CN112434118A - Shadow index and creation method, system, query method and system - Google Patents

Abstract

The invention discloses a Shadow index and creation method, a Shadow index and creation system, an inquiry method and an inquiry system, and relates to the field of space keyword inquiry. The Shadow index provided by the invention comprises the following components: a first layer for storing static objects in base units corresponding to leaf nodes; and the second layer is used for inserting the mobile object into the node created in the second layer by taking the leaf node of the first layer as an entrance. The query method comprises the following steps: when inquiring the mobile object expected by the user, if the leaf node R of the first layer is indexed by Shadow_iAnd disjoint with refined query result region, then R is pruned at the first and/or second level of the Shadow index_iAnd the search space of its children; calculating the node of the second layer of the Shadow indexAnd (3) reducing unnecessary searching spaces of nodes and groups according to a normal distribution principle by the probability that the user expected object appears in the corresponding basic unit within a period of time. The invention can solve the problem of Top-k WSKM query of the mobile object.

Description

Shadow index and creation method, system, query method and system

Technical Field

The invention relates to the field of space keyword query, in particular to a method and a system for Shadow index and creation, and a query method and a system.

Background

With the widespread use of mobile devices and the popularity of LBS (Location Based Services), a large number of mobile objects with spatial text features are generated. Top-k (find the Top k names in a set, k being a positive integer) SK (Spatial key) query is widely studied in academia and industry as one of the most important query forms in LBS. Top-k SK inquiry takes space coordinates, a group of keywords and k values as inquiry requirements, and returns k objects which are most matched with the inquiry requirements.

In some cases, when a user initiates a Top-k SK query, some of the user's desired objects (missing objects) may not appear in the query result set. The user may want to be sure why these objects disappear, whether any other unknown related objects disappear, or even question the entire query result. Therefore, it is necessary to explain the reasons for these expected objects being missing and to provide a refined query that contains all of the missing objects and the original query result objects.

For example: in a certain day, the weather is hot, john is thirsty in his company and wants a glass of milk tea. He then started to query the top three buttertea shops near his company. After receiving the query results, he unexpectedly found that neither a good mobile milky tea booth nor a milky tea shop he frequented were in the result set. John wants to know why the object he wants to query is missing and how to obtain a refined query keyword so that all missing stores and other potentially better options appear in the refined query result set.

The John's question in the above example is called the "Why-not" (why not) question. In contrast to other approaches that answer the "Why-not" question, such as operation recognition, ontology, and database modification, the query optimization approach can retrieve all missing objects for the user by adjusting the original query requirements. Chen et al answers the question "Why-not" in the spatial keyword Top-k query by adjusting the weights between spatial correlation and text similarity. Later, Zhao et al, Wang et al, and ZHEN et al dealt with the "Why-not" question in geo-social spatial keyword queries, SPARQL queries, and group queries, respectively.

However, existing research has focused primarily on the "Why-not" problem in Top-k SK queries against static objects.

Top-k space keyword queries for moving objects return moving or static Top-k objects according to a ranking function that comprehensively considers the spatial distance and text similarity between the query and the moving object. When a user initiates a Top-k SK (also referred to as Top-k SKM) query of a mobile object using some unreasonable queries, some mobile objects desired by the user (referred to as missing objects) may not appear in the query result set, and the user may want why they do not appear, which is referred to as the "Why-not" problem in SK queries for mobile objects, also referred to as Top-k WSKM queries.

In the process of implementing the invention, the inventor finds that at least the following problems exist in the prior art:

there are four factors that make it more difficult to answer the "Why-not" (Top-k WSKM) question on a mobile object Top-k space keyword query:

first, different moving objects have different moving patterns, such as moving direction and moving speed, and it is a challenge how to reasonably define the moving patterns of the moving objects.

Secondly, the probability that a moving object appears in a certain area at a certain time is a continuous variable, and how to establish a probability density function to calculate the probability that a moving object appears in a certain area within a period of time is also a considerable problem.

Third, in the process of refining query processing, some original query result objects and missing objects may leave the query point, and how to prune the search space as much as possible while ensuring that these moving objects are not pruned is also a considerable problem.

Finally, a large number of objects move in the system, which means that there are often insertions and deletions of moving objects in the index before a refined query is executed, and frequent insertions and deletions waste time, but the user needs to obtain the query result as soon as possible, which is a contradiction.

The applicant has not found the research results of the "Why-not" problem in Top-k space keyword query (Top-k WSKM query) for mobile objects at present.

Disclosure of Invention

The present application aims to overcome the above-mentioned drawbacks of the background art, and provides a method, a system, a query method and a system for Shadow indexing and creating, which can solve the problem of "Why-not" in Top-k space keyword query of a mobile object (i.e. Top-k WSKM query).

In a first aspect, a Shadow index is provided, including:

a first layer for: storing the static object in a base unit corresponding to a leaf node;

a second layer for: and taking the leaf node of the first layer as an entrance, and inserting the mobile object into the node created in the second layer.

