associated with L will be denoted by
We assume that the empty disjunction is a contradiction. PL (or simply by P if the tuple of labels is clear from the
We shall use the symbol ⊥ to represent a contradiction. context).2
Given a set , we will denote by () the power set of
, i.e. the set of all subsets of . Given a set  a binary Example 1. Given the tuple of labels L =≪ age,
relation ⪯ on  is: gender, civil status, nationality≫ ;  =
- reflexive if and only if  ⪯  for all  ∈ ; {0, 1, 2, ..., 150},  = {male, female,
- transitive if and only if it holds that if  ⪯  and  ⪯  , other},   = {single, married,
dithen  ⪯  , for all , ,  ∈ ; vorced, widowed, other},  = {English,
- antisymmetric if and only if it holds that if  ⪯  and Portuguese, Argentinian}, the following are
ex ⪯  , then  =  , for all ,  ∈ . amples of profiles: ⟨25, , , ℎ⟩,
- total if and only if it holds that either  ⪯  or  ⪯  ⟨45,  , ,  ⟩.
for all ,  ∈ .
      </p>
      <p>
        To express properties of a profile (and therefore the
possibility of changing it), we need to define a formal
language:
Definition 2. [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ] Given a tuple of labels L =≪
1, 2, ...,  ≫ , for each  ∈ {1, ..., }, let  be
the domain associated with the label . The alphabet of
symbols of the language ℒL (or simply ℒ) associated with
      </p>
      <p>L that we will consider is:</p>
      <p>A non-strict pre-order is a reflexive and transitive
relation. A non-strict order is a non-strict pre-order which is
also antisymmetric. Given a non-strict pre-order ⪯ on a
set , the associated strict part ≺ is defined by  ≺  if
and only if  ⪯  and  ̸⪯  , for all ,  ∈ .  ⋍ 
will be used to denote that  ⪯  and  ⪯  .</p>
      <p>Given a set  a strict order ≺ on  is an irreflexive
(i.e,  ̸≺  , for all  ∈ ) and transitive relation on .1
A strict order is total if and only if it holds that either
 ≺  ,  ≺  or  =  for all ,  ∈ . Throughout
the text, whenever we refer to pre-orders (respectively,
orders) without mentioning whether or not they are strict,
we will be referring to non-strict pre-orders (respectively,
non-strict orders).</p>
      <p>Let A be a set and Γ be a finite subset of A. Given a
total strict order ≺ on A, the minimum of Γ with respect
to ≺ is denoted by (Γ , ≺ ) and is defined as follows:</p>
      <p>= (Γ , ≺ ) iff
 ∈ Γ and  ≺  for all  ∈ Γ ∖ { }.</p>
      <p>Given a pre-order ⪯ on A, the set of minimal elements
of Γ with respect to ⪯ is denoted by  (Γ , ⪯ ) and is
defined as follows:
 (Γ , ⪯ ) = { ∈ Γ :  ̸≺ , for all  ∈ Γ }.</p>
      <sec id="sec-2-1">
        <title>We note that if ⪯ is a total pre-order, then</title>
        <p>(Γ , ⪯ ) = { ∈ Γ :  ⪯ , for all  ∈ Γ }.</p>
        <sec id="sec-2-1-1">
          <title>2.2. Profiles Definition</title>
          <p>In this subsection, we present the formal definition of a
profile and introduce appropriate language and semantics
for formalizing the dynamics of profiles.
1Every irreflexive and transitive relation on a set  is also
antisymmetric.
1. 1, 2, ...,  (labels);
2. = (symbol of equality)
3. (,) (punctuation symbols);</p>
          <p>4. , , ... (elements of ⋃︀ );</p>
          <p>=1
5. ⊥ (symbol of contradiction);
6. ¬, ∧, ∨, →, ↔ (symbols of connectives).</p>
          <p>
            Definition 3. [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] Let L =≪ 1, 2, ...,  ≫ be a tuple
of labels. For each  ∈ {1, ..., }, let  be the domain
associated with the label .
          </p>
          <p>An atomic formula in ℒL is defined by: if  is a label
occurring in L and  ∈ , then  =  is an atomic
formula of ℒL.</p>
          <p>A well-formed formula (wff) of ℒL is defined by:
1. Every atomic formula of ℒL is a wff of ℒL.
2. If  and  are wffs of ℒL, so are (¬), ( ∧ ),
( ∨ ), ( → ) and ( ↔ ).</p>
          <p>
            In the following definition we recall from [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] the
semantics for profile dynamics.
          </p>
          <p>
            Definition 4. [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] Let L =≪ 1, 2, ...,  ≫ be a
set of labels. For each  ∈ {1, ..., }, let  be the
domain associated with the label . A profile  =
⟨1, 2, ..., ⟩ is said to satisfy a formula  , denoted by
 |=  , if it can be shown inductively to do so under the
following conditions:
2Note that PL is finite.
1.  |= ( = ) iff  = ;
2.  |= (¬ ) iff  ̸|=  ;
3.  |= ( ∧  ) iff  |=  and  |=  ;
4.  |= ( ∨  ) iff  |=  or  |=  ;
5.  |= (
6.  |= (
→  ) iff  ̸|=  or  |=  ;
↔  ) iff ( |=  iff  |=  ).
denote that  is a tautology.
‖ ⊥ ‖ = ∅
          </p>
          <p>We say that  is a model of  if and only if  |=  .
The set of models of  is denoted by ‖ ‖. It holds that
. A set of profiles Γ is said to satisfy  if and
only if every profile in Γ is a model of  . We say that  is
a tautology if and only if ‖ ‖ = PL. We will use |=  to</p>
          <p>The following definition introduces the notion of Γ
faithful pre-order.
Γ -faithful if it satisfies:
Definition 5.</p>
          <p>
            [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] Let L be a tuple of labels and Γ be a
non-empty subset of PL. A binary relation ⪯ Γ on PL is
1. If  ∈ Γ and  ∈ Γ , then  ≺ Γ  does not
hold.
