Considering Structural and Vocabulary Heterogeneity in XML Query: FPTPQ and Holistic Evaluation.

CONSIDERING STRUCTURAL AND VOCABULARY
HETEROGENEITY IN XML QUERY: FPTPQ AND
HOLISTIC EVALUATION.
Brice Nguefack1
, Maurice Tchoupé Tchendji2
and Thomas Djotio Ndie3
3
National Advanced School of Engineering, University of Yaoundé I, Yaoundé,
Abstract
The integration of XML data sources which have different schemas/DTD can originate structural
and vocabular heterogeneity. In this context, it is difficult to write satisfiable queries. As a
solution, many Information Systems focus on building approximate evaluation techniques for
exact queries. As a project, we build flexible and preference XML query languages and associated
evaluation algorithms. In this paper, we propose the Flexible Preference Tree Pattern Query
(FPTPQ), a new TPQ that allows multiple items/names (resp. paths) for the same node, in order
to integrate (resp. to locate) all the different instances of the database nodes. The FPTPQ enable
to have preference nodes and ordering operators among label items and paths. We also provide
a holistic algorithm that evaluates the FPTPQ and capitalises the preferences to determine the
best available solutions. Illustrations and experimentations are realized to show the effectiveness
of our solutions.
Keywords
XML Query with structural preference, Structural heterogeneity, vocabular heterogeneity, flexible query,
holistic matching algorithm, ranked results.
1. Introduction
XML has become the standard format for information representation and data exchange among
different systems. For interoperability needs, data of various sources which have been modelled
independently, can be merged or exploited simultaneously. Generally, the integration of XML
data that come from different sources may cause heterogeneity problems [1]. Heterogeneity in the
domain of databases have different appreciations [2, 3]. Amount the types of heterogeneity, we
denote structural heterogeneity and vocabular heterogeneity. The first refers to the fact that the
same information instances may have different paths whose name and number of the nested
database tree nodes (tag) are different. The second signifies that the instances of the same object
(tag, attribute, text value) are represented with different names, that can be synonyms (e.g. skill,
competency and expertise), abbreviations (course, crse), or any other group of similar words
(hours and times, football and soccer). This problem has serious impact on the querying process.
As a matter of fact, during the exact matching of exact queries, since all the constrains are
considered as filters, a single erroneous object name among numerous ones means total failure,
1&2
Department of Mathematics and Computer Science, University of Dschang,
Cameroun
1
DOI:10.5121/ijdms.2021.13601
Dschang, Cameroon
International Journal of Database Management Systems (IJDMS) Vol.13, No.5/6, December 2021

The consideration of only one denomination among all the object instance names or paths may
result to an incomplete set of solutions. For example, Let’s consider the merging of the documents
uwm.xml and wsu.xml available on [4] which contain respectively the schedule of the University
of Wisconsin-Milwaukee (UWM) and the Washington State University (WSU) courses. Table 1
shows fragments of each of these documents. Even though these documents have been built for
the same goal and for the same subject (courses scheduling), some information instances are
contained in different tag (with similar names) or/and are located with different paths. The
correspondent of the uwm.xml tags course, course_listing and hours are respectively crs, course
and times in wsu.xml. The starting time of a course is located by the path
root/course_listing/section_listing/start in uwm.xml and by root/course/time/start in wsu.xml.
The integration of these two databases creates structural heterogeneity and vocabular
heterogeneity. A non-aware user who needs the title of all the courses may have as result only the
ones of WSU courses if he uses the tag course without considering course_listing. He will face
the same consequences if all the different paths of the needed database node are not considered
in the query.
The characterization of the searched information alongside with the query evaluation process
most consider these issues in order to find the complete set of solutions available for the user in
the database. In the literature, some authors focus on tailoring a homogeneous database from the
heterogeneous XML document collection. Other authors make some mapping between data
sources schemas, that help to construct from each source document (resp query), a target one
which is more appropriated for the querying process [5] [6]. In [7], AlHamad et al. produce a
global schema for the entire database, alongside with mapping procedures between it and the
multiple sources schemas, in order to provide a homogeneous view over heterogeneous XML data
Table 1. Fragment of the XML files of the two USA university course schedule, available
on [4].
a) Fragment of uwm.xml b) Fragment of wsu.xml
<root>
<course_listing>
<note></note>
<course>400-327</course>
<title>CONTEMPORARY FRENCH …</title>
<credits>3</credits>
<level>U</level>
<restrictions>; (HU) PREREQ: FRENCH 303(215)
…</restrictions>
<section_listing>
<section_note></section_note>
<section>Se 001</section>
<days>TR</days>
<hours>
<start>12:30pm</start>
<end>1:45pm </end>
</hours>
<bldg_and_rm>
<bldg>CRT</bldg>
<rm>B13 </rm>
</bldg_and_rm>
<instructor>Alkhas Alkhas</instructor>
</section_listing>
</course_listing>
…
</root>
<root>
<course>
<footnote></footnote>
<sln>10637</sln>
<prefix>ACCTG</prefix>
<crs>230</crs>
<lab></lab>
<sect>01</sect>
<title>INT FIN ACCT</title>
<credit>3.0</credit>
<days>TU,TH</days>
<times>
<start>7:45</start>
<end>9</end>
</times>
<place>
<bldg>TODD</bldg>
<room>230</room>
</place>
<instructor>B.
MCELDOWNEY</instructor>
<limit>0112</limit>
<enrolled>0108</enrolled>
</course>
…
</root>
2

[8], [9]. Further works [10], use ontology, web semantic [11], thesaurus, passed user experiences
and other artificial intelligence techniques to automatically rewrite query. The purpose is to obtain
from the user query, a more suitable one, with the respect of the database structure/format [12].
The resulting queries are more likely to produce solutions than the user one, since the user may
not be familiar to database schema and vocabulary. These techniques are used in mediation
systems. A rewriting system can produce as much targets queries as there are data sources or
clusters with different schemas. Consequently, multiple evaluations are needed. The flexibility
features are used after the formulation of the query by the user. The cited works do not allow the
user to propose himself, some additional label items that may be useful to determine alternative
solutions, or the complete set of the needed solutions. In fact, the existing query languages do not
allow to have multiple paths to localized the occurrences of the same object, or multiple (similar)
words as a node label, where each word can be one of the names used for the same database node.
Moreover, the user may have in mind some potential related substitute words or objects which
can help the evaluation process to also select alternative solutions in case the ideal query (query
that is supposed to give the user a maximum satisfaction) does not have database images. But he
will still be obliged to write other queries by continuously make several adjustments from the
initial one, replacing some labels by others, sometime without changing the initial query
organization. If this situation is more common in e-commerce context, it is a general problem for
database querying system. Note that, as an XML document has a tree shape, XML queries that
are written with the most popular languages XPath and XQuery can be represented as a
combination of one or many trees, called Tree Pattern Queries (TPQ).
Different types of tree pattern query exist in the literature [13]. The most expressive ones enable
the utilization of the wildcard * to match any single database node, or the Ancestor-Descendant
(//) operators to allow the matching of the same object occurrences that have different paths. As
example, If the user wants to select the titles of books and articles in the database dblp, he may
use the wildcard “*” to have the paths dblp/*/title. To select all the start time of the courses in
merged_wsu-uwm.xml, he may use root//start. But these operators are responsible of many
useless solutions and does not enable any preference order. dblp/*/title will select the title of all
the documents (articles, books, but also improceedings, master thesis and PhD thesis). root//start
will select all the occurrences of the node start which are the descendant of the root, no matter
what is in-between. Rather than using the wildcard and the A-D operators, a list of additional
elements can be added as replacement items to query node labels/paths. This type of flexibility
ensures that only the needed solutions are returned to the user. In some case, the replacement
items can allow the selection of alternative solutions that are closed to user needs (example room
if studio are not available). Such tree pattern is not useful only for the user. It can also be used to
represent the results of a query reformulation [14] [15] [16] by a mediator system.
Flexibilities have been imported in database query process to write soft queries and allow more
solutions possibilities. The concept of preference query, also none as bipolar query, is used to
write query that have two parts: a first part has “must be satisfied” constraints, and a preference
part is made of soft constraints. Contrary to the first part whose constraints are considered as filter
(conditions whose every solution must absolutely satisfy), the satisfaction of the preference part
is optional, but it enhances the correspondent solution value. i.e., the solutions which also satisfy
the second part are must likely to be preferred by the user than those who satisfy only the
obligatory constraints. None of the existing TPQ, even preference one, allow the integration of
difference instance names for the same object, as node label. None of them allow multiples paths
for the same query leaf node. In this paper, we propose (in section 3) a more general Tree Pattern
Query called Flexible Preference Tree Pattern Query (FPTPQ). The FPTPQ enable replacement
3

