Recursive Graph Deduction and Reachability Queries Yangjun Chen Dept. Applied Computer Science, University of Winnipeg 515 Portage Ave. Winnipeg, Manitoba, Canada R3B 2E9

Outline Motivation Graph deduction -Basic definitions -Critical nodes and critical subgraphs -Evaluation of reachability queries Recursive graph deduction (RGD) -Recursive deduction -Evaluation of reachability queries based on RGD Conclusion

Motivation Efficient method to evaluate graph reachability queries Given a directed acyclic graph (DAG) G, check whether a node v is reachable from another node u through a path in G. Application XML data processing, gene-regulatory networks or metabolic networks. It is well known that XML documents are often represented by tree structure. However, an XML document may contain IDREF/ID references that turn itself into a directed, but sparse graph: a tree structure plus a few reference links. For a metabolic network, the graph reachability models a relationship whether two genes interact with each other or whether two proteins participate in a common pathway. Many such graphs are sparse.

A simple method - store a transitive closure as a matrix Motivation c b a d e G:G: c b a d e G*: M = abcdeabcde abcdeabcde M* = abcdeabcde abcdeabcde O(n 2 ) space query time: O(1)

Question: Is it possible to reduce the size of M*, but still have a constant query time? Motivation

Graph deduction Basic definitions a b c k d r h e f g i j Let G be a sparse graph. we will first find a spanning tree T of G. The spanning tree of G is represented by the solid arrows, which covers all nodes of G.

tree edges (E tree ): edges appearing in T. cross edges (E cross ): any edge (u, v) such that u and v are not on the same path in T. forward edges (E forward ): any edge (u, v) not appearing in T, but there exists a path from u to v in T. back edges (E back ): any edge (u, v) not appearing in T, but there exists a path from v to u in T. i a b c k d r h e f g j Graph deduction Edge classification

Graph deduction Tree encoding Let G be a DAG. we will first find a spanning tree T of G. Each node v in T will be assigned an interval [start, end), where start is v’s preorder number and end - 1 is the largest preorder number among all the nodes in T[v]. So another node u labeled [start’, end’) is a descendant of v (with respect to T) iff start’  [start, end). i [3, 4) j [11, 12) [9, 12) [5, 9) k d r [8, 9) h e f c b a [10, 11) [6, 9) [7, 8) [4, 5) [2, 4) [1, 5) [0, 12) g

Graph deduction Tree encoding Let v and u be two nodes in T, labeled [a, b) and [a’, b’), respectively. If a  [a’, b’), v is a descendant of u. In this case, we say, [a, b) is subsumed by [a’, b’). Also, we must have b  b’. Therefore, if v and u are not on the same path in T, we have either a’  b or a  b’. In the former case, we say, [a, b) is smaller than [a’, b’), denoted [a, b)  [a’, b’). In the latter case, [a’, b’) is smaller than [a, b). i [3, 4) j [11, 12) [9, 12) [5, 9) k d r [8, 9) h e f c b a [10, 11) [6, 9) [7, 8) [4, 5) [2, 4) [1, 5) [0, 12) g

Graph deduction Critical nodes and critical subgraph We denote by E’ the set of all cross edges. Denote by V’ the set of all the end points of the cross edges. That is, V’ = V start  V end, where V start contains all the start nodes while V end all the end nodes of the cross edges. V start = {d, f, g, h} V end = {c, k, e, d, g} i j k d r h e f c b a g

Definition 1 (anti-subsuming subset) A subset S  V start is called an anti-subsuming subset iff |S| > 1 and no two nodes in S are related by ancestor-descendant relationship with respect to T. {d, f} {d, g} {d, h} {f, g} {f, h} {g, h} {d, f, g} {d, f, h} {d, g, h} {f, g, h} {d, f, g, h} anti-subsumming subsets: Graph deduction Critical nodes and critical subgraph i j k d r h e f c b a g

Definition 2 (critical node) A node v in a spanning tree T of G is critical if v  V start or there exists an anti-subsuming subset S = {v 1, v 2,..., v k } for k  2 such that v is the lowest common ancestor of v 1, v 2,..., v k. We denote V c the set of all critical nodes. In the graph, node e is the lowest common ancestor of {f, g}, and node a is the lowest common ancestor of {d, f, g, h}. So e and a are critical nodes. In addition, each v  V start is a critical node. So all the critical nodes of G with respect to T are {d, f, g, h, e, a}. h i j k d r e f c b a g Graph deduction Critical nodes and critical subgraph V c = {d, f, g, h, e, a}.