          </p>
          <p>2. If  ∈ Γ and  ∈ PL ∖ Γ , then  ≺ Γ  .</p>
          <p>If Γ is a singleton, say { }, then we will omit the
braces in the subscript of the binary relation mentioned in
the above definition by writing ⪯  instead of ⪯
will also write  -faithful instead of { }-faithful. Note
{ }. We
that if Γ =</p>
          <p>{ }, and ≺  is a strict order on PL, then the
ifrst condition of Definition 5 follows trivially, since
is irreflexive, and the second condition can be rewritten
≺ 
as  ≺   for all  ∈ PL ∖ { }.
2.2.1. Model 1. From One Profile to One Profile
In this subsection we present the first model for profile
dynamics. In this model, we revise a profile by a formula
of the language obtaining as output a profile.</p>
          <p>Definition 6.</p>
          <p>
            [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] Let L =
≪
 ⊙  =  ⊙ ≺  .
≺  on PL defined as follows:
≪
1, 2 ≫
and 1 = {, } and
2. There are six profiles associated with
2 = {1, 2, 3} be the domains of, respectively, 1 and
L. Let  be
the profile
          </p>
          <p>⟨, 1⟩. Consider the P-faithful total strict order
⟨, 1⟩ ≺  ⟨, 1⟩ ≺  ⟨, 2⟩ ≺  ⟨, 2⟩ ≺  ⟨, 3⟩ ≺  ⟨, 3⟩.
Let  be the sentence 1 =  and  be the sentence 2 =
2. It holds that  ⊙ ≺   = ⟨, 1⟩ and  ⊙ ≺   = ⟨, 2⟩.
2.2.2. Model 2. From a Set of Profiles to a Set of</p>
          <p>Profiles
We now present a model which addresses the problem of
revising a set of profiles. A non-singular set of profiles
of a given agent provides only partial information about
that agent. For example, the following set of profiles
{⟨25, , ⟩, ⟨25, , ⟩}
that the agent is a 25 years old man but we do not know if
expresses
he is single or married.</p>
          <p>Definition 7.</p>
          <p>
            [
            <xref ref-type="bibr" rid="ref6">6</xref>
            ] Let L =
≪
          </p>
          <p>1, 2, ...,  ≫
such that for all sentences  :
of labels and let Γ be a non-empty subset of PL. Let ⪯ Γ be
a Γ -faithful binary relation on PL. The ⪯ Γ-based SPtoSP
profile revision on Γ is the operator ⊙ ⪯ Γ : ℒL →  (PL)
be a tuple</p>
        </sec>
      </sec>
    </sec>
    <sec id="sec-3">
      <title>3. Profiles Dynamics</title>
      <p>In this section we present definitions and axiomatic
characterizations of some operators of profile revision. Given
a profile  (respectively, a set of profiles Γ ) we shall
designate by profile revision operator (or simply by profile
regarding those operators:
each sentence  associates a profile (respectively, a set of
profiles), which we shall denote by  ⊙  (respectively
Γ ⊙  ). Then, we find answers to the following questions
formula  which confirms it?</p>
      <p>(i) Does it hold that, for any two user profiles 1 and
2 (respectively, sets of profiles Γ 1 and Γ 2), there exists
a profile revision operator ⊙ and a formula  such that
1 ⊙ 
= 2 (respectively, Γ 1 ⊙</p>
      <p>= Γ 2)?
(ii) When 1 and 2 (respectively, Γ 1 and Γ 2) are such
that the answer to question (i) above is positive for a
certain (fixed) operator ⊙ , how can we obtain a change
for all sentences  :
revision induced by ≺  is the operator ⊙ ≺  such that revision) on  (respectively, on Γ ) a function ⊙
that to
(iii) Given profiles 1, . . . , , 1, . . . , 
(respectively, sets of profiles</p>
      <p>Γ 1, . . . , Γ , Φ 1, . . . , Φ ), are
there profile revision operators ⊙ 1, . . . , ⊙  and a
sentence  such that for each  ∈ {1, . . . , }, it holds that
 ⊙  
(iv)
=  (respectively, Γ  ⊙</p>
      <p>= Φ )?</p>
      <p>When 1, . . . , , 1, . . . ,  (respectively,
which confirms it?
Γ 1, . . . , Γ , Φ 1, . . . , Φ ) are such that the answer to
question (iii) above is positive for a certain (fixed)
operators ⊙ 1, . . . , ⊙ , how can we obtain a change formula</p>
      <sec id="sec-3-1">
        <title>3.1. Model 1. From One Profile to One</title>
      </sec>
      <sec id="sec-3-2">
        <title>Profile</title>
        <p>
          The following postulates, which are based on the modified
version of the AGM revision postulates and the update
will be useful towards the axiomatic characterization of
the   profile revision operators.
postulates proposed by Katsuno and Mendelzon [
          <xref ref-type="bibr" rid="ref7 ref8">7, 8</xref>
          ], ⊙ : ℒL → PL be a profile revision operator on  that
(P1) If ‖ ‖ ̸= ∅, then  ⊙  |=  .
(P2) If ‖ ‖ = ∅, then  ⊙  =  .
(P3) If  |=  , then  ⊙  =  .
(P4) If ‖ ‖ = ‖ ‖, then  ⊙
        </p>
        <p>=  ⊙  .
(P5) If  ⊙  |=  and  ⊙  |=  , then  ⊙  =  ⊙  . ‖ ‖ ⊆ PL ∖ { }.
(P6)  ⊙ ( ∨  ) =  ⊙  or  ⊙ ( ∨  ) =  ⊙  .</p>
        <p>Postulate P1 states that any consistent new information
will be satisfied by the outcome.</p>
        <p>P2 asserts that if the new
P3 states that if the input is already inferred from the
information is contradictory, then no change is performed. (P2) and (P3).
starting profile, then no change occurs.</p>
        <p>P4 expresses
the principle of irrelevance of the syntax. Postulate P5
states that if the outcome of revising a profile</p>
        <p>
          by 
implies  and the outcome of revising that profile by 
implies  , then the outcomes of those two revisions are
identical. This condition appears as (U6) in KM-update
[
          <xref ref-type="bibr" rid="ref8">8</xref>
          ], as condition (C7) in [
          <xref ref-type="bibr" rid="ref9">9</xref>
          ] in a belief revision context
and as a conditional logic axiom (CSO) in [
          <xref ref-type="bibr" rid="ref10">10</xref>
          ]. Finally,
P6 states that in the case of a revision by a disjunction,
one of the disjoints will be preferred in the outcome.