items for node labels, and multiple paths to locate the multiple instances of the same database
object. The items can be of equal value or classified in preference order using ordering operators,
when the replacement items are the attributes of alternative solutions that have difference
preference values, according to the user. For example, a user who need a room can add studio as
alternative. Attributes of room will be more rated than the attributes of studios. The FPTPQ
improves the preference operator of the language prefSXPath, proposed in our preview article
[17]. For the evaluation of the FPTPQ, we proposed (section 4.4), the holistic matching algorithm
FlexPrefTreeMatch which is an improved version of TreeMatch [18]. The matching of the query
is paired with the calculation of each solution weight, utilized to determine the best solutions.
Illustrations (Section 5) and experimentations (section 6) are made, in order to show the
effectiveness of the FPTPQ and the algorithm FlexPrefTreeMatch: that is to show how the
complete solution set is returned, and how the useless solutions caused by * and // are avoided.
2. Preliminaries
In this section, we define the concepts related to XML databases querying process, that are useful
to understand this work.
2.1. XML and heterogeneous database.
An XML database is a collection of XML documents. An XML document [19] Consists of a set
of hierarchical tags that describe the data they contain. An XML document must be well-formed,
that is it must be in accordance with XML recommendations. The structure of XML documents
is fundamentally tree oriented, so it can be modeled as a rooted tree 𝑡 = (N, E) where N is a set
of nodes labeled with the tag name for internal nodes, data or attributes for leaf nodes. E is the set
of edges, each one represented as a couple (𝑛𝑖, 𝑛𝑗 ) ∈ (𝑁 × 𝑁) that connect a node 𝑛𝑖, to 𝑛𝑗.
DTDs or XML schemas when used, impose a structure/format to the XML database documents.
They define the nature and the type of the elements that may be included in a valid document of
the collection, and the way these elements are nested. An XML database can therefore be
represented as a forest. XML documents validated against a DTD or XML Schema are said to be
"Valid", The corresponding database object instances usually have uniform structure and same
tag names, and can be considered as homogeneous. In the other hand, the integration of many
XML data sources with different structure/format may result to a heterogeneous database. XML
database heterogeneity can be interpreted in different way. In this paper, we are interested in
vocabular heterogeneity and structural heterogeneity [20] . We talk of vocabular heterogeneity
when the same object instances are expressed by different tag names, synonyms, abbreviations or
other languages borrowed words. Structural heterogeneity refers to the fact that the same
information instances may be located with more than one paths. Heterogeneity makes it difficult
to write satisfiable query, using the common XML query languages. As XML document, XML
queries can be modeled as a combination of trees, called Tree Pattern Queries.
2.2. XML query and Tree Pattern Query (TPQ)
To extract specified data from an XML database, many query languages have been developed.
The most famous are XPath [21] and XQuery [22]. A common feature of these languages is a
possibility to formulate paths in the database tree or forest. Such a path is a sequence of tree nodes
from the root to the searched element occurrences. Regular expressions of XML query languages
provide valuable methods for paths specifications based on XPath, and some formular to join the
path images. XML queries can be translated into one or many trees, called Tree Pattern Query
4

[13] (TPQ) with the respect of the structure and the complexity of the query. Each TPQ is used
to represent a useful fragment of the principal query. A TPQ is a tree 𝑡𝑄 = (𝑁𝑄, 𝐸𝑄) where 𝑁𝑄
is a set of nodes containing the root of 𝑡𝑄 and 𝐸𝑄 is the set of edges represented as couple
(𝑛𝑄𝑖
, 𝑛𝑄𝑗
) ∈ (𝑁𝑄, 𝐸𝑄) that connect the query nodes 𝑛𝑄𝑖
to 𝑛𝑄𝑗
. Several TPQ models exist in the
literature with different kind of features, some are more expressive than others, according to the
operators and relations they offer. One of the first tree pattern queries is the Tree Algebra for
XML Tree Pattern Query (TAXTPQ) [23]. Its main features are the ancestor-descendant (A//D)
and parent-child (P/C) relations [24]. If had brought the basics features for the other TPQ. The
TAXTPQ is too rigid and the user has to master the structure of the database in order to adapt his
request accordingly. The absent of only one edge or node in a potential solution tree prevents it
from being in the final result of the matching, even if the candidate subtree is "almost perfect”
[13]. As a response, Chen et al proposed the Generalized Tree Pattern (GTP) [25], a TPQ that
enable some edges or nodes to be optional, by associating a mandatory/optional status to them, in
other to increase the possible matched subtrees in the database tree. More than a limitation
Through the Annotated Pattern Tree, Paparisos et al [26] allow the addition of a specification to
an edge (u, v), which specifies how many matches to node v are to be obtained from each match
with node u. Moreover, Lu et al [18] have proposed the Extended XML tree pattern that enable
more relaxation with the wildcard node “*”. The wildcard can match any single tree element. It
is usually used when the associate’s element is unknow or is not important. The risk with this
operator is the abundancy of the corresponding matching element alongside with the query
solutions. In fact, it uses to originate many useless solutions that imply costly filtering.
Sometimes, the user or the potential mediator system could have some additional elements that
may reinforce the satisfiability of the query, when there is no assurance that the first (principal)
searched element will be available. In this context, rather than using the wildcard and give a totally
freedom to the matching algorithm to select anything, it is better to insert the additional elements
as replacement items that will help to produce alternatives solutions which are near to the user
initial needs. But the existing tree pattern does not allow it. The optional operator “?” is a
beginning of a solution, but allow only an element to be optional [26]. With the logical operator
OR and XOR, Izazi et al [27] allow the selection of one of two proposed sub path that finished at
different leaf nodes. They don’t integrate the fact that only some internal nodes/paths may need
some replacement items. For example, if we consider the two documents of Table 1, we see that
the starting time of the two documents courses are inside the tag start, while the parent tags are
different, hours for the first document, and times for the second. The TPQ of Izazi et al does not
allow to have hours and times as items of the same node.
2.3. Exact query, Flexibles and preference queries languages
Exact query is considered as query where all the constraints most obligatory be satisfied. All the
constraints are considered as filter. The none satisfaction of all the conditions by a solution
disqualify it from being in the final result set. Flexible queries are relaxed, allow soft matching
and favour more solutions possibilities. Queries containing the operators “?” and “*” can be
consider as flexible queries, since the first accept the fact that an element of the image tree can be
absent and the second can be matched by any single database node (when different database nodes
can be the image of the same query node). The wildcard is responsible of big solution set. At the
end of the matching, all the solutions of such queries are considered to be of equal weight and are
returned to the users without any ranking, obviously with all the useless solutions.
5