On the basis of the technical scheme, the search space of the first layer of the Shadow index is divided into a plurality of basic units, and all static objects in the search space are indexed by using a quadtree.

On the basis of the above technical solution, the node of the second layer of the Shadow index stores information of the mobile object, including an object identifier, an object position, an object keyword, and a probability that the object appears in a basic unit corresponding to the node within a period of time.

On the basis of the above technical solution, the nodes of the second layer of the Shadow index are allocated to a plurality of groups according to the distance between the nodes and the groups, and each group stores the following information: group identification, group location, and group pointer to its neighboring group.

On the basis of the technical scheme, the distance between any two objects in the group does not exceed the length of the group, and the length of the group is calculated according to the position of the group.

In a second aspect, a method for creating a Shadow index is provided, including the following steps:

determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

In a third aspect, a system for creating a Shadow index is provided, including:

an entrance determination unit for: determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

a node creation unit for: and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

In a fourth aspect, a Top-k WSKM query method based on the above Shadow index is provided, which includes the following steps:

when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children.

On the basis of the technical scheme, the method further comprises the following steps:

and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

In a fifth aspect, a Top-k WSKM query system based on the above Shadow index is provided, which includes:

a first clipping unit to: when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children;

a second clipping unit to: and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

Compared with the prior art, the method has the advantages that:

1. the application proposes the Why-not problem for Top-k space keyword queries on mobile objects, also known as Top-k WSKM queries. To the best of the applicant's knowledge, this Top-k WSKM query problem was first addressed by the present application.

2. The application provides a new Shadow index to organize text information, spatial information and moving mode information of an object. By analyzing the movement pattern of the moving object and calculating the probability that the moving object appears in a certain area within a certain time, the Shadow index can help the user capture the refined query with the minimum cost at the fastest speed.

Drawings

FIG. 1 shows a query q and 18 objects o in an embodiment of the invention₁To o₁₈Schematic diagram distributed in search space.

Fig. 2 is a schematic diagram of a possible trajectory of a moving object moving from a point a to a point B in the embodiment of the present invention.

FIG. 3 is a graph of a calculated arc in an embodiment of the invention

Schematic drawing of auxiliary lines in the area of the moving area enclosed by the straight line AB.

FIG. 4 is an arc in an embodiment of the invention

And the relation between the half of the corresponding circumferential angle and the sine value of the circumferential angle is shown schematically.

Fig. 5 is a schematic diagram of a base unit to which a moving object may move and a position curve to which a moving object may move at different times in the embodiment of the present invention.

Fig. 6 is a diagram illustrating a result of spatial partitioning fig. 1 by using a Shadow index in the embodiment of the present invention.

Fig. 7 is a schematic structural diagram of a Shadow index in the embodiment of the present invention.

Detailed Description

Note that: the "Why-not Top-k space keyword query on Mobile object" and the "Why-not' question in Top-k space keyword query of Mobile object" in the present application mean the same meaning, abbreviated as: "Top-k WSKM query"; and "Top-k space keyword query of moving object", abbreviated as: the 'Top-k SKM query' is simply differed by one 'W', and the 'W' represents 'Why-not'.

Aiming at the problem of 'Why-not' (Top-k WSKM query) in the Top-k SKM query of a mobile object, the embodiment of the application provides a two-layer index named Shadow, which comprises the following steps:

a first layer for: storing the static object in a base unit corresponding to a leaf node;

a second layer for: and taking the leaf node of the first layer as an entrance, and inserting the mobile object into the node created in the second layer.

The embodiment of the application further provides a method for creating the Shadow index, which comprises the following steps:

determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

An embodiment of the present application further provides a system for creating a Shadow index, including:

an entrance determination unit for: determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

a node creation unit for: and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

As a preferred embodiment, the search space of the first layer of the Shadow index is divided into several elementary units, and all static objects in the search space are indexed using a quadtree.

As a preferred embodiment, the node of the second layer of the Shadow index stores information of the mobile object, including an object identifier, an object position, an object keyword, and a probability that the object appears in a basic unit corresponding to the node within a period of time.

As a preferred embodiment, the nodes of the second layer of the Shadow index are assigned to a plurality of groups according to their distance from the group, each group storing the following information: group identification, group location, and group pointer to its neighboring group. The distance between any two objects in the group does not exceed the length of the group, which is calculated from the group position.

The embodiment of the application further provides a Top-k WSKM query method based on the Shadow index, which comprises the following steps:

when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children;

and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

The embodiment of the present application further provides a Top-k WSKM query system based on the above Shadow index, including:

a first clipping unit to: when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children;

a second clipping unit to: and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

The present application first defines a movement pattern of a moving object. Given a probability density of a moving object appearing at a point in a region at a given time, the probability of the moving object appearing in the region at a point in time and for a period of time, respectively, can be calculated. The application constructs Why-not problem (Top-k WSKM query) of Top-k space keyword query on a mobile object, and provides a cost model to measure the modification degree of a refined query relative to an original query, so as to obtain the refined query with the minimum modification cost.

