KR101271277B1 - Methods for indexing for top-k queries - Google Patents

Abstract

Disclosed is a method of generating a tuple hierarchy according to an embodiment of the present invention. A method of generating a tuple hierarchy according to an embodiment of the present invention includes obtaining a skyline for a specific tuple set, and repeating until all tuples of the tuple set are obtained as skylines, thereby obtaining at least one skyline. Step 2 is a step of establishing a relationship by deriving tuples of other skylines dependent on each tuple of each skyline obtained in the first step, and obtaining a convex skyline for each skyline obtained in the first step. Repeating until all tuples of each skyline are obtained as convex skylines, thereby obtaining at least one convex skyline for each skyline and each tuple of each convex skyline obtained in the third step A fourth step of establishing a relationship by deriving tuples of other subordinate convex skylines The. Therefore, according to the present invention, efficient query processing is possible by processing a ranking query according to a dependency relationship based on a dual hierarchy structure using a skyline and a convex skyline.

Description

How to process ranking queries {METHODS FOR INDEXING FOR TOP-K QUERIES}

The present invention relates to a method for processing a ranking query, and more particularly, to a method of generating a tuple hierarchy for processing a ranking query and a query processing method for finding the top K tuples according to user preferences.

Conventional techniques for efficiently processing rankings and queries can be classified into three types as follows. First, as a method of processing ranking queries based on a hierarchical structure, tuples are accessed in a hierarchical unit by establishing a relationship for each hierarchical structure. [Reference 1] Y.-C. Chang, L. Bergman, V. Castelli, C.-S. Li, M.-L. Lo, and J. R. Smith, “The onion technique: Indexing for linear optimization queries” in Special Interest Group on Management on Data, pp. 391-402 and 2000 proposed an efficient ranking query processing method based on hierarchical index structure in database system. This method is based on the Convex hull used in geometry as a single structure and uses only the relationships between layers. [Ref. 2] D. Xin, C. Chen, and J. Han, “Towards robust indexing for ranked queries” in Very Large Databases, pp. 235-246 and 2006 proposed a methodology for reducing the number of tuples that make up a hierarchy in order to make the hierarchical approach more efficient.

The second method is to construct a layer and then use a tuple relationship between adjacent layers. [Ref. 3] L. Zou and L. Chen, “" Dominant graph: An efficient indexing structure to answer top-k queries ”in IEEE International Conference on Data Engineering pp. 536-545, 2008 The relationship between tuples in adjacent layers is constructed based on the dependency relationship used in the skyline.This method handles the ranking query more efficiently than the existing hierarchy-based methods. can do.

The third is how to process ranking queries based on the list structure. This method is to find the top K tuples in the ranking query by sorting for each attribute value and accessing them in sorted order when the data is organized in a list according to each attribute. [Reference 4] R. Fagin, “Combining fuzzy information from multiple systems” in International Symposium on Principles of Database Systems, pp. 216-226, 1996, proposed a method for processing ranking queries based on list structure for the first time in database systems. [Ref. 5] R. Fagin, A. Lotem, and M. Naor, “Optimal aggregation algorithms for middleware ”, Journal of Computer and System Sciences, vol. 66, no. 4, pp. 614–656, 2003 improved [Ref. 4] and proposed a method for processing ranking queries by accessing fewer tuples only. .

Finally, the existing calculated ranking query is regarded as the concept of a view widely used in a database system, and a ranking query is processed by combining the results of the existing ranking query. [Ref. 6] V. Hristidis, N. Koudas, and Y. Papakonstantinou, “Prefer: A system for the efficient execution of multi-parametric ranked queries” in Special Interest Group on Management on Data, pp. 259? -270, 2001 first proposed a method to process the ranking queries by viewing the results of the existing ranking queries in the form of views for the first time in a database system. [Reference 7] G. Das, D. Gunopulos, N. Koudas, and D. Tsirogiannis, “Answering top-k queries using views” in Very Large Databases, pp. 451? -462, 2006 propose a method to compute a given ranking query by combining existing views.

However, these conventional methods have a problem in that the inefficiency caused by accessing unnecessary tuples cannot be eliminated when processing ranking queries.

An object of the present invention for solving the above problems is to provide a method of generating a tuple hierarchy for avoiding unnecessary tuple access and processing an efficient ranking query.