        </p>
        <p>We now present an axiomatic characterization for
  profile revision operators.</p>
        <p>
          Observation 1. [
          <xref ref-type="bibr" rid="ref6">6</xref>
          ] Let L =
≪
Observation 2. Let L =
of labels and let  ∈ PL. Let ⊙
: ℒL → PL be a profile
revision operator on  that satisfies (P1), (P2) and (P3).
 ∈ ‖ ‖.
        </p>
        <p>It holds that  ⊙</p>
        <p>=  if and only if ‖ ‖ = ∅ or
(P2) and (P3).</p>
        <p>Proof. Assume that it holds that ‖ ‖ ̸= ∅. Hence by (P1)
it follows that  ∈ ‖ ‖. The other direction follows by</p>
        <p>The following observation establishes a necessary
condition, regarding the models of  , to ensure that  ⊙  =
, where  ̸=  and ⊙ is an operator, on  , that satisfies
(P1), (P2) and (P3).</p>
        <p>Observation 3. Let L =
1, 2, ...,  ≫</p>
        <p>be a tuple
of labels and let ,  ∈ PL be distinct profiles. Let
‖ ‖ ⊆ PL ∖ { }.
satisfies (P1), (P2) and (P3). If  ⊙</p>
        <p>= , then  ∈
Proof. If ‖ ‖ = ∅, then by (P2) it follows that  = .</p>
        <p>Contradiction. Hence ‖ ‖ ̸= ∅. By (P1) it follows that
 ∈ ‖
 ‖
. Let  ∈ ‖
 ‖</p>
        <p>. If  =  , then by (P3) it
follows that  =  . Contradiction. Hence  ̸=  . Thus</p>
        <p>The following corollary establishes a necessary
condition, regarding the models of  , to ensure that  is a
solution of the system of equations of the form  ⊙  
where for each  ∈ {1, . . . , },  and  are distinct
profiles and ⊙  is an operator, on , that satisfies (P1),</p>
        <p>= ,
Corollary 1. Let L =
≪
labels and let {1, . . . , }∪{1, . . . , } ⊆
each  ∈ {1, . . . , } let ⊙  : ℒL → PL be profile
revision operators on  that satisfy (P1), (P2) and (P3). If for</p>
        <p>PL. For
all  ∈ {1, . . . , } it holds that  ⊙</p>
        <p>=  and  ̸=
, then {1, . . . , } ⊆ ‖  ‖ ⊆ PL ∖ {1, . . . , }.
be a relation on PL such that, for all ,  ∈ {1, . . . , } it
{ ∈ PL :  ≺  }.</p>
        <p>(SP1) If ‖ ‖ ̸= ∅, then Γ ⊙  ⊆ ‖  ‖.</p>
        <p>Proof. Let Ω = { ∈ PL :  ≺  }. (SP2) If ‖ ‖ = ∅, then Γ ⊙  = Γ .
(⇒) Assume that  ⊙ ≺   = . It holds that  ̸=  (SP3) Γ ⊙  ̸= ∅.
and ⊙ ≺  is a PtoP profile revision operator on  . Thus (SP4) If ‖ ‖ = ‖ ‖, then Γ ⊙  = Γ ⊙  .
‖ ‖ ̸= ∅. Hence (‖ ‖, ≺  ) = . Thus  ∈ ‖ ‖. (SP5) If Γ ∩ ‖ ‖ ̸= ∅ then Γ ⊙  = Γ ∩ ‖ ‖.
Let  ∈ ‖ ‖. We will consider two cases: (SP6) Γ ⊙ ( ∨  ) = Γ ⊙  or Γ ⊙ ( ∨  ) = Γ ⊙ 
Case 1)  = . Thus  ∈ Ω ∪ {}. or Γ ⊙ ( ∨  ) = Γ ⊙  ∪ Γ ⊙  .
Case 2)  ̸= . Thus  ∈ ‖ ‖ and  ̸=  = (SP7) Γ ⊙ ( ∨  ) ⊆ Γ ⊙  ∪ Γ ⊙  .
(‖ ‖, ≺  ). Thus  ≺  . Hence  ∈ Ω . (SP8) If Γ ⊙  |=  and Γ ⊙  |=  , then Γ ⊙  = Γ ⊙  .
In both cases it holds that ‖ ‖ ⊆ Ω ∪ {}.
(⇐) Let  ∈ ‖ ‖ ⊆ Ω ∪ {}. Hence ‖ ‖ ̸= ∅. We will (SP9) Γ ⊙  ∩ Γ ⊙  ⊆ Γ ⊙ ( ∨  ).
consider two cases:
Case 1) ‖ ‖ = {}, then by (P1) it follows that  ⊙ ≺  (SP1), (SP2), (SP4) and (SP8) are the adapted versions
 = . of (P1), (P2), (P4) and (P5) respectively. (SP3) states
Case 2) ‖ ‖ ̸= {}. Hence {} ⊂ ‖  ‖. It holds that that the outcome is always consistent. (SP5) corresponds
‖ ‖ ⊆ Ω ∪ {}. Let  ∈ ‖ ‖ ∖ {}. Hence  ∈ Ω . to the AGM revision postulate vacuity and states that
Hence  ≺  . Thus  = (‖ ‖, ≺  ). Therefore if there are profiles in Γ that satisfy the input sentence
 ⊙ ≺   = .  , then the output of the revision of Γ by  is the set
formed by those profiles. The intuition behind (SP6) is</p>
        <p>The following corollary presents a result similar to that if we wish to revise by a disjunction and there is
that illustrated in Observation 5, but where a system of some preference between the disjuncts, then this revision
equations is considered. is equivalent to revising by the preferred disjunct. In the
Corollary 3. Let L =≪ 1, 2, ...,  ≫ be a tuple of case of indifference, revising by the disjunction returns
labels and let {1, . . . , } ∪ {1, . . . , } ⊆ PL be the set of profiles consisting of the union of the revision
such that  ̸=  for all  ∈ {1, . . . , }. by each of the disjuncts. (SP7) states that if the revision</p>
        <p>
          For each  ∈ {1, . . . , } let ⊙ ≺  be a PtoP profile of a set of profiles Γ by  ∨  leads to the acceptance of
revision operator on . It holds that, a profile , then the same should happen when revising
{1, . . . , } ⊆ ‖  ‖ ⊆

⋂︀ (Ω  ∪ {}).