In order to effectively process Top-k WSKM query, the application provides a query processing method based on a three-phase framework.

The first stage is to generate some likely refinement queries with different query requirements and filter the less likely refinement queries before executing any of the likely refinement queries.

The second stage is based on spatial reduction techniques to minimize the irrelevant search space in the first layer of Shadow, and probability reduction techniques to capture the desired moving object as quickly as possible in the second layer of Shadow.

The third stage is to determine which promising refined queries are to be returned to the user.

The relevant background, problems and definitions are first explained below.

The present application gives some definitions and formally defines the Why-not problem for Top-k space keyword query on a mobile object, abbreviated as: top-k WSKM query.

Table 1 summarizes the symbols commonly used for Top-k WSKM queries and their meanings.

TABLE 1 Top-k WSKM query commonly used symbol table

The Top-k space keyword query (Top-k SKM query) on a moving object is explained below.

In existing related work, a spatial text object (simply referred to as an object) is generally defined as o ═ o.loc, o.doc, where o.loc is a spatial point and o.doc is a set of keywords. In reality, however, the objects are not always stationary and they may move continuously. Moving object o^mUsually have certain movement characteristics, such as: maximum velocity v_max(o^m) Minimum velocity v_min(o^m) And the actual speed v at time t^t(o^m). These properties are called objects o^mMobility capability MA at time t^t(o^m) Is defined as (v)_max(o^m),v_min(o^m),v^t(o^m))。

Please note that: in the real world, moving objects typically have their destinations, so the present application assumes that each moving object moves toward its destination without changing direction or returning.

Thus, given a spatial point o.loc, a set of keywords o.doc and mobility MA^t(O)), the object O ∈ O may be expressed as O ═ o.loc, o.doc, MA^t(o))。

Note that: when MA is used^t(o_i) Is empty, object o_iIs static.

Given a query point q.loc, a set of keywords q.doc, and two values α and k, then the Top-k spatial keyword query for moving objects (Top-k SKM query) q ═ q.loc, q.doc, α, k >, k, the k best objects are retrieved from the set of objects O according to a ranking function that takes into account the spatial distance and text similarity between query q and object O, where α is a smoothing parameter that satisfies 0 ≦ α ≦ 1.

The present application uses a widely used ranking function to measure the similarity between query q and object o, as follows:

Rank(q,o)＝α*(1-SD(o,q))+(1-α)*ST(o,q) (1)

where SD (o, q) is the normalized spatial distance between query q and object o, and ST (o, q) is the normalized text similarity between query q and object o.

Wherein D is_E(q, o) is the Euclidean distance between query q and object o, MaxD_ERepresenting the maximum distance between any two objects in the set O of objects.

The description is given by way of example. Referring to FIG. 1, a query q and 18 objects o are distributed in a search space₁To o₁₈Referring to fig. 1, normalized spatial distance and text similarity information between query q and each object are shown, where α is a smooth parameter satisfying that α is greater than or equal to 0 and less than or equal to 1, and when a user starts Top-3 SKM query, and α is 0.5 as input, object o is returned₂、o₅And o₁₃As a result of the query, because these objects have a higher ranking score than other objects.

The Why-not problem on the mobile object Top-k space keyword query, referred to as Top-k WSKM query, is explained below.

When a user initiates a Top-k query (Top-k SKM query) of a spatial keyword of a mobile object, q ═ c, doc, k, α), k objects are returned to form an original query result set^OAnd R is shown in the specification. If certain query parameters are incorrectly set in the query, one or more user-desired objects, referred to as missing objects, may accidentally disappear. This application represents the missing object set as^mO。

In the existing "Why-not" (why not) problem study, all objects in O were considered static objects. The model of the application can directly handle the simple situation, the mobile capability MAt (o) is directly set to be null, and the corresponding refined query result set is called as a static refined query result set^sR。

User obtains original result set^OR＝{o₂,o₅,o₁₃}, set^mO＝{o₄,o₉}. If all the objects in FIG. 1 are static, then existing methods can be used to return a refined query q ═ to the user (loc, doc, 7)0.6) to obtain a static refined query result set^sR＝{o₂,o₅,o₁₃,o₄,o₃,o₈,o₉}。

However, some objects are often moving, and the existing method cannot directly handle the "Why-not" problem in Top-k SKM query. For this reason, the application focuses on the Why-not problem of the mobile object Top-k space keyword query, referred to as Top-k WSKM query for short.

Note that: in this application, only k and α are adjusted to capture the inclusion^oR and^mrefined query result set of O^rR, some of which are moving objects.

For ease of expression, the present application first gives the following definitions.

Definition 1.(Original Query Result Area,OA for short).

Given a Top-k SKM query q＝(loc,doc,k,a)with a query result set ^oR,there is an object o_i∈^oR,

d(q,o)≤d(q,o_i).Then the original query result area can be defined as a circular area with the query point q.loc as the center and d(q,o_i)as the radius.