Another object of the present invention for solving the above problems is to provide a ranking query processing method for avoiding unnecessary tuple access and efficient processing of ranking queries.

An embodiment of the present invention for achieving the above object, the first to obtain a skyline for a specific set of tuples, and to obtain at least one skyline by repeating until all the tuples of the set of tuples is obtained as a skyline Step 2 is a step of establishing a relationship by deriving tuples of other skylines dependent on each tuple of each skyline obtained in the first step, and obtaining a convex skyline for each skyline obtained in the first step. Repeating until all tuples of each skyline are obtained as convex skylines, thereby obtaining at least one convex skyline for each skyline and each tuple of each convex skyline obtained in the third step A tuple hierarchy comprising a fourth step of establishing a relationship by deriving tuples of other dependent convex skylines Provides a structure creation method.

Another embodiment of the present invention for achieving the above object is a method of clustering a specific tuple set that is classified into at least one group according to the distribution of attribute values and a relationship is established between dependent tuples, the highest formed first among the groups Deriving a group, clustering the most significant group to generate at least one cluster, generating a virtual tuple having the best attribute value for each of the generated clusters, and in the second step A tuple clustering method includes a third step of deriving tuples of the uppermost group subordinate to each generated virtual tuple and establishing a relationship.

An embodiment of the present invention for achieving the above another object is a method of processing a ranking query according to user preference for a specific tuple set is classified into at least one group according to the distribution of the attribute value and the relationship is established between the dependent tuples The first step of accessing the highest group among the groups, putting tuples of the group into a priority queue, selecting a tuple having the highest user preference from the priority queue, adding to the result array and removing from the priority queue. And a third step of accessing tuples dependent on the tuple selected in the second step among the tuples in the specific tuple set, and putting the accessed tuples into the priority queue. And repeating the third step, wherein the number of tuples in the result array is determined after the second step. It provides a method of processing a ranking query, characterized in that iterates until a certain number is reached.

By using the method of generating a tuple hierarchy and a ranking query processing method according to the present invention as described above, a hierarchical structure of a double layered tuple is created based on a skyline and a convex skyline, and a double hierarchy. By setting the relationship between tuples between adjacent hierarchies based on the structure, efficient query processing is possible by searching the top K tuples along the path based on the relationship between tuples based on the hierarchy when processing ranking queries based on user preferences. It's effective. In addition, by adding a virtual tuple by clustering the above-described dual hierarchical structure, it is possible to perform faster query processing by shortening an access path of tuples when querying for tuple search.

1 is a database schema illustrating a tuple set for explaining an embodiment of the present invention.
FIG. 2 is a graph illustrating a result of obtaining a skyline and a convex skyline with respect to the tuple set of FIG. 1.
3 is a flowchart illustrating a process of generating a hierarchy of tuples according to an embodiment of the present invention.
FIG. 4 is a tuple hierarchy diagram illustrating a result of applying the hierarchical structure generating method according to the present invention to the tuple set of FIG. 1.
5 is a flowchart illustrating a process of processing a ranking query according to an embodiment of the present invention.
6 is a data structure diagram showing data generated in each process of processing a ranking query according to an embodiment of the present invention.
7 is a flowchart illustrating a tuple clustering process according to an embodiment of the present invention.
8 is a graph illustrating a process of applying a tuple clustering method according to an embodiment of the present invention.
FIG. 9 is a tuple hierarchy diagram illustrating a result of applying a tuple clustering method according to an embodiment of the present invention to the tuple hierarchy structure of FIG. 4.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, A, B, etc. may be used to describe various elements, but the elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

When a component is referred to as being "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in between. Should be. On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and are to be interpreted as either ideal or overly formal in the sense of the present application Do not.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In order to facilitate the understanding of the present invention, the same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

First, a skyline, a convex skyline, and a dependency relationship are defined as concepts that are basically used to explain a method of generating a tuple hierarchy according to an embodiment of the present invention.

Table 1 below is a table for explaining the terms used while describing one embodiment of the present invention.

1) Skyline

A skyline may mean a set of tuples that are not dependent on a set of attributes when a set S of all tuples is given. For the set S of all tuples, the skyline is denoted as SKY (S).

2) Convex Skyline

The convex skyline may mean a set of tuples having the highest priority for any user's preference expressed as a linear combination function in the skyline. The convex skyline is denoted CSKY (S) for the set S of all tuples.