Preference queries, more that flexible queries also include operators useful for the calculation of
the best solutions. As example we consider a tourist looking for a room in a luxury hotel with
swimming pool and beach. Even if hotels with beach are difficult to find, they are more likely to
be preferred by the users. If it is possible to consider the beach as optional (in flexible query), in
the final query solution set, all the hotels room’s will be returned in a random order, with no
consideration of if they have a beach or not. The best formulation of that query is "I want a room
in a luxury hotel, with a preference for hotels that have a swimming pool and a beach". In this
case, it is better to use a preference query language [28] [29] [30] or the bipolar queries languages
of our previews works [17] [31] [32] which make it possible to write queries with two parts: a
first part containing the obligatory constraints that most necessarily be satisfied and a second part
containing the elements of preferences whose availability magnify the corresponding solutions.
At the end of the matching, the solutions that are not dominated, those which satisfy all the
obligatory constraints and incorporate more elements of preferences than the others, are returned.
Pareto's dominance concept through the skyline [33] [34] [35] [36] operator is commonly used to
compare the solutions. But more is still to be done to make XML queries languages and TPQ
more flexible even though they integrate preferences. The non-existence of database schema may
induce the representation of an information instance by different words, in the same collection of
documents. Moreover, in the context of heterogenous database, XML query language must allow
the integration of all the different representations of an information in the query (for focused
flexibility) while minimizing the utilization of the operators * and // which cause huge number of
useless solutions.
3. Flexibles Preferences Tree Pattern Queries (FPTPQ)
3.1. motivations
Vocabular heterogeneity and structural heterogeneity make difficult to write query that capture
all the different representations of the database tags. Several issues are considered:
Issue 1: The database mays contain some words (tags) and their synonyms, abbreviations, other
language borrowed word. For example, in the database of a country labor’s ministry, a football
club can be considered as an enterprise, whereas in the database of the Football League the word
used for the corresponding tag is "club"; For those who are used to American language, the
collective sport where the ball is moved only with foot is called soccer rather than football in
European country. Another example is the utilization of the words option and specialty to indicate
the area of study. The same object instance may have different paths in the database. As example
root/course_listing/section_listing/start and root/course/time/start in merged_wsu-uwm.xml. In
this issue, the listed elements are similar, and thus have the same values. We only need to precise
the different word/path instances no matter the order. The purpose is to select all the available
solutions. Here we introduce the expression Flexible Node (FN). A FN is a node which have
more than one label item. Its label items are separated by the operator “|”. In the query S2 of
Figure 1, the node whose label is (country | location), is a flexible node. The purpose is to
maximize the satisfiability of the constrains associated to that node.
Issue 2: This issue concerns user preference node label which has multiple representations or
paths in the database. Here, the objective is to maximize the satisfiability of the preference
constraints. The list of the similar elements should be listed alongside with the preference
operator. The associated node is called a Flexible Preference Node (FPN). A FPN is a preference
node which has more than one label. The preference operator remains “!” as proposed in ours
preview work [17]. In the last query of Figure 1, the node whose label is (proc1|proc2)!, is a FPN.
6

Issue 3: In some circumstances as e-commerce, the attributes (constraints) for alternative
solutions can be added as replacement items alongside with preference order, in the same query.
So that, if the ideal (user first choice) solution is missing, alternatives ones will be selected. The
established preference order is used to calculate the best solutions among all the available ones.
For example, in the last query, the node whose label is (Proc1 | proc2)>! is an Ordered Flexible
Preference Node (OFPN). In an OFPN, the node label items are classified in ascending or in
decreasing order, thus (Proc1 | proc2)>! Can also be written as (Proc2 | proc1)<!. In this node,
the label item proc1 is more preferred than proc2. The solutions whose tree has Proc1 will have
a greater preference value than those whose tree has proc2.
The different types of nodes can be combined inside the same query.
3.2. Flexibles Preferences Tree Pattern Queries (FPTPQ): language.
It is already possible to propose many preference alternatives for data values (processors name,
screen size, etc.) using Sara Cohen preference language [30]. None of the existing TPQ allow to
do so for internal node labels. In this section, we present the Flexible Preference Tree Pattern
Query (FPTPQ), a model of Tree Pattern Query which gives the possibility of proposing many
paths or/and many label items (words) for some query node, alongside with on-demand ordering
operators which are useful for the calculation of solutions preference weights. This is another way
to enable more flexibility at the level of the query writing module. Since the user can be the one
to propose replacement items, the associated results are likely to satisfy him and save the querying
systems from multiples query execution and useless filtering operations. To express preferences
inside the query, we extend the prefSXpaths language proposed in our previews work[16]. As
another example, let’s consider the queries of Figure 1, from Sara Cohen et al [30]: In S1, a tourist
(Sam) needs a hotel to stay at when attending a conference in China. Ideally, Sam would like a
cheap (at most 1000 RMB per night) hotel in China. Since he would like to taste the local food,
he would like Chinese food to be served at the hotel. Sam needs an Internet connection, to keep
in touch. Finally, Sam will be bringing his wife and new baby, and so will need a crib. Despite it
is represented as a preference query type, in a heterogeneous database, the crib can be expressed
by its synonym cradle. In the place of the tag country, the database designer could have used the
word location. In order to maximize the satisfiability, synonyms or equivalent word (cradle and
location) can be added as replacement item for the label country and crib in order to have query
S2. In some cases, the main label item, better contribute for user satisfaction than the other
replacement items.
In a preference query, a priority/preference order can be set among the replacement items of
preference nodes. In this case, the operators “<” (resp “>”) are added before “!” to indicate that
the replacement item are listed in ascending (resp in decreasing) order of preference.
Figure 1. Sara Cohen preferences Vs flexible preferences
7

Figure 2-b show how to represent a FPTPQ (of Figure 2-a) that has FN and FPN, using their
associated variable. Figure 2-d do the same for the FPTPQ of Figure 2-c which has both FN and
OFPN. All the other nodes constraints are expressed as in the Extended Tree pattern query [18].
The preference operator still remains “!”. We can now have preferences nodes that have
replacement items. The binary operator "|" is used as separator of multiple paths or label items.
As label of a preference query node, (p1|p2|…|pn)! means that in the absence of P1, the items
P2,…,pn will equally replace it. Figure 3-a show the tree representation of the query
A[(D|E)!/F]/B!/C. In this query, node 2 is a preference node, D and E are both label items of a
preference node, with the same weight. In the absence of D, E will equally be considered. In the
query of Figure 3-b, the preferences nodes items are ordered. This means there are classify in
increase order like (C|D)<! or in decreasing order like (G|H)>!. Here, the label item D is more
valuable than C. The solutions whose tree carry D will be more valuable than those whose tree
carry C. In the query of Figure 3-b, the item H can replace G, but with a lower preference value.
3.3. Assignation of preferences values to preference node items
We need to know that preferences values are assigned to preferences node items, according to the
type of preference node there are associated to. All the items of FPN are assigned the default
preference value which is “1” like in the example of Figure 3-a. For OFPN that carry a lower
operator (<), the label items from the first, are assigned preference values respectively from the
integer 1 to N, where N is the number of items. When the preference node has the operator “>”,
preference values are assigned to its label items from “N” to “1”. In Figure 3-b, the node 3 has
the operator “<” which mean its label item are classify in ascending order. So, the label item C
has “1” as preference value and the label item D has “2” as preference value. For the node 6, the
label items are classified in descending order; its values then start from “2” for G, to “1” for H.
4. Evaluation of the FPTPQ: FlexPrefTreeMatch
In this section, we present an evaluation approach of the FPTPQ. The purpose is to minimize the
FPTPQ by removing all the labels items that do not appear in the database, then the minimized
Figure 2. Expressing the FPTPQ with variables.
Figure 3. FPTPQ with different types of preference nodes
8