Definition 1: the Original Query Result Area (OA) is abbreviated. Given a Top-k WSK query q ═ (loc, doc, k, α), the query result set is^oR, has an object o_i∈^oR, i are positive integers, for any

d(q,o)≤d(q,o_i). The original query result area OA is defined as d (q, o) centered around the query point q.loc_i) Is a circular area of radius.

Definition 2.(Actual Refined Query Result Area,AA for short)

Given a top-k SKM query q with a query result set ^oR and a missing object set ^mO,there are a moving object o^m _k∈^oR∪^mO and a static object o_j∈^oR∪^mO,

moving objects o^m∈^oR∪^mO–{o^m _k},we have MAX{v_min(o^m _k)*(t_t-t_s)+d(q,o^m _k),d(q,o_j)}≥v_max(o^m)*(t_t-t_s)+d(q,o^m),where t_s is the starting time of q and t_t is the starting time of the refined query of q.MAX{a,b}returns the maximum value between a and b.Then the actual refined query result area can be defined as a circular area with the query point q.loc as the center and MAX{v_min(o^m _k)*(t_t-t_s)+d(q,o^m _k),d(q,o_j)}as the radius.

Definition 2: the Actual Refined Query Result Area (Actual referred Query Result Area), is abbreviated as AA.

Given a query with a set of query results^oTop-k SKM query q and deletion object set of R^mO, then there is a moving object O^m _k∈^oR∪^mO and a static object O_j∈^oR∪^mO, j is a positive integer, for any moving object O^m∈^oR∪^mO–{o^m _kIs of MAX { v }_min(o^m _k)*(t_t-t_s)+d(q,o^m _k),d(q,o_j)}≥v_max(o^m)*(t_t-t_s)+d(q,o^m) Wherein, t_sFor Top-k SKM query q start time, t_tThe start time of the refined query of q, MAX (a, b) being the greater between return a and b, then the actual refined query result area AA is defined as MAX { v } v centered around the query point q.loc_min(o^m _k)*(t_t-t_s)+d(q,o^m _k),d(q,o_j) Is a circular area of radius.

Definition 3: the independent Area (Irrelevant Area) is abbreviated as IA.

Given an AA, the irrelevant area is defined as the area outside the AA.

For ease of description, the original query and the refined query are represented below with different symbols: original query q ═ c, doc, k₀,α)，k₀Is the k value of the original query q, α is the α value of the original query q, the refined query q ═ (loc, doc, k ', α'), k 'is the k value of the refined query q', and α 'is the α value of the refined query q'.

To measure the refined query q ' ═ (loc, doc, k ', α ') relative to the original query q ═ (loc, doc, k)₀α) of the modification degree cost (q, q'), defining the following modification cost model:

where β ∈ (0,1) is a weight to indicate the user's preference for adjusting k or α, Δ α_maxIs the alpha maximum possible adjustment value and ak is the k maximum possible adjustment value.

Definition 4.(Why-Not questions on Top-k Spatial Keyword Query over Moving Objects,Top-k WSKM query for short)

Given an original Top-k SKM query q＝(loc,doc,ko,α),an original query result set ^oR and a missing object set ^mO,the Top-k WSKM query returns a refined query q’＝(loc,doc,k’,α’)with the minimum modified cost according to Eqn.(2)and its result set ^rO

^oR∪^mO.

Definition 4: and (4) carrying out Why-not problem of mobile object Top-k space keyword query, namely Top-k WSKM query.

An original Top-k SKM query q ═ (loc, doc, k) is given₀α), a set of original query results^oR and a set of missing objects^mO, Top-k WSKM query returns result set thereof^rO comprises^oR∪^mO, according to the modification cost model formula (2), the refined query q ' with the smallest modification cost is (loc, doc, k ', α ').

The moving capability of the moving object is explained below.

Each moving object

There is a destination and its mobility. Due to moving objects

Having maximum speed

And minimum speed

Thus moving the object

The distance of movement in the delta t time interval is

Within the range.

For ease of discussion, the present application further assumes that a moving object, no matter how fast it is moving, will move towards its destination and reach its destination at a particular point in time.

FIG. 2 shows a possible trajectory of a moving object moving from point A to point B, see FIG. 2 for a moving object

Moving from point A to point B if it continues at maximum speed

The movement is carried out in such a way that,need to be at time t_tWhen the point B is reached, its moving track is an arc AB and is shown as

Similarly, moving objects

At a speed of

When moving, the moving track is a straight line AB.

At a speed of

While moving, the object is moved

The movement at time t is confined to two arcs

Within the enclosed range, called

The moving area of (2).

Please note that: t e [ t ∈ ]_s,t_t]And an

Thus, the length of segment AB is

Arc of

Is approximately equal to

How to calculate the area of the above-described moving region is explained below.