3) dependency

On the other hand, the fact that any tuple T depends on another tuple T 'means that T has a value equal to or less than all of the attribute values of T', and T for any one attribute. It means to have a smaller value. Such dependency may be consistently applied according to the direction of the maximum value MAX and the minimum value MIN in the numeric data according to the user's preference.

Hereinafter, an example of obtaining a skyline and a convex skyline will be described with reference to the drawings.

1 is a database schema illustrating a tuple set for explaining an embodiment of the present invention.

FIG. 2 is a graph illustrating a result of obtaining a skyline and a convex skyline with respect to the tuple set of FIG. 1.

Referring to FIG. 1, a total of 11 tuples are shown, and each tuple has attribute values according to the identifier 110, the attributes D1 120, and the attributes D2 130.

In addition, referring to FIG. 2A, the skylines {a, f, b, g obtained as a result of obtaining a skyline having coordinates D1 120 and D2 130 with respect to the tuples of FIG. , c} (210). At this time, d, e, h, i, k are dependent on a. That is, the tuples described above cannot have a better value than a in any user preference function. And j is dependent on b. This set of nondependencies is the skyline, which is {a, f, b, g, c}.

In addition, referring to FIG. 2A, as a result of obtaining a skyline with respect to the remaining tuples {d, e, i, j, h, k}, the skyline {d, e, i, j} 220 is obtained. As a result of obtaining the skyline with respect to the remaining tuple {h, k}, the coarse layer formed by acquiring the skyline {h, k} 230 is shown. Detailed description thereof will be described later.

Meanwhile, referring to FIG. 2B, the skyline {a, b, c} obtained as a result of obtaining a convex skyline having coordinates D1 120 and D2 130 with respect to the tuples of FIG. 1. Shows (240). It may mean that no tuple in any user preference function has a better value than any of a, b, and c.

In addition, as a result of obtaining the convex skyline with respect to the remaining tuples {d, e, f, g, h, i, j, h, k}, the skyline {d, f, g} 250 is obtained and again. As a result of repeatedly calculating the convex skyline with respect to the remaining tuples}, the fine layers formed by obtaining the skylines {e, j} 260, {h, i} 270, and k {280} are shown.

4) full dependence

Full dependence means that when two tuples T ∈ Li, T '∈ L (i + 1) are given and T is dependent on T', T is totally dependent on T '.

Referring to Figures 1 and 2 (a), a has a better value for all attributes than tuples d and e of adjacent coarse layers 220, where a is totally dependent on d and e. can do.

5) partially dependent

As described above, one layer Li refers to a coarse layer (for example, a layer formed of a skyline), and the coarse layer may be divided into several fine layers (for example, the layer formed of the convex skyline described above). . At this time, the sparse layer is represented by Li = {Li1, Li2, ... Lik}, and any fine layer is represented by Lij. Any fine layer Lij may represent a set of tuples.

When expressing any fine layer Lij as a convex skyline, each face constituting the convex skyline means a partial dependent set for tuples specific to Li (j + 1). The partial dependent set is denoted by EDS = {T1, T2, ..., Td}.

Given two tuples T ∈ Lij, T '∈ Li (j + 1), it means that T is partially dependent on T' when T is included in the partial dependency set on T '.

For example, as can be seen in Figures 1 and 2 (b), one of a and b is necessarily better than f for any user preference for the adjacent fine layer f. That is, a has a better value than f based on the attribute D1 120, and b has a better value than f based on the attribute D2 130. A subordinate subset represents a set of tuples that are subdependent, that is, {a, b} are subdependent to f and subdependent to f.

That is, the line connecting a and b in Fig. 4 (b) is located below f, in which case {a, b} becomes a subordinate subset of f, and one of a and b is always f regardless of user preference. It will have a better value. It can likewise be described that {b, c} is a subordinate subset for g.

As shown in FIG. 4 to be described later, in one embodiment according to the present invention, a partial dependency relationship appears in a fine layer in a sparse layer, and is indicated by a dotted arrow.

On the other hand, an embodiment of the present invention proposes a dual layer structure that defines a relationship between a skyline and tuples of adjacent layers obtained as convex skylines, that is, a coarse layer composed of a sparse layer composed of skylines and convex skylines in the sparse layer. . In this case, the sparse layer in the dual layer structure may be represented by the entire level L = {L1, L2, ..., Lk}, and any adjacent layer may be represented by Li, and may represent a set of tuples.