FPTPQ is matched with an extended Dewey based index using a holistic algorithm that is based
on treeMatch [18]. During the matching, the preference value of each solution is also calculated.
4.1. The algorithm TreeMatch
We used the algorithm TreeMatch as the backbone of our proposed algorithms because it is able
to optimally (in term of I/O complexity) processed the Extended Tree Pattern Query, which is one
of the most featured and flexible TPQ, since it allows negation function, order-based axis and
wildcards. In fact, treeMatch has one of the larger optimality classes in terms of input-output. The
optimality class of a tree pattern matching algorithm represent the set of TPQ it is capable of
optimally evaluate. Since the FPTPQ integrate replacement items, preference operator, and
ordering among items, some modifications have been made on TreeMatch for its evaluation. The
evaluation process of a FPTPQ start with some precomputing which consist of the minimization
of the FPTPQ label items, node’s identification, attribution of preference values to each node
item, the determination of all the paths associated to each query leaf node, alongside with their
corresponding preference values.
4.2. Precomputing: FPTPQ minimization, node identification and calculation of
preference values of all paths.
Before the evaluation, the FPTPQ label item are minimized, using the database tag names list. In
fact, all the label items that do not appear in the database tag list are removed from the FPTPQ.
the FPTPQ query nodes are then numbered, using integer values that are used to identify them.
Because a node label can have many items, it is not appropriated to identify them with their label
or with one of their label items, since each item may be manipulated separately. For example, if
a node label is “A|B”, A|B or A are not more appropriate identifier for it, since A|B is too long
and each database index list contains the occurrences of only one object. Integer identifiers are
associated to the node with the respect of their position. The Figure 3 show how the nodes of
different types of queries are numbered. The couple (i, lj) is used to refers to the label item (name)
lj of the node whose id is j. During the evaluation process, the preference value of each solution
is progressively calculated using the preferences values associated to each query branching node
that are stored in the preference locate match table. The description of this table is shown in
section 4.3, alongside with the procedure that need it.
4.3. Used Data structures
The inputs of the algorithm treeMatch are a FPTPQ and the lists Tq associated to each label item
q of the minimized query. Tq contains the extended Dewey label of all the tag occurrences whose
name is q. eq is used to refer an element of a Tq list. Cur(Tq) is used to denoted the current element
pointed by the cursor of Tq. The procedure advance(Tq) is used to advance the cursor of Tq to the
next element. Like for the algorithm TreeMatch in [18], a set Si is associated to each query
branching node. Here, “i” no more the branching node label, but its id. Each element eq of the set
is a triplet (label, intVector, outputList) where label is the extended Dewey label of eq. intVector
is a vector of integer whose size is equal to the number of descendants of q. Compared to bitVector
used by TreeMatch, intVector has many rules. Its first rule is to tell whether eq has the proper
children or descendant with the respect of the query (as bitVector for TreeMatch). Its second
function is to save the current (partial) preference value of the potential solutions. Indeed, each
integer of intVector represent the preference value of all the its associated descendant subtree.
Given a child node qc of q, let consider intVector(eq)[qc]= V: V>=0 if and only if there is a
9

database element eqc such as eq and eqc satisfy the query relationship between q and qc and V is the
preference value of the subtree rooted by qc; V= -1 if not. OutputList contains the elements that
potentially contribute to final query answers.
4.4.Algorithms flexPrefTreeMatch and associated procedures and functions.
4.4.1. The principal algorithm: flexPrefTreeMatch
Line 1 locate the first match label of each query leaf node. If a leaf node is a FN then all its label
items first matches would be located and the one with the minimum match label will be first
processed. Now the function prefGetNext select among all the query leaf nodes, the one which is
going to be processed (the one that has the item with the minimum match label). prefGetNext
return a couple (fid, fact) where fid is the id of the next leaf node to be processed, and fact is the label
item of fid , that has the minimum current matchLabel. The purpose of line 4 and 5 is to insert the
potential matching element inside the outputList of NDB(fid). After the treatment of fid, the cursor
of Tfact is advanced to the next element. Line 7 update the set encoding and line 8 locate the next
matching element to individual root to leaf path. Finally, line 9 do the appropriate update for the
final solutions; for FPTPQ which has preferences, the weighted solutions table is constructed.
4.4.2. PrefGetNext and other procedures and functions used by flexPrefTreeMatch
The index used by flexPrefTreeMatch remains the Tq lists, where each Tq list contains the
Extended Dewey label of all the database occurrences of q. A list 𝑇𝑞 is visited only if q is the label
item of a query leaf node. Initially a pointer is positioned at the first element of 𝑇𝑞.
PrefLocateMatchLabel is a very important procedure, its purpose is to locate the first elements
whose path match one of the individual root-leaf query paths with the respect. In spite, for FPTPQ,
many paths can be associated to a query leaf node. The preference locate match table is used to
associate to each query leaf id, all its corresponding root to leaf paths in other to facilitate the
matching. Table 2 associate to each leaf node item of the query of Figure 5-b all its corresponding
root to leaf paths. For leaf node which are flexibles, all its label replacement items have to be
considered during the matching. Let consider a leaf node whose label items are q1..qn; during the
matching, if the current matchLabels of these items are respectively e1..en, then, the label with the
minimum (by lexicographical order) will be first selected to make sure that the evaluation is being
made by lexicographical order.
The function minMatchLabel(n) return the minimum of the current matchLabels of all the label
items of the node n. The function minItem (n) return amount the label item of n, the one that has
the minimum current match label. That is nimin | cur (Tnimin )= minMatchLabel(n).
Algorithme1: flexPrefTreeMatch
1: PrefLocateMatchLabel(Q);
2: while (¬end(root)) do
3: (fid, fact) = prefGetNext(topBranchingNode);
4: if (fid is a return node)
5: addToOutputList(NAB(fid), cur(Tfact));
6: Advance (Tfact); // read the next element in Tfact
7: prefUpdateSet(fid, fact ); // update set encoding
8: prefLocateMatchLabel(Q); // locate the next element with matching path
9: emptyAllSets(root);
10