=1
Observation 6. [
          <xref ref-type="bibr" rid="ref6">6</xref>
          ] Let L =≪ 1, 2, ...,  ≫ be a
tuple of labels, Γ be a non-empty subset of PL and ⊙ be a
profile revision operator on Γ . The following conditions
are equivalent:
        </p>
        <p>The following observation states that when revising a
set of user profiles Γ by a formula  , through an operator
that satisfies some of the postulates introduced above, the
set of user profiles stays unchanged iff  has no models
or contains all the profiles in Γ .</p>
        <p>The following observation states the bounds (in terms
of set inclusion) of the models of a change formula  in
order to obtain a given set of profiles Φ as output of a
  revision of a set profiles Γ by  .</p>
        <p>Observation 10. Let L =≪ 1, 2, ...,  ≫ be a
tuple of labels. Let Γ be a non-empty subset of PL. Let
⊙ : ℒL →  (PL) be a profile revision operator on Γ
that satisfies (SP1), (SP2) and (SP5). If Γ ⊙  = Φ and
Φ ̸⊆ Γ , then Φ ⊆ ‖  ‖ ⊆ PL ∖ Γ .</p>
        <p>Proof. Assume that ⊙ satisfies (SP1), (SP2) and (SP5)
and Γ ⊙  = Φ and Φ ̸⊆ Γ . By (SP2) it follows that
‖ ‖ ̸= ∅. Thus by (SP1) it yields that Φ ⊆ ‖  ‖. On
the other hand, it holds that Γ ⊙  ̸= Γ ∩ ‖ ‖. Thus, by
(SP5), Γ ∩ ‖ ‖ = ∅. Therefore, Φ ⊆ ‖  ‖ ⊆ PL ∖ Γ .</p>
        <p>It follows from the above observation that, if its
preconditions are held, then ‖ ‖ is unique whenever Φ = PL ∖Γ .</p>
        <p>The following example illustrates the usefulness of the
result of the above observation.</p>
        <p>Example 3. Let PL = {1, 2, 3, 4, 5}, Γ =
{2, 3}, Φ 1 = {1, 4} and Φ 2 = {1, 2}. Hence
PL ∖ Γ = {1, 4, 5}. It holds that Φ 1 ̸⊆ Γ and</p>
        <p>The following three corollaries follow from
Observation 10.</p>
        <p>Corollary 4 states that if Γ and Φ are two nonempty
sets of profiles such that Φ is not contained in Γ , and Φ is
obtained from Γ when revising it by  through a SPtoSP
profile revision operator, then Φ and Γ are disjoint sets of
profiles.</p>
        <p>Proof. (⇐) If ‖ ‖ = ∅, then by (SP2) it follows that
Γ ⊙  = Γ . Assume now that ‖ ‖ ̸= ∅ and Γ ⊆ ‖  ‖.</p>
        <p>Then, Γ ∩ ‖ ‖ = Γ ̸= ∅. Thus, by (SP5), it follows that
Γ ⊙  = Γ .
(⇒) Assume that Γ ⊙  = Γ . Consider that ‖ ‖ ̸= ∅.</p>
        <p>Hence, by (SP1), it follows that Γ ⊙  ⊆ ‖  ‖. Thus
Γ ⊆ ‖  ‖.</p>
        <p>Observation 8. Let L =≪ 1, 2, ...,  ≫ be a tuple
of labels and let Γ be a non-empty subset of PL. Let Φ 2 ̸⊆ Γ .
⊙ : ℒL →  (PL) be a profile revision operator on Γ that If Γ ⊙  = Φ 1, then {1, 4} ⊆ ‖  ‖ ⊆ { 1, 4, 5}.
satisfies (SP1), (SP2) and (SP5). It holds that Γ ⊙  = Γ Thus either ‖ ‖ = {1, 4} or ‖ ‖ = {1, 4, 5}.
if and only if ‖ ‖ = ∅ or Γ ⊆ ‖  ‖. There is no sentence  such that Γ ⊙  = Φ 2 since,
although Φ 2 ̸⊆ Γ , it cannot hold that {1, 2} ⊆ ‖  ‖ ⊆
{1, 4, 5}.</p>
        <p>Given a set of profiles Γ , the following observation
illustrates the properties that a change formula  must
satisfy in order for it to hold that Γ ⊙  = Φ , for a given
set of profiles Φ ⊂ Γ .</p>
        <p>Corollary 4. Let L =≪ 1, 2, ...,  ≫ be a tuple
of labels. Let Γ and Φ be non-empty subsets of PL such
that Φ ̸⊆ Γ . Let ⊙ : ℒL →  (PL) be a profile revision
Observation 9. Let L =≪ 1, 2, ...,  ≫ be a tuple operator on Γ that satisfies (SP1), (SP2) and (SP5). If
of labels and let Γ be a non-empty subset of PL, and Γ ⊙  = Φ , then Φ ∩ Γ = ∅.