On how to create a dual hierarchy Example .

3 is a flowchart illustrating a process of generating a hierarchy of tuples according to an embodiment of the present invention.

FIG. 4 is a tuple hierarchy diagram illustrating a result of applying the hierarchical structure generating method according to the present invention to the tuple set of FIG. 1.

Referring to FIG. 3, a process of generating a hierarchical structure of tuples according to an embodiment of the present invention includes tuple set input step S310, skyline acquisition step S320, total dependency setting step S330, and convex sky. And a line acquiring step S340 and a partial dependency setting step S350.

3 and 4, each step of the process of generating a hierarchy of tuples according to an embodiment of the present invention can be described as follows.

The tuple set input step S310 may be a step of inputting a set of data to be used to generate a hierarchy of tuples according to the present invention in order to efficiently process a ranking query. The set of tuples may be tuples stored in a particular database or tuples extracted from a particular repository through any query.

Skyline acquisition step (S320) obtains a skyline for a specific tuple set obtained in the tuple set input step (S310), but repeats until all the tuples of the tuple set are obtained as a skyline at least one skyline It may be a step to obtain.

In particular, when obtaining a skyline, the skyline acquiring step (S320) acquires a skyline with respect to a specific tuple set received, and generates a skyline with respect to the remaining tuples not included in the acquired skyline among the specific tuple set. While obtaining, it is possible to obtain a hierarchy composed of skylines by repeating until all tuples of the specific tuple set are obtained as skylines.

For example, referring to FIG. 4, an L1 410 formed of {a, f, b, g, c} is formed by obtaining a skyline for the entire tuple. Next, the L2 420 formed of {d, e, i, j} is formed by obtaining a skyline for the remaining tuples except for the L1 410. The L3 430 formed of {h, k} is formed by obtaining a skyline for the remaining tuples except for the L1 410 and the L2 420.

The overall dependency establishing step (S330) may be a step of deriving tuples of other skylines totally dependent on each tuple of each skyline obtained in the skyline obtaining step (S320) and establishing a relationship. Meanwhile, another skyline may be a skyline nearest to each of the above skylines.

For example, as shown in FIG. 4, tuples of the skylines L1 410, L2 420, and L1 430 are connected by solid arrows. As described above, each tuple of each skyline is connected by a solid arrow with all subordinate tuples among the tuples of the nearest skyline. That is, as illustrated in FIG. 2, tuple a is completely dependent on tuples d and e while all attributes have better values than tuples d and e of adjacent coarse layers. In Fig. 4 this overall dependency is shown by a solid arrow.

Also, solid arrows ad, ai, and ae indicate that tuple a of L1 410 is totally dependent on tuples d, e, and i of L2 420, and solid arrows eh and ek are L2 420, respectively. The tuple e of may mean that the tuple e is entirely dependent on the tuples h and k of the L3 430. On the other hand, c can mean that there are no tuples that depend entirely on adjacent sparse hierarchies.

The convex skyline acquisition step (S340) is to obtain a convex skyline for each skyline obtained in the skyline acquisition step (S320), iterate until all tuples of each skyline is obtained as a convex skyline, each skyline It may be a step of obtaining at least one convex skyline with respect to.

That is, in the convex skyline acquisition step (S340), the convex skyline is obtained for each skyline, and the convex skyline is also obtained for the remaining tuples not included in the convex skyline, and all tuples of the skyline are convex skylines. Iterate over to obtain a layer composed of convex skylines.

For example, FIG. 4 shows the results of obtaining the convex skylines for the skylines obtained in the skyline acquisition step S340, respectively. That is, convex skylines L11 411 and L12 412 are obtained for the skyline L1 410, and convex skylines L21 421 and L22 422 are obtained for the skyline L2 420. It shows that convex skyline L31 431 has been obtained for skyline L3 430. Also, these convex skylines may be fine layers in which data is more densely distributed than the sparse layers of the skyline.

The partial dependency establishing step S350 may be a step of deriving tuples of other convex skylines that are dependent on each tuple of each convex skyline obtained in the convex skyline obtaining step S340 and establishing a relationship. In this case, the other convex skyline may be a convex skyline obtained from the same skyline as each convex skyline, and may be a convex skyline closest to each convex skyline.