Give the current matchLabel e of the node label n, flexMB (n, b) return all the matchLabel of n
that cover (are ancestors or parent of) e. flexMB help to make sure that if many occurrences of the
branching node (b) carry e, we always start to process the deepest one, since the evaluation is
bottom-up. The function minValue(v) return the minimal value of the intVector v. The function
prefValue(eq', eq), return the preference value of the path that link eq' to eq. It is equal to the
preference value of eq added to the sum of all preference node value between eq' and eq. This
function is used by the procedure updateAncestorSet to set the integer value of the nearest ancestor
branching node set. Assume that q is a branching node and qi is it children, intVector (eq, eqi) =
prefValue(eq, eqi). Like the matching process, the calculation of the final solution preference value
(weight) of a node is bottom-up. This mean that, during the matching, the preference value of a
branching node matchLabel is equal to the sum of all the preference values of its children subtree
images. The function flexSatisfyTreePattern(eqi, eq) test whether the document element eqi is
covered by the branching node matchLabel eq; it return true if intVector (eq)[eqi]>=0;
Function prefGetNext(n)
1: if (isLeaf(n) then
2: return (𝑛, 𝑛𝑚𝑖𝑛)| 𝑛𝑚𝑖𝑛 = minItem(n)
3: else
4: for each ni ϵ NDB(n) do
5: (mi, fi) = prefGetNext(ni)
6: if ( isBranching(ni) and ¬empty(Sni)
7: return (mi, fi)
8: else ei = max{p | p ϵ flexMB(ni, n)};
9: end for
10: max = maxArgi {ei};
11: for each ni ϵ NDB(n) do
12: if (∀𝑒 ϵ flexMB(𝑛𝑖, n): e ∉ ancestors(𝑒𝑚𝑎𝑥))
13: return (mi, fi);
14: endif
15: end for
16: 𝑚𝑖𝑛 = minarg𝑖 {𝑓𝑖 |𝑓𝑖 is not a return node};
17: for each e ϵ flexMB(nmin; n)
18: if (e ϵ ancestors(emax) updateSet(Sn; e);
19: end for
20: return (mi, fmin);
21: end if
Function flexMB (n, b)
1: if (isBranching(n)) then
2: Let e be the maximal element in set Sn
3: else Let e = minMatchLabel(n);
4: Return a set of elements a that is an ancestor of e such that a can match node b item in
the path solution of e to path pattern pn
11

4.4.3. Computation of the best results
After the matching by the function flexPrefTreeMatch, the weighted solution table is constructed
using the set of the top branching node. The weighted solution table contains only the solutions
that integrate at least the obligatory constraints. It associates to each solution its corresponding
preference weight. The solutions are sorted by increase order of weight. In fact, the weight of
every solution of the final outputList is the sum of all the integer (weight of the children subtree)
of its corresponding intVector. The solutions are inserted in the table with the respect of theirs
weight values in order to avoid a sorting operation, so that at the end of the insertion, the table is
already sorted. The top-K best solutions (with the highest preference values) are returned to the
user. The next section shows some illustrations of different FPTPQ execution process.
5. Illustration of the evaluation of FPTPQ by treeMatch
5.1.Illustration of flexPrefTreeMatch for queries with flexible nodes.
Function prefSatisfyTreePattern(eqi, eq)
1: if (intVector(eq, eqi) >=0) return true;
2: else return false;
Procedure prefUpdateSet(q; e)
1: prefCleanSet(q, e);
2: add e to set Sq; //set the proper intVector(e) using the locate match table.
3: if (¬isRoot(q) ˄ (minValue(intVector(e))>= 0)) then
prefUpdateAncestorSet(q);
Procedure addToOutputList(q, eqi)
1: for each eq ϵ Sq do
2: if (prefSatisfyTreePattern(eqi, eq)) then outputList(eq). add (eqi);
Procedure prefCleanSet (q; e)
1: for each element eq ϵ Sq do
2: if (prefSatisfyTreePattern(eq,e))
3: if (q is a return node)
4: addToOutputList(NAB(q), e);
5: if (isTopBranching(q)
6: if (there is only one element in Sq)
7: output all elements in outputList(eq);
8: else from the set Sq construct the weighted solution table
9: delete eq from set Sq;
Procedure prefUpdateAncestorSet(q)
1: /*assume that q′ = NAB(q)*/
2: for each e ϵ Sq′ do
3: if (intVector (e, q) = -1) then
4: intVector(e, q) = prefValue(e, q);
5: if (¬isRoot(q) ˄ (minValue(intVector(e))>= 0))
6: prefUpdateAncestorSet(q');
12

Figure 4: Illustration: evaluation of a FPTPQ that has a flexible node
Initially, the procedure PrefLocateMatchLabel locate B1 for the node 2 and for the node 3, C1
and D1 are located, but C1 is selected first because it has the minimum matchLabel. The function
prefGetNext return (3, C) because C1 is deeper than B1. Then A2 is added in S1 with the intVector
“-10” to indicate that at this moment, the second child of A2 has been found. The cursor of TA is
advanced. At the next stage, B1 and D1 are read by PreflocateMatchLabel and (2, B) is returned
by PrefGetNext, A1 is added in the set with the bit vector "0-1" and outputList <0.0> (since B is
a return node); the cursor of TB is advanced. Early in stage 3, B2 and D1 are read and PrefGetNext
return (2, B) because B2 is the deepest and has the minimum matchLabel. B is also a return node,
so B2 is added in the outputList of its corresponding matchLabel ancestor element (A2). The
intVector of A2 is updated from "-10" to "00". The last element to be read is D1, and the set S1 is
updated. That is the intVector of A is update from “0-1” to “00”. The query does not have any
preference node, as the label items of the same flexible node, C and D are of equal values. A node
(of id -1) is created to merges all the outputList of the set S1: S-1 = {<-1, "0", (0.0, 0.1.0)>}. The
set {B1(0.0), B2(0.1.0)} is returned. Notice that B1 is also returned only because of a replacement
label item (D) has been added to the FPTPQ for the node 3. Using the Extended Tree Pattern, the
user may have been partially satisfied or would have been obliged to write and execute almost the
same query, replacing C with D as the label of node 3.
5.2. Illustration of flexPrefTreeMatch: evaluation of queries with ordered
preference node label items.
The Figure 5-a shows an example of a FPTPQ which has two ordered flexible preference nodes,
lets describe its evaluation process using the algorithms treeMatch. First of void the query nodes
are numbered (Figure 5-d), the labels Id are minimized (since the item C does not exist in the
database, it is deleted) and the preference locate match table of Table 2 is constructed from it. The
sets S1 and S4 are associated respectively to the branching nodes 1 and 4. The intVectors in blue
shows the preference weight values of each path, calculated from the leaf matchLabel to the
associated branching node matchLabel. In the first stage, B1, F1 and H1 (H1 is selected before
G1 because it has the minimum match Label by lexicographical order) are read by
prefLocateMatchLabel. A1 is added to the set S1 with “0.0.0” (matchLabel of B1) in his
outputList, since B is the return node of the query. At this moment A1, only the first child of A1
has been read, Its corresponding intVector is “0,-1”. E1 is added to the set S4 with the intVector
“0,1” whose guarantee that E1 matches all its corresponding subtree, since E1 carries two children
F1 which is not a preference node item, and H1 whose preference value is 1. Therefore, the
intVector A1 (the corresponding NAB matchLabel) is updated to “0,3”. The integer 3 is the
preference value of the subtree rooted by D1. Later, F2 is read and it corresponding NAB (E2) is
inserted in S4 with the intVector “0,-1”. When F2 is read, intVector(E2) become “0,1” since E2
has his two children and the second is the item of the preference node 6, whose preference weight
is equal to “1”. Later when B2 is read, A2 is added to the set S1 with the intVector “0,1” and B2
in its outputList. Afterwards F3, G1, B3 are read in this order, followed by the insertion of
13