FIG. 3 illustrates calculating an arc

Auxiliary lines drawn in the area of the moving area enclosed by the straight line AB, see FIG. 3, angle γ representing an arc

Half of the corresponding circumferential angle, then:

by simplification, we can get:

since γ ∈ (0, π/2), a value γ can be found₁E (1, pi/2) ensures

FIG. 4 shows an arc

Half of the corresponding circumferential angle is related to its sine value, see figure 4,

the area of the moving region of (2) is calculated as follows:

in this way, two arcs can be calculated

The formula expression of (1).

Due to space limitations, a detailed calculation process is not given here, but both are represented as

And

then

The area of the moving region of (2) can be calculated as follows in another way:

the probability density and probability are explained below.

Due to moving objects

Is not fixed, so

The position moved to at time t should be on the curve i. As is well known, spatial indexing is based on dividing the entire search space into elementary units. Thus, for any given spatial index, the curve l may appear in one or more base cells.

Let the curve l appear in n elementary units at time t, n being a positive integer, called C₁,C₂,…,C_nThe parts of the curve that appear in these units are respectively called l₁,l₂,…,l_nThen, there are:

and l_i>0。

If it is not

The probability density of a point appearing on the curve l at time t is

Then

Appears in_iThe probability of a segment is:

note that:

due to moving objects

The possible locations at different points in time are composed of different curves whose intersections with the basic cells can be obtained, so that t can be calculated₁To t₂Occurring in one basic unit during a time period

The probability of (c) is:

FIG. 5 illustrates a moving object

From t_sTime t_tMoving trajectory of time, see fig. 5, moving object

Respectively at t₁Time t and₂the time of day occurring in the basic unit C₂And C₁In t₃Time of day, moving object

The position curve of (A) appears in the basic cell C₁And C₂In each case is l₁And l₂。

If l₁And l₂Has the same probability density of falseIs all provided with

Then the object is moved

Appears in₁The probability of (c) is:

moving object

Appears in₂The probability of (c) is:

at different time points t e [ t ∈ ]₂,t₃]Moving an object

Present in the basic cell C₁Are different, moving objects

From t₂Time t₃The time of day occurring in the basic unit C₁The probability of (1) is:

in addition, the present application may use another method to calculate the value at [ t₁,t₂]During which the object is moved

The probability of occurrence in the elementary cell is briefly described below.

In one aspect, as described above, two arcs

Can be expressed by two functionsAre calculated and respectively recorded as

And

on the other hand, the basic unit C_iUsually carrying its spatial information, can be organized as a function g (C)_i) Then according to

And g (C)_i) Can calculate the moving object

Motion trail and basic unit C_iIn a time period t₁,t₂]Probability of intersection, i.e.

And g (C)_i) The intersection area of (a).

The probability distribution function and the normal distribution are explained below.

How to calculate the moving object is explained

After a certain period of probability of occurrence in a basic cell, the application gives the following probability distribution function:

this probability distribution function can be used in two ways:

1) given two points in time t_iAnd t_jWherein t is_s≤t_i<t_j≤t_tAnd t can be calculated_iTo t_jThe probability of a moving object occurring within the base unit;

2) given a starting time t_sIf a moving object is desired

The probability of occurrence in a basic unit is greater than a certain value, and a critical time point t can be calculated_kSo that at t_sMoving the object by t time period

Appears at C_iIs greater than this value, where t ∈ (t)_k,t_t)。

As described above, at time t, curve i is divided into n parts: l₁,l₂,…,l_nEach part having a moving object

The probability that appears on it. In the same way, the present application can calculate t by equation (3)_iTo t_jDuring the period, the object is moved

Present in the basic cell C₁,C₂…, where t is_s≤t_i<t_j≤t_t. This means that at t_iTo t_jMeanwhile, the probability that each moving object appears in a different basic unit is different.

If the missing object is a moving object, the refined query must return it and the different locations and probabilities that it occurred. Note that the probability that a moving object appears at a certain point in time is a probability density, which has no practical meaning. In the experiment, the application calculates the probability of the occurrence of a mobile object in a basic unit, and uses the probability value as the probability of the occurrence of the mobile object at any point in the basic unit.

However, a conflict arises when the probability of a moving object appearing in a base unit is very small and the base unit is very far from the query point over a period of time.

On the one hand, if one wants to ensure that each refined query captures the moving object that the user needs with 100% probability, one needs to access primitives and other unnecessary search space that are far from the query point, which takes time.

On the other hand, if the refined query does not have access to all the primitives where the mobile object appears with a certain probability, then the refined query has a probability that the user-desired object cannot be retrieved.

To ensure that the probability is as small as possible, and to resolve this conflict, the present application uses the 3 σ principle of positive too distribution. About normal distribution

Where μ is a mean value, σ is a standard deviation value, and x ═ μ is a symmetry axis of the image.

The 3 σ principle of the positive-Taiwan distribution is: the probability of the numerical distribution in (μ - σ, μ + σ) is 0.6826, the probability of the numerical distribution in (μ -2 σ, μ +2 σ) is 0.9544, and the probability of the numerical distribution in (μ -3 σ, μ +3 σ) is 0.9974. Therefore, it is considered that the value of P (. mu. -3. sigma.,. mu. + 3. sigma.) is 0.9974, and the values of Y are almost entirely concentrated in the range of (. mu. -3. sigma.,. mu. + 3. sigma.). The possibility of exceeding this range is only less than 0.3%.