Critical node recognition Algorithm critical-node-recognition(T) 1. Mark any node in T, which belongs to V start. 2. Let v be the first marked node encountered during the bottom-up searching of T. Create the first node for v in G c. 3. Let u be the currently encountered node in T. Let u’ be a node in T, for which a node in G c is created just before u is met. Do (4) or (5), depending on whether u is a marked node or not. 4. If u is a marked node, then do the following. (a)If u’ is not a child (descendant) of u, create a link from u to u’, called a left-sibling link and denoted as left-sibling(u) = u’. Graph deduction

Critical node recognition Algorithm critical-node-recognition(T) (continued) (b)If u’ is a child (descendant) of u, we will first create a link from u’ to u, called a parent link and denoted as parent(u’) = u. Then, we will go along a left-sibling chain starting from u’ until we meet a node u’’ which is not a child (descendant) of u. For each encountered node w except u’’, set parent(w)  u. Set left- sibling(u)  u’’. Remove left-sibling(w) for each child w of u. 5.If u is a non-marked node, then do the following. (c)If u’ is not a child (descendant) of u, no node will be created. (d)If u’ is a child (descendant) of u, we will go along a left-sibling chain starting from u’ until we meet a node u’’ which is not a child (descendant) of u. If the number of the nodes encountered during the chain navigation (not including u’’) is more than 1, we will create new node in G c and do the same operation as (4.b). Otherwise, no node is created. Graph deduction

Sample trace … u’’ u u’’ is not a child of u. u’ link to the left sibling … u’’ u u’ ddfdf g (c)(b)(a) e h g f d a (f) df g e df g e h (e) (d) Graph deduction i j k d r h e f c b a g

Tree deduction Let T be a spanning tree of G. Denote by T r a reduction of T obtained by removing all those nodes v  V c  V end. Deleting a node v entails connecting v’s parent to each of v’s children. So, removing a node in this way corresponds to the elimination of a tree edge. Example: T r obtained by removing the nodes b, r, i, and j one by one. (Note that none of them belongs to V c  V end. V c = {a, d, e, f, g, h} and V end = {c, d, e, g, k}.) Graph deduction a c k d h e f g Tr:Tr: i j k d r h e f c b a g

Critical subgraph Definition 4 (critical subgraph) Let G(V, E) be a DAG. Let T be a spanning tree of G. The critical subgraph G c of G with respect to T is graph with node set V(T r ) and edge set E(T r )  E cross. a c k d h e f g Gc:Gc: The reachability of any two nodes can be checked by using T or G c.

Graph deduction i [3, 4) j [11, 12) [9, 12) [5, 9) k d r [8, 9) h e f c b a [10, 11) [6, 9) [7, 8) [4, 5) [2, 4) [1, 5) [0, 12) g r f ?  r d ?  a c k d h e f g Gc:Gc:

Graph deduction Evaluation of reachablity queries Definition 5 (anchor nodes) Let G be a DAG and T a spanning tree of G. Let v be a node in T. Denote by C v all the critical nodes in T[v]. We associate two anchor nodes with v as below. i)A node u  C v is called an anchor node (of the first kind) of v if u is closest to v. u is denoted v*. ii)A node w is called an anchor node (of the second kind) of v if it is the lowest ancestor of v (in T), which has a cross incoming edge. w is denoted v**. Example. r* = e. It is because node e is critical and closest to node r in T[r]. But r** does not exist since it does not have an ancestor which has a cross incoming edge. e* = e** = e. That is, both the first and second kinds of anchor nodes of e are e itself.

Example. r* = e. It is because node e is critical and closest to node r in T[r]. But r** does not exist since it does not have an ancestor which has a cross incoming edge. e* = e** = e. That is, both the first and second kinds of anchor nodes of e are e itself. Graph deduction Evaluation of reachablity queries i j k d r h e f c b a g f** = e

Graph deduction Evaluation of reachablity queries Definition 6 (non-tree labels) Let v be a node in G. The non-tree label of v is a pair, where -x = v* if v* exists. If v* does not exists, let x be the special symbol “-”. -y = v** if v** exists. If v** does not exist, let y be “-”.