“0.2.2.0” (matchLabel of E3) in S4 and the insertion of “0.2” (matchLabel of A3) in S1
respectively with the intVectors ‘0,2’ and ‘0,4’. Then, F4 and B4 are read, E4 is added in S1 with
the intVector ‘0,0’, even if it does not have a child G or H. The reason is that node 6 is a preference
node, its satisfaction is not compulsory in a solution tree. A4 is added to S1 with the intVector
“0,0” since it does not carry any preference node item. Finally, B5 is read, and its NAB A5 is
added to S1 with the intVector “0,-1”. The execution is stopped, since all the elements of the input
list associated to the query leaf label items have been read.
At the end of the matching process, the weighted solutions table of Table 3. is constructed from
the top branching node set S1. We can see that the solution 0.2.0. (B3) has the greatest preference
value and can then be considered as the best (most preferred) solutions. The solutions are printed
per order of user preference. The possibility can be given to the user to precise the number (K) so
that the top-K solutions will be returned.
With these illustration examples, we can show mayor differences between The FPTPQ and the
other tree pattern queries. The replacement items enable more flexibility and more satisfiable
results for query, than the existing most famous and featured TPQ like GTP [25] and the extended
tree pattern query [18]. In fact, since they do not allow replacement items, the user would have
written a query with only the first item of each flexible node. The consequence is that no solution
would have been returned for the example of figure 3, since the database does not have any node
(tag) named “C”. Without the replacement items H and D, only the solution B4 which has the
lowest solution weight would have been returned. Moreover, if node 3 was not a preference node,
Figure 5. Illustration: evaluation of a FPTPQ with ordered preference node label items.
Table 2: Preference locates match table for the
FPTPQ of Fig.6.
Leaf
node
id
Label
items
Root to
leaf paths
preference at
the level of
the NAB “1”
preference at
the level of
the NAB “4”
2 B A/B 0 -
5
F
A/D/E/F 2 0
A/E/F 0 0
G
A/D/E/G 2 + 2 2
A/E/G 0 + 2 2
6
H
A/D/E/H 2 + 1 1
A/E/H 0 + 1 1
E
A/D/E 2 + 0 0
A/E 0 + 0 0
Table 3: Weighted solutions table for the
example of Figure 5.
Solution id name weight position
0.2.0. B3 4 1
0.0.0 B1 3 2
0.1.0 B2 1 3
0.3.0 B4 0 4
14

B4 would not have been selected. Unless all the flexible nodes were replaced by the wildcard “*”
which is responsible of a many useless solutions. This shows how the combination of replacement
items and preference nodes contribute to ensure user satisfiability via the proposition of numerous
solutions that remain close to the user needs.
Since none of the existing tree pattern query allow flexibles nodes, flexible preference nodes and
flexible ordered preference node like describe in section 3.2. If We can assert that a new class of
query have been created, the class of query which contain these three types of nodes, we cannot
yet assert that flexPrefTreeMatch has a new optimality class. To do so, further analysis need to
be done. Our main purpose was to propose the PFTPQ alongside with evaluation algorithms.
6. Experimentations: evaluation of FPTPQ with FlexPrefTreeMatch
In this section, we compare the evaluation results of the FPTPQ against the evaluation results of
a TPQ written for the same need. We used the XML datasets DBLP (regular structure imposed
by a DTD), Treebank (irregular, with no DTD), uwm.xml and wsu.xml which are two databases
with different structures that was merged to produce merged_wsu-uwm.xml which suffer of
structural and vocabular heterogeneity. These datasets are available on the University of
Washington XML Data repository [4]. In fact, the FPTPQ help to express all the desired
information set through replacement items, contrary to the other TPQ like the GTP [25] or the
extended tree pattern query [18] whose execution may result to an empty or incomplete solution
set. Moreover, these TPQ may originate many useless solutions through the utilization of the
wildcard “*” and the A-D (“//”) operator. This experiment is used none to compare algorithms
execution times, but to show how the FPTPQ and FlexPrefTreeMatch may be effectively used to
express queries which produce only or the complete set of desired solution, when the replacement
items are well inserted.
The percentage (P) of useless solution is calculated based on the exact number of available
solutions. Thus, if the number of returned useless solution is greater, P will be greater than 100%.
6.1. FPTPQ contribution for the searching of all the needed solutions.
Table 4. Comparison of FPTPQ (with appropriate replacement items) evaluation results with the
evaluation results of classic TPQ.
N°
XML data
(.xml)
Evaluation results of TPQ with
only one label
Evaluation results of FPTPQ with
appropriate replacement items
Number
of
lacking
solutions
% of
lacking
solutions
Tree Pattern Query
(TPQ)
Number
of
solutions
FPTPQ with appropriate
replacements items
Number
of
solutions
1
merged_wsu-
uwm
TQ1: root//credit 3924 FQ1: root//(credit|credits) 6036 2112 34,99%
2
merged_wsu-
uwm
TQ2: root//sect 3924 FQ2: root//(sect|section) 8499 4575 53,83%
3
merged_wsu-
uwm
TQ3: root/course_listing
[//hours/start]//instructor
45 75
FQ3: root/(course_listing|course)
8499 3924 46,17%
4
merged_wsu-
uwm
TQ4:
root/course/place/bldg
3924
FQ4: root/((course/place)
|(course_listing//bldg_and_rm))/bldg
8499 4575 53,83%
5 dblp TQ5: dblp/book/title 845 FQ5: dblp/(book|article)/title 112454 111609 99,25%
6 dblp TQ6: dblp/book/author 1153
FQ6:
dblp/(mastersthesis|book)/author)
1158 5 0,43%
7 dblp
TQ7:
dblp/mastersthesis/author
5
FQ6:
dblp/(mastersthesis|book)/author)
1158 1153 99,57%
15

The FPTPQ through replacement items can help to completely express the user needs inside one
query and therefore, reduce multiple query adjustment and execution. As query example, let’s
consider a user who wants to print the credit of all the courses present in the document
merged_wsu_uwm.xml (the merged result of wsu.xml and uwm.xml). Because the occurrences
of credit are represented inside two differences tags (credit and credits), the query TQ1 of Table
4 will produce only 3924 titles over the needed 6036 that are returned by the FPTPQ FQ1. The
same issue is caused by the query TQ2 that print 3924 section nodes over the available 8499 that
are completely returned with the execution of the FPTPQ FQ2.
Let consider now a user who want the titles of all the books and articles of dblp.xml. With the
TPQ TQ5, only the titles of books (only 845 solutions over 112454 needed) will be returned, and
the user will be obliged to write another query, replacing “book” by “article” in order to obtain
the other 112454 titles. Only one execution is need with the query FQ5 to produce the complete
set of solutions. The last column of Table 4 show that an appropriate utilisation of replacement
items in a FPTPQ has helped to select more than 99% of needed solutions that have not been
selected with the utilization of the classic TPQ.
6.2. The FPTPQ help to avoid useless solutions caused by the wildcard “*”
During the evaluation of a TPQ, the wildcard can be matched by any single node. Consequently,
it causes many useless intermediate results and unsatisfiable solutions. When the user is aware of
all the words used to express all the instances of an object (document tag), it better to use them
rather than “*”. The Table 5 shows FPTPQ (with appropriate multiple label items) evaluations
results compare to TPQ where “*” have been used in the place of the multiple items. In the
document merged_wsu_uwm.xml, courses are now described inside two tags: course (from
uwm.xml file) and course_listing (from wsu.xml file).
Table 5: Avoiding useless solutions with FPTPQ over TPQ with wildcard "*".
N°
XML
database
(.xml)
Execution of TPQ with the wildcard
(“*”)
Corresponding FPTPQ with appropriate
replacement items.
Useless
solutions
caused
by “*”
% of
useless
solutions
caused
by “*”
TPQ with "*"
Number
of
solutions
FPTPQ with replacement items.
Number
of
solutions
1
merged_wsu-
uwm
SQ1: root/*[//hours/start]//
instructor
8499
8499 0 0%
2
merged_wsu-
uwm
SQ2: root/*//*/bldg 8499
FQ4: root/((course/place)|(course_
listing//bldg_and_rm))/bldg
8499 0 0%
3
merged_wsu-
uwm
SQ3: dblp/*/title 328859 FQ5: dblp/(book|article)/title 112454 216405 192%
4 dblp SQ4: dblp/*/author 716488 FQ6: dblp/(mastersthesis|book)/author 1158 715330 61773%
5 Treebank_e SQ5: //PP[//VP/IN]/*/VBN 676 FQ7: //PP[//VP/IN]/(ADJP|VP)/VBN 96 580 604%
6 Treebank_e SQ6: //*[//VP/IN]/*/VBN 28314
FQ8: //(PP|SBARQ)[//VP/IN]/
(ADJP|VP)/VBN
89 28 225 31713%
7 Treebank_e SQ7: //*[//VP/IN]/NP 435689 FQ9: //(ADJP|NP)[//VP/IN]/NP 98352 337 337 343%
Replacing the wildcard “*” of the query SQ1 by these two tags names to obtain FQ3 will originate
only the needed solutions, since no other tag can be matched as a course. But the other tree pattern
queries of Table 5. show how “*” caused overabundant useless solutions, even for simple queries.
In the merged document, not only the courses can be scheduled in a building (bldg). Different
sections of the same course can be scheduled in different building. With the operator “*”, a user
who need only the buildings of courses may be obliged to find between other useless ones (the
building of section). Moreover, preference operator (course_listing|course)<!,
16