The 3 σ principle means that a variable x that does not belong to (μ -3 σ, μ +3 σ) is a small probability event that will not occur under normal circumstances. Thus, if the object is moved

The sum of the probabilities occurring in some elementary units is greater than P (mu-3 sigma)<x ≦ μ +3 σ), i.e., 0.9974, then no access to other ones is required

The basic units which appear with a certain probability can be used for filtering out unnecessary search space so as to improve the query processing efficiency.

The following describes a Top-k WSKM query method based on Shadow index.

Knowing how to compute the probability that a mobile object will appear in a basic unit over a period of time, and knowing the fact that refining a query does not require finding the mobile object that the user needs with 100% probability, the following focus is to design a novel index to efficiently store and process the objects. If existing methods are used to hold objects (static or mobile) in order to answer a Top-k WSKM query, it is necessary to first delete or insert the mobile object from the index, then update the index accordingly, and finally perform a refined query on the modified index. This has two disadvantages:

1) time consuming, especially when large numbers of moving objects are far from the query point;

2) if a mobile object and its information are deleted, inserted, and the index structure is updated, some mobile objects may move to other locations during execution of the refined query, which may affect the accuracy of the refined query results.

To overcome these disadvantages, the present application proposes an index named Shadow, and the structure of the Shadow index is described in detail below.

The Shadow index contains two levels. In the first level, the entire search space is divided into several elementary units as described above, and all static objects in the search space are indexed using one quadtree. Leaf nodes in the first tier of Shadow store static objects in base units corresponding to the leaf nodes. Therefore, the first layer of the Shadow index is not affected no matter how the moving object moves.

In the second layer of the Shadow index, all moving objects are organized.

Inserting the first moving object into the second layer of the Shadow index, specifically comprising the steps of:

first, assuming that the mobile object is static, find the leaf node to be inserted into the first layer of the Shadow index;

then, in a second layer of the Shadow index, a node is created by using node id different from leaf nodes in the first layer and the same node position information;

finally, the mobile object is inserted into the node.

When other mobile objects are to be inserted into the second layer of the Shadow index, whether a node of the second layer of the Shadow index to be inserted exists needs to be determined, and if so, the objects only need to be inserted into the node; otherwise, a node needs to be constructed in the second layer of the Shadow index, and then the object is inserted into the node.

Please note that: if the number of objects in the node of the second layer of the Shadow index exceeds the maximum capacity of the node, the node needs to be divided into 4 sub-nodes, and the specific method is the same as that in the quadtree.

Each node in the second layer of the Shadow index stores information about the mobile objects it contains, including an object identification o^mId, object location o^mLoc, object keyword o^mDoc and object o^mAt [ t ]_i,t_j]Probability of occurrence in the basic unit corresponding to the node within the time period:

and inserting all the moving objects into the second layer of the Shadow index, and distributing all nodes of the second layer of the Shadow index into a plurality of groups according to the distance between the nodes and the groups. For each group, the following information is stored: group identifier G_aId, group position G_aLoc and group pointer G to its neighboring group_aP, can be according to group position G_aLoc calculates the length of the group and the distance between any two objects in the group does not exceed the group length. Note that: each group has the same upper group length limit. When a group reaches the upper limit of the length, it is not assigned to the group no matter how close a node in the second level of the Shadow index is to the group.

Referring to FIG. 6, FIG. 6 shows the result of partitioning all objects in FIG. 1 using the Shadow index, where there are 18 objects o in FIG. 1₁-o₁₈Wherein o is₂、o₃、o₉、o₁₀、o₁₈Are moving objects and the rest are static objects.

Referring to fig. 7, fig. 7 shows the structure of the Shadow index constructed for all the objects in fig. 1, each static object is allocated to its basic unit, and each mobile object can also find the corresponding basic unit. Note that in this example, it is assumed that each basic unit can hold information of four objects at most.

Referring to fig. 7, a first layer of the Shadow index stores information of static objects, a second layer of the Shadow index stores information of moving objects, and each leaf node of the first and second layers corresponds to one of the basic units of fig. 6.

Due to moving object o₉、o₁₀And static object o₈、o₁₁、o₁₂In the same basic unit, it is necessary to first find the storage static object o at the first layer of the Shadow index₈、o₁₁And o₁₂Then creates a node in the second layer of the Shadow index to store the mobile object o₉And o₁₀. From storage of static objects o₈、o₁₁And o₁₂The node constructs a pointed storage moving object o₉And o₁₀As one of the entries into the second level of the Shadow index. Due to moving object o₉And o₁₀Storage node A of (1) is more than the moving object o₂And o₃Is closer to the mobile object o₁₈Thus a and C are assigned to the second group and B is assigned to the first group.

The specific step of creating the Shadow index may refer to algorithm 1.

