Multi-location visibility query processing using portion-based transactional modeling and pattern mining

Abstract

Visibility computation is critical in spatial databases for realizing various interesting and diverse applications such as defence-related surveillance, identifying interesting spots in tourist places and online warcraft games. Existing works address the problem of identifying individual locations for maximizing the visibility of a given target object. However, in case of many applications, a set of locations may be more effective than just individual locations towards maximizing the visibility of the given target object. In this paper, we introduce the Multi-Location Visibility (MLV) query. An MLV query determines the top-k query locations from which the visibility of a given target object can be maximized. We propose a portion-based transactional framework and coverage pattern mining based algorithm to process MLV queries. Our performance evaluation with real datasets demonstrates the effectiveness of the proposed scheme in terms of query processing time, pruning efficiency and target object visibility w.r.t. a recent existing scheme.

Appendices

Appendix: A coverage patterns

The concept of coverage has been used to the solve set cover problem in set theory (Chvatal 1979) and node cover problem in graph theory (Garey et al. 1974). By extending the notion of coverage, a model for extracting the knowledge of coverage patterns (CPs) (Srinivas et al. 2012) was proposed to extract coverage value of distinct sets of data items or patterns from a transactional database.

Now we shall explain the concept of coverage patterns. Given a transactional database D, each transaction is a subset of a set I of m items $\{i_{1},i_{2},i_{3},\ldots , i_{m}\}$. A pattern P is a subset of items in I i.e., $P\subseteq I$ where $P=\{i_{p},i_{q},\ldots ,i_{r}\}$ and $1\le p\le q\le r \le m$ and p, q, r are integers. $T^{i_{r}}$ is the set of transactions in which item $i_{r}$ is present, hence $|T^{i_{r}}|$ is the cardinality of $T^{i_{r}}$.

The fraction of transactions containing a given item $i_{r}$ is designated as Relative Frequency$RF(i_{r})$ of $i_{r}$. $RF(i_{r})=|T^{i_{r}}|/|D|$. A given item is considered as frequent if its relative frequency is greater than that of a threshold value, which we designate as minRF. Coverage Set CSet(P) of a pattern $P=\{i_{p},i_{q},\ldots ,i_{r}\}$ is the set of all the transactions that contain at least one item from the pattern P i.e., $CSet(P)=T^{i_{p}} \cup T^{i_{q}},\ldots ,\cup T^{i_{r}}$. Coverage Support CS(P) is the ratio of the size of CSet(P) to the size of D i.e., $CS(P)=|CSet(P)| / |\textit{D}|$.

In order to add a new item to the pattern P such that the coverage support increases significantly, the notion of overlap ratio is introduced. (This is possible in the case when the number of transactions, which are common to the new item and the pattern P, is low.) Overlap ratio$OR(\textit{P})$ of a pattern P is the ratio of the number of transactions that are common between CSet($P-i_{r}$) and $T^{i_{r}}$ to the number of transactions in $T^{i_{r}}$, i.e., $OR(P)=(|CSet(P-i_{r}) \cap (T^{i_{r}})|) /(|T^{i_{r}}|)$. A high CS value indicates more number of transactions and a low OR value means less repetitions among the transactions. A pattern is interesting if its CS is greater than or equal to the user-specified minimum CS threshold value (i.e., minCS) and its OR is less than or equal to user-specified maximum OR threshold value (i.e., maxOR). Given minRF, minCS and maxOR, pattern $P=\{{i_{p}},{i_{q}},\ldots ,i_{r}\}$ is regarded as a coverage pattern CP if $RF({i_{k}})\ge minRF \, \forall i_{k} \in P, CS({P}) \ge minCS$ and $OR(P) \le maxOR$.

Table 2 Transactional database

For example, consider D depicted in Table 2 to understand the notion of coverage patterns. In D, the set of items $I=\{a,b,c,d,e,f,g,h\}$ and the number of transactions i.e., $|\textit{D}|=10$ with $\hbox {TIDs}=1,2,3,\ldots ,10$. Let $minRF=0.3$, $minCS =0.6$, $maxOR=0.8$. Consider pattern $P=\{b,c,d\}$; $T^{b}=\{1,2,4,7\}$, $T^{c}=\{1,2,5,7,9,10\}$, $T^{d}=\{5,7,8\}$. $|T^{b}|=4$, $|T^{c}|=6$, $|T^{d}|=3$; $RF(b)=4/10=0.4$, $RF(c)=6/10=0.6$, $RF (d)=3/10=0.3$. $\hbox {CSet}(P)=\{1,2,4,5, 7,8,9,10\}$. $RF(b)\ge minRF$, $RF(c) \ge minRF$, $RF(d) \ge minRF$; $CS(P)=8/10= 0.8\ge minCS$; $OR(P)=(|(T^{b}\cup T^{c})\cap (T^{d})|)/ (|T^{d}|)$; $OR(P)=2/3=0.6 \le maxOR$. Hence, $\{b,c,d\}$ is a coverage pattern.