course|course_listing)>!) can be added to give a privilege to the buildings which are carried by
the item (tag) “course”. i.e., the building of the UWM courses. In dblp dataset, the cited
documents can be articles, improceedings, master thesis, PhD thesis and books. The query SQ4
is written using the extended tree pattern query for a user who wants only the titles of books and
articles. “*” causes the selection 216405 (328859 - 112454) unneeded ones.
The utilisation of FPTPQ to write query FQ5 (by replacing “*” with book|article) help the
FlexPrefTreeMatch to return only the needed solutions (only the titles of books and articles).
Queries SQ5, SQ6, SQ7 produce a huge number of useless solutions. This is because of the
heterogeneity of the dataset Treebank.
Table 6. Advantages of replacement items over the A-D operator
N°
XML
database
Result of TPQ with A-D (//)
Relations
Results of FPTPQ with replacement items in the
place of “//”
Number
of useless
solutions
caused by
“//”
% of useless
solutions
caused by
“//”
Tree pattern query
with “//”
Number
of
solutions
FPTPQ with needed replacements
items
Number
of
solutions
1
merged_
wsu-uwm
AQ1:
root[//hours/start]//
instructor
8499
[//hours/start]// instructor
8499
0 0%
2
merged_
wsu-uwm
AQ2: root//bldg 8499
FQ4: root/((course/place)|
(course_listing//bldg_and_rm))/bldg
8499 0 0%
3 dblp AQ3: dblp//title 328859 FQ5: dblp/(book|article)/title 112454 216 405 192%
4 dblp AQ4: dblp//author 716488 FQ6: dblp/(mastersthesis|book)/author) 1158 715 330 61773%
5 Treebank_e AQ5: //S/VP//NP/NNP 55288 FQ11: //S/VP/(VP|PP)/NP/NNP 6518 48 770 748%
6 Treebank_e
AQ6:
//PP[//VP/IN]//VBN
6262 FQ7: //PP[//VP/IN]/(ADJP|VP)/VBN 96 6 166 6423%
The chart of Figure 6 show the comparison result of the number of solutions returned by TPQ that
have “*” and FPTPQ where “*” have been substituted by the appropriate replacement items. The
FPTPQ contains only the needed and appropriate items. Line 1 and 2 of the table show that the
number of solutions returned by the TPQ and the FPTPQ are equals. This is because the two
needed nodes label “course_listing” and “course” are the only possible ones which can be the
image of “*”. The query FQ11 shows how a FPTPQ can avoid more than 61773% of useless
solutions caused by “*” (if the TPQ SQ4 is used) in highly irregular databases.
Figure 6. Showing how FPTPQ help to avoid
useless solutions caused by “*”.
Figure 7. Showing how FPTPQ help to
avoid useless solutions caused by “//”.
17

6.3. Advantages of a proper utilization of replacement items in a FPTPQ rather
than A-D operators.
When the schema is absent, a distance between XML nodes (number of node than are in-between)
may not be known, then the utilization of the ancestor-descendant operator (//) is justified. But an
inappropriate utilization of this operator causes the selection of solutions whose paths have
different lengths and different nodes labels. The execution of a query like A//B imply the selection
of A/B, A/*/B, A/*/*/B, …, where * can be anything. The execution of this operator produces
many useless results linked to unnecessary paths. The Table 6 shows how useless intermediate
solutions can be totally avoided when the replacement items are correctly inserted inside the
FPTPQ. The execution of the query AQ3, AQ4, AQ5 and AQ6 produce a huge quantity of useless
solutions. AQ4 induce 61773% (calculated base on the number of needed solution) of useless
solutions. This huge quantity of useless solutions is due to the fact that all the database instance
of the return node are read with not enough information to filter them. In fact, there are 328859
documents in dblp and the operator “//” of the query AQ3 (resp AQ4) allows the selection of their
title (resp of all the authors), no matter the type of document. When only the titles (resp authors)
of books and articles are needed, it is preferable to use query FQ5 (resp FQ6). The execution of
AQ1 and AQ2, do not generate useless solutions because all the instructors of the database are
needed (the number of instructors needed is equal to the number of instructors replaced by //).
The chart Figure 7 shows the higher percentage of useless solutions engender by “//”. We can see
that four of the six queries produce over 150% of useless solutions. Query AQ4 engendered more
than 60000% of useless solutions. The consequence of such enormous quantity of useless solution
is that it may confuse the user and make him abandon its searching.
To conclude this section, we can assert that a proper utilization of replacement items in a FPTPQ
help to express all the needed solution, and avoid incomplete results that are caused by TPQ, since
they do not allow multiple items (terms) to represent all the names of the difference tag instances.
Even if the operators “*” and “//” remain useful, mainly when the database schema is unknown,
they cause huge number of unsatisfiable solutions. When the replacements items are known and
are added in the correct places of the FPTPQ, unsatisfiable solutions are totally avoided, and only
the useful solution set is returned.
7. Conclusion
Obtaining satisfiable queries for XML databases that have structural and vocabular heterogeneity
remain an important challenge. We proposed the Flexible Preference Tree Pattern Query
(FPTPQ), a TPQ that allows to have multiple items as node label and multiples paths to locate the
same query node, ordering and preference operators. The FPTPQ can be used in any XML
database, to characterize in the same query the user both first choice solution and alternatives
ones. The FPTPQ enhance the satisfiability of both preference and non-preference nodes. For the
evaluations of FPTPQ queries, we proposed the holistic algorithm FlexPrefTreeMatch which
match the FPTPQ with the database index based on extended Dewey labelling scheme, while
calculating each solutions preference weight. Illustrations and experimentations verify the
effectiveness of the FPTPQ and the correctness of the algorithm flexPrefTreeMatch. More type
of flexibilities and preferences are being integrated in our project, to improve XML query
languages.
18