The inputs to algorithm 1 are a set of objects O containing static objects and moving objects, a queue Q for storing all moving objects, and an upper bound Gl for limiting the group length of the group size_maxAnd outputting the Shadow index. The first and second layers of Shadow are created by commands lines 4 and 5-12, respectively.

Algorithm 1：Creating Shadow Algorithm

1 Input：an object set O,a queue Q storing the moving objects,the upper limit Gl_max of group length；

2.Output:Shadow；

3.begin

4Build a quadtree for static objects in O to form the level 1 of Shadow；

5while Q is not empty do

6.o_i ^m＝Out_Queue(Q)；

7Find the left node R_i whose corresponding basic cell includes o_i ^massuming that o_i ^m is a static object；

8.if R_i has a pointer linked to the node R_i’in level 2 then

9.Insert o_i ^m into R_i’；

10.else

11.Create a node R_i’linked by R_i in level 2,and Insert o_i ^m into R_i’；

12.Group all nodes in level 2 to ensure that the length of each group does not exceed Gl_max；

13.Return Shadow；

Algorithm 1: shadow index creation algorithm:

1.Input:an object set O,a queue Q storing the moving objects,the upper limit Gl_max of group length；

2.Output:Shadow；

3.begin

4. establishing a quadtree for the static object in the O to form the level 1 of Shadow;

while (Q is not empty) do

6.{o_i ^m＝Out_Queue(Q)；

7. Suppose o_i ^mFor static objects, find includes o_i ^mCorresponding leaf node R of the basic unit_i；

8.if(R_iWith one pointing to a node R of the second layer_iThe pointer of `), then

{ will o_i ^mInsertion of R_i’；}

10.else

{ creation of one R at the second layer_iDirected node R_i', and will o^m _iInsertion of R_i’；}}

12. Grouping all nodes in the second layer to ensure that the length of each group does not exceed Gl_max；

13. The Shadow index is returned.

The following describes a space reduction technique.

The present application introduces several arguments to effectively cut unnecessary search space for each refined query q' to be examined. First, the previously defined AA (actually refined query result area) is used to prune unwanted nodes in the first tier of the Shadow index.

Introduction 1: node R in the first tier for a given Shadow index_iAnd refining AA of query q', if R_iAnd AA do not intersect, then R_iAnd its children will be pruned.

And (3) proving that: let R be_iContaining the result object o_iThen d (q', o)_i) No greater than a radius of AA. Due to R_iNot intersecting with AA, R_iAll of the objects in (a) are not within the range of AA, o_iNor is it. Thus, d (q', o)_i) A radius greater than AA, which contradicts the assumption. Thus, R_iAnd its sub-nodes can be safely cut down.

Please note that: this pruning technique may also be used in the second layer of the Shadow index to filter undesired nodes and groups. Second, in the second layer of the Shadow index, the 3 σ principle of normal distribution can be utilized to filter unnecessary nodes and groups.

2, leading: set of nodes in the second layer given Shadow index { R }₁,R₂，…R_iH, a user-desired moving object o_i ^mIf the sum of probability values of some nodes in the node set is larger than P (mu-3 sigma) of normal distribution<x is less than or equal to mu +3 sigma), then in the processing o_i ^mIn the process of (A) does not needOther nodes and groups are accessed.

And (3) proving that: proof of this lemma can be derived directly from the 3 σ principle of normal distribution and is therefore omitted.

How to answer the "Why-not" question in the Top-k SKM query is explained below.

Some user-desired objects may be missing from the query result set after the original query is executed. The main goal of the present application is to obtain a refined query with minimal modification cost, whose result set contains all original query result objects and all missing objects required by the user.

Algorithm 2 in this application presents the processing steps for answering why-not questions (i.e., Top-k WSKM queries) in Top-k SKM queries using the Shadow index. Since some existing methods have studied how to adjust α and k to answer the "why-not" problem in the spatial keyword Top-k query in the case where all objects are static, the present application mainly discusses a different part from the existing static algorithms.

The inputs to algorithm 2 are a Shadow index, an initial query q ═ loc, doc, k, α, a set of promising refined queries, an actual refined query result area AA, an initial query result set^OR, a set of missing objects^mO, time to execute the current refined query t, starting time t_sAnd an end time t_tAnd outputting an optimal refined query q' and a result set thereof.

First refine a query result set^rAnd R and the node set RS are set to be null, and respectively store the object meeting the requirement of the refined query and the leaf node of the first layer of the Shadow index (line 4) as the entrance of the second layer of the Shadow index. Next, a refined query is taken from the set of promising refined queries and its cost is calculated. If the cost of refining a query is lower than any other refined query that was previously executed, the refined query will be executed. Otherwise, execution of the refined query is terminated and the next likely query execution is selected (line 5).

According to lemma 1, the irrelevant tree branches and nodes corresponding to the search space that does not intersect AA in the first level of the Shadow index may be filtered (row 6), and then a set of nodes RS of promising static objects and leaf nodes containing these objects may be found (row 7). All promising static objects are now captured and the desired moving object is then derived from the remaining steps of algorithm 2.