Φ ⊂ Γ . Let ⊙ : ℒL →  (PL) be a profile revision
operator on Γ that satisfies (SP1), (SP2), (SP3) and (SP5).</p>
        <p>The following corollary establishes upper and lower
It holds that Γ ⊙  = Φ if and only if Γ ∩ ‖ ‖ = Φ and bounds for the set of models of formulas that are solutions
Φ ̸= ∅.
of a “system of equations" of the form Γ  ⊙  
where for each , Φ  ̸⊆ Γ  ̸= ∅ and ⊙  satisfies (SP1),
shown above).</p>
        <p>= Φ , that  ⪯ Γ  if  =  (since ⪯ Γ is reflexive, as

=1
labels. For each  ∈ {1, . . . , } let Γ  be a non-empty
subset of PL and ⊙  : ℒL →  (PL) be a profile revision
operator on Γ  that satisfies (SP1), (SP2) and (SP5). If,
for all  ∈ {1, . . . , }, it holds that Γ  ⊙   = Φ  and
Φ  ̸⊆ Γ , then ⋃︀ Φ  ⊆ ‖  ‖ ⊆ PL ∖ ⋃︀ Γ .</p>
        <p>=1</p>
        <p>We will now prove that ⪯ Γ is transitive. Let ,  ,  ∈
PL be such that  ⪯ Γ  and  ⪯ Γ  we intend to
prove that  ⪯ Γ .</p>
        <p>It follows trivially if  =  or  = . It also follows
Where ⪯ Γ is such that  ⪯ Γ  iff either: (i)  ∈ Γ or follows that  ̸∈  (‖ ‖, ≺ Γ). Contradiction.</p>
        <sec id="sec-3-2-1">
          <title>3Note that ⪯ Γ is total.</title>
          <p>Case 1)  ∈ Φ .</p>
          <p>Case 1.1)  ∈ Γ . Hence Φ
Therefore  ∈ Γ . Hence, by definition of ⪯ Γ, it follows
that  ⪯ Γ .</p>
          <p>Case 1.2)  ̸∈ Γ . Hence, by definition of ⪯ Γ, it follows
Case 2)  ̸∈ Φ . Then  ∈ PL ∖ Γ . Hence  ̸∈ Γ ∪ Φ .</p>
          <p>Thus, by definition of ⪯ Γ, it follows that  ⪯ Γ .</p>
          <p>Thus in all cases it follows that  ⪯ Γ . Hence  ̸≺ Γ
 . It yields that  ∈ ‖ ‖. Thus Φ ⊆</p>
          <p>(‖ ‖, ≺ Γ).</p>
          <p>Let  ∈  (‖ ‖, ≺ Γ) and assume towards a
contradiction that  ̸∈ Φ . Let  ∈ Φ ⊆ ‖  ‖. From  ∈
 (‖ ‖, ≺ Γ) it follows that  ∈ ‖ ‖ ⊆ Φ</p>
          <p>∪ PL ∖ Γ .</p>
          <p>Thus  ̸∈ Γ ∪ Φ . Therefore,  ⪯ Γ . Furthermore it
holds that  ̸⪯ Γ  . Hence  ≺ Γ . From which it
∩ Γ ̸
= ∅</p>
          <p>. Thus Φ ⊆ Γ .</p>
          <p>The following example clariefis the importance of the
condition “either Γ
∩ Φ =</p>
          <p>∅ or Γ ⊆ Φ " included in the
hypothesis of the statement of the previous observation.
and Φ =
Φ ∪ PL ∖
Example 4. Let PL = {1, 2, 3, 4}, Γ =</p>
          <p>{1, 2}
{1, 3}. Hence Γ ∩ Φ ̸</p>
          <p>= ∅ and Φ ̸⊆ Γ . Let
 be a formula such that ‖ ‖ = Φ . Hence Φ ⊆ ‖  ‖ ⊆
Γ . However, since Γ</p>
          <p>∩ ‖ ‖ ̸= ∅ it follows by
Γ ⊙  = Γ ∩ ‖ ‖ = {1} ̸= Φ .
(SP5) that for any SPtoSP profile revision operator ⊙ ,</p>
          <p>Given sets of profiles Γ 1, . . . , Γ , Φ 1, . . . , Φ , the
following corollary, which follows directly from Observation
∪ Φ . Hence  ⪯ Γ .
11, establishes the bounds, in terms of set inclusion, for
the models of  in order to ensure the existence of a SP- be such that ‖ 2‖ = Φ . It holds that (PL ∖Φ)
∩Φ =
∅ and
toSP revision operator ⊙  on Γ  such that Γ  ⊙   = Φ  Φ ⊆ ‖  2‖ ⊆ Φ
∪ PL ∖ (PL ∖ Φ) . Hence by Observation
=1
∅ orΦ  ⊆ Γ .</p>
          <p>If ⋃︀ Φ  ⊆ ‖  ‖ ⊆

=1
for all  ∈ 1, . . . , .</p>
          <p>Corollary 7. Let L =
≪
of labels. For all  ∈ {1, . . . , } let Γ  and Φ  be
nonempty subsets of PL such that it holds that either Γ ∩Φ  =</p>
          <p>= Φ . However, as the following observation asserts,
it is always possible, by means of a procedure of (at most)
two-steps to transform a non-empty set of profiles Γ into
any non-empty set of profiles Φ ̸= PL.
Φ ̸= PL.4 It holds that either:
Observation 12. Let L =
≪
Γ , therefore it is not possible to apply ⊙ 1 to the
in general, ⊙ 1 ̸= ⊙ 2.
outcome of Γ ⊙ 1  (unless Γ = Γ</p>
          <p>⊙ 1  ). Hence,
Proof. Case 1) Γ ∩ Φ =</p>
          <p>∅ or Φ ⊆ Γ . Let  be such that
Φ ⊆ ‖  ‖ ⊆ Φ ∪ PL ∖ Γ . Then according to Observation
11 there exists a SPtoSP profile revision operator ⊙ such
that Γ ⊙</p>
          <p>= Φ .
cases:
Case 2) Γ ∩ Φ ̸</p>
          <p>= ∅ and Φ ̸⊆ Γ . By hypothesis it holds
that Φ ̸</p>
          <p>= PL, Γ ̸= ∅ and Φ ̸= ∅. We will consider two
Φ ̸⊆ Γ ).</p>
          <p>Case 2.2) Γ ∖ Φ =
Case 2.1) Γ ∖Φ ̸= ∅. Let  1 be such that ‖ 1‖ = Γ ∖Φ . It
holds that Γ ∖
Φ ⊆ Γ and Γ ∖
Φ ⊆ ‖  1‖ ⊆</p>
          <p>(Γ ∖Φ)
Hence by Observation 11 there exists a SPtoSP profile
revision operator ⊙ 1 such that Γ ⊙ 1  1 = Γ ∖
be such that ‖ 2‖ = Φ . It holds that Γ ∖</p>
          <p>Φ ∩ Φ =
Γ ⊙ 1  1) such that (Γ ⊙ 1  1) ⊙ 2  2 = Φ .