References
[1] D. U. a. E. Z. Luigi Pontieri, "An approach for the extensional integration of data sources
with heterogeneous representation formats," Data & Knowledge Engineering, vol. 45, pp.
p291-331, 2003.
[2] S. B. S. C. V. D. A. A. F. F. G. e. a. Domenico Beneventano, “Semantic Integration and
Query Optimization of Heterogeneous Data Sources,” in Advances in Object-Oriented
Information Systems, 2002.
[3] L. G. DeMichiel, "Resolving database incompatibility: an approach to performing
relational operations over mismatched domains," Vols. 1, p 485-493, 1989.
[4] U. o. W. X. D. Repository, 14 March 2020. [Online]. Available:
http://aiweb.cs.washington.edu/research/projects/xmltk/xmldata/www/repository.html#a
uctions.
[5] L. L. a. J. L. R. Rada CHIRKOVA, "Tractable XML data exchange via relations,"
Frontiers of Computer Science, pp. 243-263, 01 06 2012.
[6] D. C. L. L. a. M. F. Amano Shun'ichi, "On the tradeoff between mapping and querying
power in XML data exchange," 09 2010.
[7] A. a. H. Ahmad, "XML-Based Data Exchange in the Heterogeneous Databases
(XDEHD)," International journal of Web & Semantic Technology, vol. International
journal of Web & Semantic Technology, pp. 11-24, 07 2015.
[8] K. S. a. M. Peter, "Integrating Unnormalised Semi-structured Data Sources," 2005.
[9] A. A. a. P. Jaroslav, "A Mediation Layer for Heterogeneous XML Schemas.," IJWIS, vol.
1, pp. 25-32.
[10] C. N. a. C. R. Tekli Joe, "Building Semantic Trees from XML Documents," Journal of
Web Semantics, vol. 37, 03 2016.
[11] T. Mohammad, "Understanding Semantic Web and Ontologies: Theory and
Applications," Journal of Computing, 06 2010.
[12] C.-B. N. C. F. a. R. O. Dernaika Farah, "Semantic Mediation for A Posteriori Log
Analysis," in ARES '19: Proceedings of the 14th International Conference on Availability,
Reliability and Security, 082019.
[13] H. M. a. D. Jérôme, "A Survey of XML Tree Patterns," EEE Transactions on Knowledge
and Data Engineering, vol. 25, pp. 29-46, 01 2013.
[14] S. B. Sven Groppe, "Query Reformulation for the XML standards XPath, XQuery and
XSLT," in DBLP, Berlin, January 2004.
[15] H. B. a. O. K. Saber Benharzallah, "Reformulating XQuery queries using GLAV mapping
and complex unification," vol. Volume 28, no. 1, January 2016.
19

[16] P. Y. B. C. a. S. S. B. a. J. Xu, "AutoG: A Visual Query Autocompletion Framework
forGraph Databases," Vols. 26, 347-372, 2017.
[17] Maurice Tchoupe Tchendji and Brice Nguefack, "Requêtes XPath bipolaires et
évaluation," Revue Africaine de la Recherche en Informatique et Mathématiques
Appliquées, INRIA., 2017.
[18] T. W. L. Z. B. a. C. W. J. Lu, "Extended XML Tree Pattern Matching: Theories and
Algorithms," in IEEE Transactions on Knowledge and Data Engineering, vol. 23, no. 3,
March 2011.
[19] W, "Extensible Markup Language (XML) 1.0 (Fifth Edition)," [Online]. Available:
https://www.w3.org/TR/xml/. [Accessed 12 08 2020].
[20] M. M. G. G. a. R. B. L. Ismael Sanz, "Fragment-based approximate retrieval in highly
heterogeneous XML collections," Data Knowl. Eng., vol. 64, pp. 266--293, 2008.
[21] W3C, "XML Path Language (XPath) 3.0," 08 04 2014. [Online]. [Accessed 05 04 2021].
[22] W3C, "XQuery 3.0: An XML Query Language," 08 04 2014. [Online]. Available:
https://www.w3.org/TR/xquery-30/. [Accessed 04 03 2021].
[23] L. L. S. D. T. K. Jagadish H.V., "TAX: A Tree Algebra for XML," Berlin, Heidelberg,
2002.
[24] A. N. a. H. V. Jagadish, "Evaluating Structural Similarity in XML Documents," WebDB,
pp. 61--66, 2002.
[25] H. V. J. L. V. S. L. a. S. P. Zhimin Chen, "From tree patterns to generalized tree patterns:
on efficient evaluation of XQuery," in In Proceedings of the 29th international conference
on Very large data bases (VLDB '03), 2003.
[26] W. Y. L. L. V. S. a. J. H. V. Paparizos Stelios, "Tree Logical Classes for Efficient
Evaluation of XQuery," 2004.
[27] H. T. a. H. M. S. Izadi Sayyed Kamyar, "S3: Evaluation of Tree-Pattern XML Queries
Supported by Structural Summaries," vol. 68, p. 126–145, 2009.
[28] G. K. Werner Kießling, "Preference SQL — Design, Implementation, Experiences," Hong
Kong, China, August 20-23, 2002.
[29] H. B. F. S. H. S. Kießling W., "Preference XPATH:A Query Language for E-Commerce,"
Heidelberg, 2001.
[30] S. C. a. M. Shiloach, "Flexible XML Querying Using Skyline Semantics," in Proceedings
of the 25th International Conference on Data Engineering, {ICDE} 2009, March 29 2009
- April 2 2009, Shanghai, China, 2009.
[31] L. T. a. T. T. T. Maurice Tchoupé Tchendji, "A Tree Pattern Matching Algorithm for
{XML} Queries with Structural Preferences," CoRR, vol. abs/1906.03053, 2019.
20

[32] M. T. T. a. P. J. Kenfack, "An XQuery Specification for Requests with Preferences on
XML Databases," vol. 582, no. {IFIP} Advances in Information and Communication
Technology, pp. 120--130, 2020.
[33] Y. T. a. G. F. a. a. B. S. Papadias, "Progressive skyline computation in database systems,"
ACM Trans, pp. 41–-82, 2005.
[34] S. a. K. D. a. S. K. Borzsony, "The Skyline Operator," in Proceedings - International
Conference on Data Engineering, 2001.
[35] C. K. a. T. Tzouramanis, “A Survey of Skyline Query Processing,” ArXiv, p.
abs/1704.01788, 2017.
[36] K. B. P. I. a. V. Shanthi, "Goal Directed Relative Skyline Queries in Time Dependent
Road Networks," International Journal of Database Management Systems ( IJDMS ), vol.
4, no. 2, pp. 01-12, 2012.
AUTHORS
Brice Nguefack received his master degree in computer science in
2016, from the department of mathematics and computer science,
faculty of science, University of Dschang, Cameroon. Where he is
actually a Ph.D. student. His research interests include Software
Engineering, integration of preferences in XML query language, and
XML preference query processing. (https://orcid.org/0000-0001-
5241-2750)
Maurice Tchoupé Tchendji received his Ph.D degree in computer
science in 2009, from the University of Yaounde 1, Cameroon and
the University of Rennes 1, France. He is actually a senior lecturer in
the department of mathematics and computer science, Faculty of
Science, University of Dschang, Cameroun. His research interests
include Distributed Computing, Theory of Computation, Software
Engineering, integration of Preferences in XML query language, and
XML preference query processing. (https://orcid.org/0000-0002-9208-6838)
Thomas Djotio Ndie received his Ph.D degree in computer science,
in 2009, from National Advanced School of Engineering, University of
Yaounde 1, Cameroon. He is actually an Associate Professor in the
same University. He is also the team leader of the LIRIMA project
name “Internet of Things for Developing countries”. His research
interests include Software Engineering, Wireless Mesh Network,
Wireless Sensor Network, Intrusion Detection System, Open Wireless
Access for Developing Countries. (https://orcid.org/0000-0002-6300-6237)
21

Considering Structural and Vocabulary Heterogeneity in XML Query: FPTPQ and Holistic Evaluation.

More Related Content

Similar to Considering Structural and Vocabulary Heterogeneity in XML Query: FPTPQ and Holistic Evaluation. (20)

Recently uploaded (20)

Considering Structural and Vocabulary Heterogeneity in XML Query: FPTPQ and Holistic Evaluation.