Leaf node R in RS if RS is not empty_iIs fetched as an entry (using the pointer stored on this leaf node) to access the nodes and groups in the second level of the sharow index. Then at R_iAnd finding the mobile object meeting the requirement of the refining query in the linked nodes. If the probability value of the objects appearing in the search space corresponding to the node is larger than the P (mu-3 sigma) of the normal distribution<x ≦ μ +3 σ), then these moving objects are added to^rAnd R is shown in the specification. If one of the objects is not satisfied, the other nodes in the same group that the node is in will be accessed first. If necessary, nodes in other groups near the group are accessed until the probability requirement for the object is met (rows 8-10).

When RS is empty, calculate^rRanking scores of all objects in R, and keeping k best objects. Finally, a refined query is returned and^rR。

and 2, algorithm: top-k WSKM query algorithm based on Shadow index:

Algorithm 2:Shadow Based Algorithm

1.Input:Shadow,q,a promising refined query set,AA,^oR,^mO,t,t_s,t_t；

2.Output:Best refined query q’＝(loc,doc,k’,α’),^rR；

3.begin

4.

5.Execute a refined query in the promising refined query set,whose cost is less than any other refined queries previously executed；

6.Prune the irrelevant nodes in level 1 of Shadow corresponding to the search space that has no intersection with AA；

7.Find promising static objects and the node set RS of the leaf nodes containing these objects；

8.while RS is not empty do

9.Take out a leaf node R_i from RS and access the level 2 of Shadow；

10.Find promising moving objects that are not in ^rR and satisfy the refined query requirements,and insert them into RS according to the requirement of 3σprinciple；

11.Computer the ranking scores of all objects in ^rR and retain top-k’objects；

12.Return q'＝(loc,doc,k',a'),^rR；

the Top-k WSKM query algorithm 2 based on the Shadow index comprises the following specific contents:

input, Shadow, q, a set of desired refinement queries, AA,^oR,^mO,t,t_s,t_t；

output the best refinement query q ' ═ c, doc, k ', α '),^rR；

3.begin

4.set ^r

5. executing one of the set of desired refined queries at a cost less than any other refined queries previously executed;

6. deleting the irrelevant node of the first Shadow layer corresponding to the search space without intersection with AA;

7. finding a set of nodes RS of promising static objects and leaf nodes containing these objects;

while (RS non-null) do

{ removal of leaf node R from RS_iAnd accessing a second layer of the Shadow index;

10. find out of position^rPromising moving objects in R that satisfy the refining query requirements and insert them into R according to the requirements of the normal distribution 3 sigma principleS；}

11. Computing^rRanking all the objects in the R, and reserving Top-k' objects;

12. returning q ' (loc, doc, k ', α '),^rR。

it will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (10)

1. A Shadow index, comprising:

a first layer for: storing the static object in a base unit corresponding to a leaf node;

a second layer for: and taking the leaf node of the first layer as an entrance, and inserting the mobile object into the node created in the second layer.

2. The Shadow index of claim 1, wherein: the search space of the first layer of the Shadow index is divided into a number of elementary units, and all static objects in the search space are indexed using a quadtree.

3. The Shadow index of claim 1, wherein: the node of the second layer of the Shadow index stores information of the mobile object, including an object identifier, an object position, an object keyword and a probability that the object appears in a basic unit corresponding to the node within a period of time.

4. The Shadow index of claim 1, wherein: the nodes of the second layer of the Shadow index are assigned to a plurality of groups according to the distance between the nodes and the groups, and each group stores the following information: group identification, group location, and group pointer to its neighboring group.

5. The Shadow index of claim 4, wherein: the distance between any two objects in the group does not exceed the length of the group, which is calculated from the group position.

6. A method for creating a Shadow index is characterized by comprising the following steps:

determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

7. A system for creating a Shadow index, comprising:

an entrance determination unit for: determining a leaf node used for storing a static object in a first layer of a Shadow index as an insertion entry, and inserting the static object into the leaf node;

a node creation unit for: and creating a node for storing the mobile object in the second layer of the Shadow index according to the node id different from the leaf node in the first layer and the same node position information, and inserting the mobile object into the node.

8. The Shadow-indexed Top-k WSKM query method based on claim 1, characterized by comprising the following steps:

when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children.

9. The method of claim 8, further comprising the steps of:

and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

10. A Shadow-indexed Top-k WSKM query system based on claim 1, comprising:

a first clipping unit to: when inquiring the mobile object expected by the user, if the leaf node R of the first layer of the Shadow index_iDisjoint from the refined query result region, R is pruned in the first and/or second layers of the Shadow index_iAnd the search space of its children;

a second clipping unit to: and calculating the probability of the node of the second layer of the Shadow index appearing in the user expected object in the corresponding basic unit within a period of time, and reducing the search space of unnecessary nodes and groups according to the normal distribution principle.

Images