Φ ⊆ ‖  2‖ ⊆ Φ</p>
          <p>∪ PL ∖ (Γ ∖ Φ) . Hence by Observation
11 there exists a SPtoSP profile revision operator ⊙ 2 (on
Φ . Let  2</p>
          <p>∅ and
∅. Hence Γ ⊆ Φ . Thus Γ ⊂ Φ (since
Let  1 be such that ‖ 1‖ = PL ∖Φ . It holds that PL ∖Φ ̸=
∅, Γ ∩PL ∖Φ =
∅ and PL ∖Φ ⊆ ‖  1‖ ⊆
(PL ∖Φ)</p>
          <p>∪PL ∖Γ .</p>
          <p>Hence by Observation 11 there exists a SPtoSP profile
4If Φ = PL and Γ = PL the result also holds.
Γ ⊙ 1  1) such that (Γ ⊙ 1  1) ⊙ 2  2 = Φ .
11 there exists a SPtoSP profile revision operator ⊙ 2 (on</p>
          <p>Given two sets of profiles Γ and Φ , and a  
revision operator ⊙</p>
          <p>on Γ induced by a specific (known)
that:
pre-order ⪯ , the following observation establishes the
upper and lower bounds for the set of models of a change
formula  for it to hold that Γ ⊙  = Φ .</p>
          <p>Observation 13. Let L =
≪
∪PL ∖Γ .  ∈ Φ ⊆ ‖  ‖ such that  ≺ Γ  . Contradiction since
we need to prove that Φ =
(⇐) From Φ ̸
= ∅ it follows that ‖ ‖ ̸= ∅</p>
          <p>. Hence,
 (‖ ‖, ⪯ Γ). Let  ∈
 (‖ ‖, ⪯ Γ). It holds that  ∈ ‖ ‖ ⊆ Ω</p>
          <p>∪ Φ .
Suppose towards a contradiction that  ̸∈ Φ . Hence  ∈ Ω .</p>
          <p>Thus there exists  ∈ Φ ⊆ ‖  ‖ such that  ≺ Γ .</p>
          <p>Contradiction, since  ∈  (‖ ‖, ⪯ Γ). Hence  ∈
Φ , from which it follows that  (‖ ‖, ⪯ Γ) ⊆ Φ .</p>
          <p>Let  ∈ Φ ⊆ ‖  ‖ and suppose towards a
contradiction that  ̸∈  (‖ ‖, ⪯ Γ). Hence there exists a
 ∈  (‖ ‖, ⪯ Γ) such that  ≺ Γ . Thus, by
hypothesis,  ̸∈ Φ . Hence  ∈ Ω . Thus there exists
 ∈  (‖ ‖, ⪯ Γ).
(⇒) If ‖ ‖ = ∅, then Γ ⊙ ⪯ Γ</p>
          <p>diction. Hence ‖ ‖ ̸= ∅
. Let  ∈ ‖ ‖
= Φ = Γ</p>
          <p>.
Contra. Assume that
 ̸∈ Φ . We intend to prove that  ∈ Ω . From  ̸∈ Φ
it follows that  ̸∈  (‖ ‖, ⪯ Γ). Hence there exists
 ∈  (‖ ‖, ⪯ Γ) = Φ such that  ≺ Γ . Thus
 ∈ Ω . On the other hand it holds that Φ ⊆ ‖  ‖</p>
          <p>.</p>
          <p>The following corollary presents a result similar to
that illustrated in Observation 13, but where a system of
equations is considered (rather than a single equation).</p>
          <p>Corollary 8. Let L =≪ 1, 2, ...,  ≫ be a tuple of
labels. For  ∈ {1, . . . , } let Γ  and Φ  be two distinct
non-empty subsets of PL. For  ∈ {1, . . . , } let ⪯ Γ
be a Γ -faithfull pre-order on Γ  such that it holds that
 ̸≺ Γ , for all  ,  ∈ Φ . Let ⊙ ⪯ Γ be a ⪯ Γ
based SPtoSP profile revision operator on Γ . It holds
that,
∀ ∈ {1, . . . , } Γ  ⊙ ⪯ Γ  = Φ</p>
          <p>
            iff

⋃︀ Φ  ⊆ ‖  ‖ ⊆
=1

⋂︀ (Ω  ∪ Φ )
=1
a desirable change, although the two are related problems
(e.g., [
            <xref ref-type="bibr" rid="ref13 ref14">13, 14</xref>
            ]). The approach most closely related to our
proposal is due to Schwind et al. [
            <xref ref-type="bibr" rid="ref15">15</xref>
            ]. In that paper the
authors considered a belief revision scenario where an
announcement (a propositional formula)  is made to a
group of agents, each of these represented by a belief base.