Referring to FIG. 4, tuples of each convex skyline L11 411, L12 412, L21 421, L22 422, and L31 431 are connected by a dotted arrow in the same skyline. That is, as described above, each tuple of each convex skyline is connected by a dotted arrow and a tuple partially dependent on one of the tuples of the adjacent convex skyline in the same skyline. The method of obtaining the partial dependency is the same as described with the definition of partial dependency.

Accordingly, the dashed arrows af and bf mean that tuples a and b of L11 411 are partially dependent on tuple f of L12 412. That is, {a, b} may be a subset of f. Likewise {b, c} may be a subset of g.

Next, an embodiment of a method for processing a tuple ranking query according to an embodiment of the present invention will be described.

First, the concepts basically used to explain a method of processing a tuple ranking query according to an embodiment of the present invention are summarized first. As described above, the coarse hierarchical structure of the tuple formed according to an embodiment of the present invention is a hierarchical structure by sequentially obtaining a skyline, and each coarse hierarchical layer has the following properties.

1) Dependency in the sparse hierarchy

For two tuples T, T 'having a total dependency, the tuple T, which is created first and belonging to the coarse hierarchy above, is preferentially accessed, regardless of user preference, than the tuple T', which is created later in the coarse hierarchy below. In this case, the access means to access the data of the corresponding tuple, take the attribute value and apply it to the user preference function to calculate the value.

Each coarse layer is divided into fine layers according to the method of generating a dual hierarchy according to an embodiment of the present invention. That is, they are stratified by finding convex skylines in order in one sparse hierarchy. The fine layer thus generated has the following properties.

2) Partial Dependency in Fine Layers

For a particular tuple T belonging to a partial dependency set with a partial dependency and a tuple T 'depending upon, a tuple T of the partial subsets that are generated first and in the finer hierarchy above is a tuple that belongs to the later finer hierarchy that is created later. T 'is preferentially accessed regardless of user preference.

For example, referring to FIG. 4, coarse layers 410, 420, and 430 formed of skylines and fine layers 411, 412, 421, 422, and 431 formed of convex skylines for each coarse layer appear. In this dual hierarchical structure, in the order in which sparse hierarchies are generated, L1 410 is approached first to form a full dependency relationship with tuples of adjacent next coarse hierarchies L2 420.

In addition, within the same coarse layer (e.g., L1 410), L11 411 is approached first in the order in which the fine layers are created, and then partially dependent on adjacent next fine layers, tuples in L12 412. Configure

3) Forall-dominance-free

For any tuple T, if there are no tuples with full dependencies in the adjacent coarse hierarchies, or if all tuples in the adjacent coarse hierarchies with full dependencies are accessed, the tuple is in full dependent release. do.

For example, in the tuples a, b, and c of L11 411 of FIG. 4, since there is no overall dependency relation from the coarse hierarchy of the upper level, all of the tuples a, b, and c are all dependent release states. However, in the case of tuple d of L21 421, d is entirely dependent from tuple a of L11 411. Thus, if a is approached first, d can be in full dependent release.

3) Exists-dominance free

For any tuple T, if there is no tuple with a partial dependency on the adjacent finer hierarchy, or if one of the tuples in the adjacent finer hierarchy with partial dependency is accessed, the tuple is partially dependent freed. It is said to be. In this case, the access means to access the data of the corresponding tuple, take the attribute value and apply it to the user preference function to calculate the value.

For example, in the tuples a, b, and c of the L11 411 of FIG. 4, since there is no partial dependency relationship from the fine layer of the upper level, all of them are partially dependent released. Further, d, e, f of L21 (421) and h, k of L31 (431) also have no partial dependency relationship from the higher level hierarchy, and thus are all partially dependent released. However, for the tuple f of L12 412, f is partially dependent from tuples a and b of L11 411. Therefore, f cannot be partially released until both a and b are accessed.

4) Unnecessary access in the dual layer

When accessing a tuple to process a ranking query, you do not need access unless the current tuple's state is not full dependent release or partial dependent release. That is, according to the method of processing the ranking query according to an embodiment of the present invention, more efficient query processing can be performed by eliminating unnecessary access.

On how to process ranking queries Example .