An apriori-like approach to extract complete sets of CPs from transactional database has been proposed in Srinivas et al. (2012). A pattern growth-like approach (Gowtham Srinivas 2015) has also been proposed to extract the complete set of CPs in an efficient manner.

Appendix: B Illustrative example of our approach

Figure 8 depicts an illustrative example of the proposed approach. Consider the target object ABCD having four sides $\{AB,BC,CD,DA\}$ with query locations ($q_{1}$ to $q_{10}$) and obstacles ($o_{1}$ to $o_{4}$). ABCD is divided into portions ($p_{1}$ to $p_{10}$).

Fig. 8

Illustrative example of our approach

We compute the VRs corresponding to each query location, as shown in Table 3. Observe that $p_{1}$ is visible from query location $q_{1}, p_{5}$ is visible from $\{q_{4},q_{6}\}$ and $p_{7}$ is visible from $\{q_{6},q_{7},q_{8}\}$. Now using the computed VRs, we generate portion transactions. Hence, the query location in portion transaction $p_{1}$ is $q_{1}$. Similarly, in the portion transaction $p_{7}$, the query locations are $\{q_{6},q_{7},q_{8}\}$, as depicted in Table 4.

Table 3 Visible regions (VRs) from query locations

Table 4 Portion transactions

Now let us consider a candidate set $c_{i}=\{q_{4},q_{6},q_{9}\}$. Let the values of minV and maxO be 0.7 and 0.4 respectively. $VR(q_{4},T)=\{p_{3},p_{4}, p_{5}\}$, $VR(q_{6},T)=\{p_{5}, p_{6},p_{7}\}$, $VR(q_{9},T)=\{p_{8},p_{9}\}$. $vis(c_{i},T)=|VR(q_{4},T)\cup VR(q_{6},T)\cup VR(q_{9},T)|/|S| =\{p_{3},p_{4}, p_{5}, p_{6}, p_{7},p_{8},p_{9}\} /10$ i.e., $vis(c_{i},T)=0.8$. $|VR(q_{4},T)|\ge |VR(q_{6},T)| \ge |VR(q_{9},T)|$. Thus, $overlap(c_{i},T)= |(VR(q_{4},T)\cup VR(q_{6},T))\cap VR(q_{9},T)| / |VR(q_{9},T)| =0/2$ i.e., $overlap(c_{i},T)=0$. Here, $continuity(c_{i},T)=true$ since portions $p_{3}$ to $p_{9}$ from the VRs of query locations in $c_{i}$ are continuous. Since $vis(c_{i},T)\ge minV, overlap(c_{i},T) \le maxO$ and $continuity (c_{i},T)=true, c_{i}$ qualifies as a candidate set.

Now let us consider $c_{i} =\{q_{6},q_{8}\}$. $VR(q_{6},T) =\{p_{5},p_{6},p_{7}\}$, $VR (q_{8},T) =\{p_{7}\}$. $vis(c_{i},T)=|VR(q_{6},T)\cup VR(q_{8},T)|/|S|=\{p_{5},p_{6},p_{7}\}$ /10 i.e., $vis(c_{i},T)=0.3.|VR(q_{6},T)| \ge |VR(q_{8},T)|$.

Thus, $overlap(c_{i},T)=|(VR(q_{6},T) \cap VR(q_{8},T)| / |VR(q_{8},T)|= 1/1$ i.e., $overlap(c_{i},T)=1$. Here, $continuity(c_{i},T)=true$ since portions $p_{5}$ to $p_{7}$ from the VRs of query locations in $c_{i}$ are continuous. Since $vis(c_{i},T)$ and $overlap(c_{i},T)$ do not satisfy the minV and maxO constraints, $c_{i}$ would not qualify as a candidate set.