          </p>
          <p>In this scenario, it is considered that  is unknown, and
that, for each agent, its previous beliefs and those obtained
after the revision by  are known. The authors
characterized the set of formulas  satisfying these requirements.</p>
        </sec>
      </sec>
    </sec>
    <sec id="sec-4">
      <title>5. Conclusion and Future Work</title>
      <p>where Ω  = { ∈ PL :  ≺ Γ , for some  ∈
Φ }.</p>
      <p>User profiles are important tools in several areas of
in</p>
      <p>We close this section remarking that an SP-to-SP revi- formation technology. Given a profile, sometimes it is
sion operator defined on a singleton is similar to a P-to-P necessary to determine a set of tasks, or pieces of training
revision operator. However, we chose to treat these opera- which can transform the profile of a user into a target
tors as distinct because they have conceptual differences. profile. In this paper, we provided a formal analysis of
In P-to-P revision, the output is always a profile, while in the process of changing a user profile (or a set of
proSP-to-SP revision, the output is a set of profiles, which ifles) into a target profile (or set of profiles) by means of
may not be a singleton, even if the operator is defined on a profile revision operator, and we identified upper and
a singleton. lower bounds for the set of formulas that cause such a
change. Analogous results are obtained for “systems of
equations" of the form  ⊙   =  and Γ  ⊙   = Φ 
4. Related Works (for every  ∈ {1, . . . , }). This work may have some
pratical applications.</p>
      <p>
        To the best of our knowledge there are not many works
that relate belief revision with user’s profiles. One of those (a) Given a set of previous and current profiles of
works is [
        <xref ref-type="bibr" rid="ref11">11</xref>
        ]. In that paper, the authors developed a ser- a community, the identification of the input that
vice recommendation agent based on belief revision logic caused a certain change can be useful for
deterto handle the non-monotonicity problem of web service mining the impact of spreading fake news or for
recommendation. They applied belief revision-based rea- predicting the benefices of an information
camsoning to determine the most suitable context for the initial paign.
service request based on the beliefs stored in the user’s pro- (b) In cognitive rehabilitation procedures where it is
ifle. After service request reasoning, the set of potential common practice to predetermine goals at the
beweb services is identified and ranked. The highest-ranked ginning of a training program.
services are considered to be the most desirable ones that (c) In systems where it may be convenient to change
match the user’s specific interests. Another paper that a user’s program interface in order to allow the
establishes a connection between belief revision and user execution of more complex interactions.
profiles is [
        <xref ref-type="bibr" rid="ref6">6</xref>
        ]. In that paper, the authors formalize the
creation, representation and dynamics of profiles from a As a future work topic, we intend to model the
modificaKnowledge-Driven perspective. In particular, they pro- tion of a user profile (set of profiles) into a target profile
posed a formal profile representation framework, based (set of profiles) through a series of minimal changes. This
on a formal language that allows to clearly represent a is useful, for example, for the application settings
menuser profile and its attributes. The authors introduced sev- tioned in items (b) and (c) above. In a cognitive
rehabilieral operators for modeling profile’s dynamics, which are tation procedure, the process that leads to rehabilitation
based on well-known operators from the belief revision is performed by small improvements and in the case of
literature, and axiomatically characterized them, includ- changes in user’s program interface, these must be small,
ing the PtoP and the SPtoSP profile revision operators that minimal or imperceptible, since abrupt changes in systems
have been used throughout this article. generally lead to rejections by the end-users.
On the other hand, the problem of identifying the change
formula in a belief revision scenario is related to announce- Acknowledgements
ments (e.g., [
        <xref ref-type="bibr" rid="ref12">12</xref>
        ]). It is important to mention that the
problem of identifying which formula causes a change is We thank the reviewers of KR2023 and also the
reviewdifferent from planning a sequence of formulas to perform ers of ENIGMA-23 for their comments on previous
ver
      </p>
    </sec>
  </body>
  <back>
    <ref-list>
      <ref id="ref1">
        <mixed-citation>
          [1]
          <string-name>
            <given-names>J.</given-names>
            <surname>Bobadilla</surname>
          </string-name>
          ,
          <string-name>
            <given-names>F.</given-names>
            <surname>Ortega</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Hernando</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Gutiérrez</surname>
          </string-name>
          ,
          <article-title>Recommender systems survey</article-title>
          ,
          <source>Knowledge-based systems 46</source>
          (
          <year>2013</year>
          )
          <fpage>109</fpage>
          -
          <lpage>132</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref2">
        <mixed-citation>
          [2]
          <string-name>
            <given-names>S.</given-names>
            <surname>Zhang</surname>
          </string-name>
          ,
          <string-name>
            <given-names>L.</given-names>
            <surname>Yao</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Sun</surname>
          </string-name>
          ,
          <string-name>
            <given-names>Y.</given-names>
            <surname>Tay</surname>
          </string-name>
          ,
          <article-title>Deep learning based recommender system: A survey and new perspectives, ACM Computing Surveys (CSUR) 52 (</article-title>
          <year>2019</year>
          )
          <fpage>1</fpage>
          -
          <lpage>38</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref3">
        <mixed-citation>
          [3]
          <string-name>
            <given-names>D.</given-names>
            <surname>Browne</surname>
          </string-name>
          ,
          <article-title>Adaptive user interfaces</article-title>
          ,
          <source>Elsevier</source>
          ,
          <year>2016</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref4">
        <mixed-citation>
          [4]
          <string-name>
            <given-names>P.</given-names>
            <surname>Lops</surname>
          </string-name>
          ,
          <string-name>
            <given-names>C.</given-names>
            <surname>Musto</surname>
          </string-name>
          ,
          <string-name>
            <given-names>F.</given-names>
            <surname>Narducci</surname>
          </string-name>
          , G. Semeraro,
          <source>Semantics in Adaptive and Personalised Systems</source>
          , Springer,
          <year>2019</year>
          .
        </mixed-citation>
      </ref>
      <ref id="ref5">
        <mixed-citation>
          [5]
          <string-name>
            <given-names>W.</given-names>
            <surname>Deng</surname>
          </string-name>
          ,
          <string-name>
            <given-names>I.</given-names>
            <surname>Papavasileiou</surname>
          </string-name>
          ,
          <string-name>
            <given-names>Z.</given-names>
            <surname>Qiao</surname>
          </string-name>
          ,
          <string-name>
            <given-names>W.</given-names>
            <surname>Zhang</surname>
          </string-name>
          , K.-Y. Lam, S. Han,
          <article-title>Advances in automation technologies for lower extremity neurorehabilitation: A review and future challenges</article-title>
          ,
          <source>IEEE reviews in biomedical engineering 11</source>
          (
          <year>2018</year>
          )
          <fpage>289</fpage>
          -
          <lpage>305</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref6">
        <mixed-citation>
          [6]
          <string-name>
            <given-names>E.</given-names>
            <surname>Fermé</surname>
          </string-name>
          ,
          <string-name>
            <given-names>M.</given-names>
            <surname>Garapa</surname>
          </string-name>
          ,
          <string-name>
            <surname>M. D. L. Reis</surname>
            ,
            <given-names>Y.</given-names>
          </string-name>
          <string-name>
            <surname>Almeida</surname>
            ,
            <given-names>T.</given-names>
          </string-name>
          <string-name>
            <surname>Paulino</surname>
            ,
            <given-names>M.</given-names>
          </string-name>
          <string-name>
            <surname>Rodrigues</surname>
          </string-name>
          ,
          <article-title>Knowledge-driven profile dynamics (</article-title>
          <year>2022</year>
          ). Submitted manuscript.