5 is a flowchart illustrating a process of processing a ranking query according to an embodiment of the present invention.

Referring to FIG. 5, a process of processing a ranking query according to an embodiment of the present invention may include obtaining a user preference step (S510), inputting a tuple set step (S520), and adding a tuple of the highest layer to a priority queue (S530). It may be configured to include a tuple adding step (S540) to the result array, and a tuple adding step (S550) in which a dependency relationship is set in the priority queue.

Referring to FIG. 5, in a tuple adding step (S540) and a tuple adding step (S550) in which a dependency relationship is set in a priority queue, the number of tuples added to the result array is the number of tuples intended according to the ranking query. Can be repeated until K is reached.

Referring to FIG. 5, each step of a process of processing a ranking query according to an embodiment of the present invention may be described as follows.

The user preference obtaining step S510 may be a step of receiving a condition for data to be searched by a user. At this time, a search range according to the attribute may be input.

The tuple set input step S520 may be a step of inputting a tuple set corresponding to data to be searched by a user. In this case, the input tuple set may be classified into at least one group according to the distribution of attribute values, and a dependency relationship between tuples may be set.

That is, tuples included in the specific tuple set are classified into at least one group obtained by skylines to form a coarse hierarchy, and tuples included in each skyline in the coarse hierarchy are at least one subgroup obtained by convex skylines. It can be classified into to form a fine layer.

Adding a tuple of the highest layer to the priority queue (S530) may be a step of accessing a group of the highest layer among the skyline groups of the tuple set received in the previous step (S520) and putting tuples of the group into the priority queue. have. In this case, the highest layer may mean a finest layer first generated in the sparse layer generated first.

Adding a tuple to a result array (S540) may be a step of selecting a tuple having the highest user preference from the priority queue, adding the tuple to the result array, and removing the tuple added to the result array from the priority queue.

The tuple adding step in which a dependency relationship is set in a priority queue (S550) accesses tuples of which tuples are recently set in a result array among tuples in the specific tuple set and whose total dependency or partial relationship is set, and accesses the tuples. May be put into the priority queue.

In this case, the tuples put in the priority queue may be full dependent and partially dependent released tuples among the accessed tuples.

In addition, the steps S540 and S550 may be repeated, but may be repeated until the number of tuples in the result array reaches K. That is, if the number of tuples to be searched is three, the number of tuples added to the result array may be repeated until the number of tuples is three.

Hereinafter, a process of processing a ranking query according to an embodiment of the present invention will be described with reference to the drawings.

6 is a data structure diagram showing data generated in each process of processing a ranking query according to an embodiment of the present invention.

Referring to FIG. 6, tuples in the priority queue 630 and the result array 640 are shown for each operation 620 of processing a ranking query. Assume that you extract the three tuples with the best attribute values based on user preference.

First, as a process of initializing the priority queue and the result array to process the rank and the query, it can be seen that the priority queue and the result array are empty (operation 1).

Next, all tuples {a, b, c} of L11 410 of FIG. 4 are accessed and the corresponding tuples are added to priority queue 630. As a result, it can be seen that priority queue 630 has {a, b, c} (operation 2).

Next, among the tuples {a, b, c} of the priority queue, {a} having the best user preference value is extracted from the priority queue 630 and stored in the result array 640 (operation 3).

Next, through operation 3, tuples connected to a taken out of the priority queue 630 and which are totally dependent and partially dependent, are searched. At this time, as shown in FIG. 4, tuples connected to a whole dependently or partially dependent are {f, d, e, i}. However, f, d, and e are all partially dependent releases, whereas i is not partially dependent release. This is because {e, j}, which is a subset of i, has not been accessed yet. Accordingly, {d, e, f}, except for tuple i, which is not partially dependent released, enters the priority queue 630 (operation 4).

Next, if the highest user preference value in the priority queue is b, then b is taken out and placed in the result array 640 (operation 5).

Then, as in operations 4 and 5, tuples {g, j} which are full dependent release and partial dependency release are found by accessing the tuples connected to b of FIG. 4 and put into the priority queue 630 (operation 6).

Next, f is found again in the priority queue with the highest user affinity value and put into the result array 640 (operation 7). At this time, the number of tuples in the result array 640 is three, which is equal to the number of tuples originally intended to be extracted, and thus the priority query algorithm is terminated.