        </mixed-citation>
      </ref>
      <ref id="ref7">
        <mixed-citation>
          [7]
          <string-name>
            <given-names>H.</given-names>
            <surname>Katsuno</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Mendelzon</surname>
          </string-name>
          ,
          <article-title>Propositional knowledge base revision and minimal change</article-title>
          ,
          <source>Journal of Artificial Intelligence</source>
          <volume>52</volume>
          (
          <year>1991</year>
          )
          <fpage>263</fpage>
          -
          <lpage>294</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref8">
        <mixed-citation>
          [8]
          <string-name>
            <given-names>H.</given-names>
            <surname>Katsuno</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Mendelzon</surname>
          </string-name>
          ,
          <article-title>On the difference between updating a knowledge base and revising it</article-title>
          , in: P. Gärdenfors (Ed.), Belief Revision, number 29 in Cambridge Tracts in Theoretical Computer Science, Cambridge University Press,
          <year>1992</year>
          , pp.
          <fpage>183</fpage>
          -
          <lpage>203</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref9">
        <mixed-citation>
          [9]
          <string-name>
            <given-names>P.</given-names>
            <surname>Gärdenfors</surname>
          </string-name>
          ,
          <article-title>Conditionals and changes of belief</article-title>
          ,
          <source>Acta Philosophica Fennica</source>
          <volume>30</volume>
          (
          <year>1978</year>
          )
          <fpage>381</fpage>
          -
          <lpage>404</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref10">
        <mixed-citation>
          [10]
          <string-name>
            <given-names>D.</given-names>
            <surname>Nute</surname>
          </string-name>
          ,
          <article-title>Conditional logic</article-title>
          , in: Handbook of philosophical logic, Springer,
          <year>1984</year>
          , pp.
          <fpage>387</fpage>
          -
          <lpage>439</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref11">
        <mixed-citation>
          [11]
          <string-name>
            <surname>R. Y. K. Lau</surname>
          </string-name>
          , L. Song,
          <source>Belief Revision for Intelligent Web Service Recommendation</source>
          , Springer Berlin Heidelberg, Berlin, Heidelberg,
          <year>2012</year>
          , pp.
          <fpage>53</fpage>
          -
          <lpage>66</lpage>
          . URL: https://doi.org/10.1007/978-3-
          <fpage>642</fpage>
          -30454-
          <issue>5</issue>
          _ 4. doi:
          <volume>10</volume>
          .1007/978-3-
          <fpage>642</fpage>
          -30454-
          <issue>5</issue>
          _
          <fpage>4</fpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref12">
        <mixed-citation>
          [12]
          <string-name>
            <given-names>X.</given-names>
            <surname>Li</surname>
          </string-name>
          ,
          <string-name>
            <given-names>D.</given-names>
            <surname>Gabbay</surname>
          </string-name>
          ,
          <string-name>
            <given-names>R.</given-names>
            <surname>Markovich</surname>
          </string-name>
          ,
          <article-title>Dynamic deontic logic for permitted announcements</article-title>
          ,
          <source>in: Proceedings of the International Conference on Principles of Knowledge Representation and Reasoning</source>
          , volume
          <volume>19</volume>
          ,
          <year>2022</year>
          , pp.
          <fpage>226</fpage>
          -
          <lpage>235</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref13">
        <mixed-citation>
          [13]
          <string-name>
            <given-names>R.</given-names>
            <surname>Booth</surname>
          </string-name>
          ,
          <string-name>
            <given-names>A.</given-names>
            <surname>Nittka</surname>
          </string-name>
          ,
          <article-title>Reconstructing an agent's epistemic state from observations about its beliefs and non-beliefs</article-title>
          ,
          <source>Journal of Logic and Computation</source>
          <volume>18</volume>
          (
          <year>2008</year>
          )
          <fpage>755</fpage>
          -
          <lpage>782</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref14">
        <mixed-citation>
          [14]
          <string-name>
            <given-names>P.</given-names>
            <surname>Liberatore</surname>
          </string-name>
          ,
          <article-title>Revision by history</article-title>
          ,
          <source>Journal of Artificial Intelligence Research</source>
          <volume>52</volume>
          (
          <year>2015</year>
          )
          <fpage>287</fpage>
          -
          <lpage>329</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref15">
        <mixed-citation>
          [15]
          <string-name>
            <given-names>N.</given-names>
            <surname>Schwind</surname>
          </string-name>
          ,
          <string-name>
            <given-names>K.</given-names>
            <surname>Inoue</surname>
          </string-name>
          ,
          <string-name>
            <given-names>S.</given-names>
            <surname>Konieczny</surname>
          </string-name>
          ,
          <string-name>
            <surname>J.-M. Lagniez</surname>
            ,
            <given-names>P.</given-names>
          </string-name>
          <string-name>
            <surname>Marquis</surname>
          </string-name>
          ,
          <article-title>What has been said? Identifying the change formula in a belief revision scenario</article-title>
          ,
          <source>in: Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, IJCAI-19, International Joint Conferences on Artificial Intelligence Organization</source>
          ,
          <year>2019</year>
          , pp.
          <fpage>1865</fpage>
          -
          <lpage>1871</lpage>
          . URL: https://doi.org/10.24963/ijcai.
          <year>2019</year>
          /258. doi:
          <volume>10</volume>
          .24963/ijcai.
          <year>2019</year>
          /258.
        </mixed-citation>
      </ref>
    </ref-list>
  </back>
</article>