Tuple Clustered  How about Example .

7 is a flowchart illustrating a tuple clustering process according to an embodiment of the present invention.

Referring to FIG. 7, a tuple clustering process according to an embodiment of the present invention includes a tuple set input step (S710), a clustering step (S720), a virtual tuple generation step (S730), and a virtual tuple relationship setting step (S740). It can be configured to include.

In addition, referring to Figure 7, each step of the tuple clustering process according to an embodiment of the present invention can be described as follows.

The tuple set input step S710 may be a step of receiving a set of tuples for clustering. In this case, the received tuple set may be a specific tuple set which is classified into at least one group according to the distribution of attribute values and a relationship is established between dependent tuples. For example, a set of tuples is a set of tuples of a dual layer structure forming a coarse layer composed of at least one skyline and forming a fine layer composed of at least one convex skyline obtained for each skyline in the coarse hierarchy. Can be.

The clustering step (S720) may be a step of deriving at least one cluster by deriving a top group according to attribute values among groups of the received tuple set and clustering the derived top group.

For example, in the above-described double hierarchical set of tuples, the top layer of the coarse layered skylines is identified, and the convex skylines, which are the highest layer among the fine layer convex skylines in the identified skyline, are identified and clustered. can do. In this case, a tuple of the most convex skyline may be divided into a plurality of clusters by using a K-Means clustering method.

The virtual tuple generation step S730 may be a step of generating a virtual tuple having the best attribute value for each generated cluster. That is, a tuple having the most excellent attribute value among the tuples included in each cluster may be derived, and a virtual tuple may be generated by giving an attribute value superior to the derived tuple.

Hereinafter, a process of grouping tuples according to an embodiment of the present invention will be described with reference to the accompanying drawings.

8 is a graph illustrating a process of applying a tuple clustering method according to an embodiment of the present invention to a convex skyline.

FIG. 9 is a tuple hierarchy diagram illustrating a result of applying a tuple clustering method according to an embodiment of the present invention to the tuple hierarchy structure of FIG. 4.

Referring to FIG. 8, the tuples of the highest fine layer 810 are divided into two clusters by tying {a, b} 811 and {b, c} 812, respectively. The points 801 and 802 for setting the virtual tuples for the clusters 811 and 812 are shown. Both points 801 and 802 are set to have better attribute values than any of the tuples a, b and c in clusters 811 and 812.

9, a new sparse layer L0 910 and a fine layer L01 911 are added to the double layered tuple of FIG. 4. Also, points 801 and 802 shown in FIG. 8 are added to the fine layer L01 911 as the virtual tuples tv1 901 and tv2 902 of FIG. 9.

That is, the entire dependency relationship may be set for the virtual tuples 801 and 802 described with reference to FIG. 8 to connect the relationship. As in the overall dependency relationship described above, the point 801 corresponding to the virtual tuple of the {a, b} 811 cluster of FIG. 8 represents tv1 901 of FIG. 9, which corresponds to {a, has a total dependency on b}. Also, the point 802 corresponding to the virtual tuple of the {b, c} 812 cluster of FIG. 8 represents tv2 902 of FIG. 9, which indicates the total dependency on {b, c} of L11 510. Have

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the present invention as defined by the following claims It can be understood that

Claims (18)

Obtaining a skyline for a specific tuple set, and obtaining at least one skyline by repeating until all tuples of the tuple set are obtained as skylines;
A second step of deriving a tuple of another skyline to which each tuple of each skyline obtained in the first step is dependent and establishing a relationship;
Obtaining a convex skyline for each skyline obtained in the first step, and repeating until all tuples of each skyline are obtained as convex skylines, thereby obtaining at least one convex skyline for each skyline Third step; And
And a fourth step of establishing a relationship by deriving a tuple of another convex skyline to which each tuple of each convex skyline obtained in the third step is dependent. The method of claim 1,
The first step is to obtain a skyline for the specific tuple set, and to obtain a skyline for the remaining tuples not included in the skyline, by repeating until all tuples of the specific tuple set are obtained as skylines. And obtaining a coarse layer composed of the at least one skyline. 3. The method according to claim 1 or 2,
The second step is to establish a relationship with the tuples of other skylines that each tuple of each skyline obtained in the first step is totally dependent, wherein the other skylines are the closest skylines to the respective skylines. How to create a tuple hierarchy. 3. The method according to claim 1 or 2,
In the third step, a convex skyline is obtained for each skyline obtained in the first step, and a convex skyline is also obtained for the remaining tuples not included in the convex skyline, and all tuples of the skyline are convex. And repeating until it is obtained as a skyline, thereby obtaining a fine layer composed of the at least one convex skyline for each of the skylines. The method of claim 1,
In the fourth step, a partial dependency relationship is established by deriving tuples of other convex skylines in which each tuple of each convex skyline obtained in the third step is partially dependent.
Wherein said other convex skyline is a convex skyline obtained from the same skyline as said convex skyline and is a convex skyline closest to each of said convex skylines. The method of claim 1,
A fifth step of deriving the highest convex skyline generated first among the convex skylines obtained in the third step and clustering the highest convex skylines to generate at least one cluster;
Generating a virtual tuple having the best attribute value for each of the generated clusters; And
And a seventh step of establishing a relationship by deriving a tuple of the highest convex skyline to which all the virtual tuples generated in the sixth step are totally dependent. The method according to claim 6,
The clustering of the highest convex skylines performed in the fifth step uses a K-Means clustering method. The method according to claim 6,
The virtual tuple is a tuple hierarchy structure generating method characterized in that the tuple having a superior attribute value than the selected tuple by selecting a tuple having the most excellent attribute value among the generated tuples. A method of processing a ranking query according to user preference for a specific tuple set classified into at least one group according to the distribution of attribute values and having a relationship between dependent tuples.
Accessing a topmost group of the groups and putting tuples of the group into a priority queue;
Selecting a tuple having the largest user preference from the priority queue, adding it to a result array, and removing it from the priority queue; And
A third step of accessing tuples to which the tuple selected in the second step among the tuples in the specific tuple set depends, and putting the accessed tuples into the priority queue,
And repeating the second and third steps, after the second step, until the number of tuples in the result array reaches a predetermined number. The method of claim 9,
Tuples included in the specific tuple set are classified into at least one group to form a coarse layer, and tuples included in each group in the coarse layer are classified into at least one subgroup to form a fine layer,
And the first step includes accessing the highest subgroup within the first highest formed group among the coarse hierarchical groups and putting tuples of the highest subgroup into a priority queue. The method of claim 10,
Wherein the coarse hierarchical groups are groups obtained by skylines, and the sub-groups of fine hierarchies are groups obtained by convex skylines. 11. The method according to claim 9 or 10,
In the third step,
Ranked query processing, characterized in that the tuple selected in the second step among the tuples in the specific tuple set accesses all dependent tuples, and puts all dependent freed tuples of the accessed tuples into the priority queue. Way. 11. The method according to claim 9 or 10,
In the third step,
Ranked query processing, characterized in that the tuple selected in the second step of the tuple in the particular tuple set accesses tuples that are partially dependent, and puts partially dependent liberated tuples of the accessed tuples into the priority queue. Way. The method of claim 10,
The specific tuple set clusters the convex skylines of the uppermost layer of the fine layer, and includes a virtual tuple having the best attribute value generated for each cluster, wherein each virtual tuple is the highest convex skyline. A ranking query processing method characterized in that a dependency is set which is dependent on a tuple of. A method of clustering a specific set of tuples classified into at least one group according to the distribution of attribute values and having a relationship between dependent tuples.
Deriving a first-most formed group among the groups, and clustering the top-most group to generate at least one cluster;
Generating a virtual tuple having the best attribute value for each of the generated clusters; And
And a third step of establishing a relationship by deriving tuples of the uppermost group that are dependent on each virtual tuple generated in the second step. 16. The method of claim 15,
The specific set of tuples forms a sparse hierarchy of at least one skyline and forms a fine hierarchy of at least one convex skyline obtained for each skyline in the sparse hierarchy,
Tuple clustering method, characterized in that the highest group derived in the first step is the highest convex skyline. 17. The method according to claim 15 or 16,
Tuple clustering method characterized in that the clustering of the highest group performed in the first step using the K-Means clustering method. 16. The method of claim 15,
The virtual tuple is a tuple clustering method, characterized in that the tuple having a superior attribute value than the selected tuple by selecting a tuple having the most excellent attribute value among the generated tuples.

Images