Graph deduction Example a b c k d r h e f g i j [5, 9) [4, 5) r* = e d** = d d is reachable from e through a path in G c. So d is reachable from r. r d ?  a c k d h e f g Gc:Gc:

a c k dhe f g a e f d k h g c acdefghkacdefghk (1, 1) (2, 3) (1, 4) (1, 2) (1, 3) (2, 2) (2, 1) (1, 5) Index(v) Graph deduction Evaluation of reachablity queries Reachability checking over G c : Decompose G c into chains:

Graph deduction Evaluation of reachablity queries Reachability checking over G: b c k 1st chain r e f 2nd chain d (2, 1) (1, 2)(3, 2)(4, -)(5, -) (2, 2) (1, 2)(3, 2)(4, -)(5, -) (2, 3) (1, 3)(3, -)(4, -)(5, -) (2, 4) (1, 3)(3, -)(4, -)(5, -) a i 4th chain (4, 1) (1, 1)(2, 1)(3, 1)(5, 1) (4, 2) (1, -)(2, -)(3, -)(5, -) h g 3rd chain (3, 1) (1, 2)(2, 2)(4, 2)(5, 1) (3, 2) (1, 2)(2, -)(4, -)(5, -) j 5th chain (5, 1) (1, -)(2, -)(3, -)(4, -) Index(v) (1, 1) (2, 4)(3, -)(4, -)(5, -) (1, 2) (2, -)(3, -)(4, -)(5, -) (1, 3) (2, -)(3, -)(4, -)(5, -) abcdefghijkrabcdefghijkr

From the above discussion, we can see that G c is much smaller than G. However, it can be observed that G c itself can be further re­duced, leading to a further reduction of space requirement. Recursive graph decomposition Recursive deduction Using the above method, we can find a series of graph reductions: G 0 = G, G 1,..., G k,(k  1) where G i is a critical subgraph of G i-1 (i = 1,..., k). In order to construct such critical subgraphs, a series of spanning trees have to be established: T 0, T 1,..., T k-1, where each T i is a spanning tree of G i (i = 0,..., k - 1), used to construct G i+1.

To check reachability efficiently, each node v in G will be asssociated with two sequences: an interval sequence and an anchor node sequence: 1)[ 0 (v),  0 (v)),..., [ j (v),  j (v)) (j  k - 1) where each [ i (v),  i (v)) is an interval generated by labeling Ti; 2)(x 0 (v), y 0 (v)),..., (x j (v), y j (v)), where each is a pointer to an anchor node of the first kind (a node appearing in G i+1 ) while each a pointer to an anchor node of the second kind (also, a node in G i+1 ). Recursive graph decomposition Recursive deduction

Recursive graph decomposition Recursive deduction G 0 : U [ 0 (u),  0 (u)) v [ 0 (v),  0 (v)) w [ 0 (w),  0 (w)) z [ 0 (z),  0 (z)) G 1 : U [ 1 (u),  1 (u)) v [ 1 (v),  1 (v)) w [ 1 (w),  1 (w)) z [ 1 (z),  1 (z)) G j : U [ j (u),  j (u)) v [ j (v),  j (v)) w [ j (w),  j (w)) z [ j (z),  j (z)) * ** * *

Recursive graph decomposition Recursive deduction Example g c k f a h e d i j k d r h e f c b a g G0:G0:G1:G1:G2:G2: ck ckck (1, 1) (1, 2) Index(v)

Recursive graph decomposition Recursive deduction abcdefghijkrabcdefghijkr [0, 12)[0, 8) [1, 5) [2, 4)[7, 8) [4, 5)[4, 6) [6, 9)[2, 8] [7, 8)[3, 6) [8, 9)[6, 8) [9, 12)[1, 8) [[10, 11) [11, 12) [3, 4)[5, 6) [5, 9) Example Interval sequence:Anchor node sequence:

Recursive graph decomposition Evaluation of reachability queries abcdefghijkrabcdefghijkr Anchor node sequence: a {1, *} c {1, **} {2, *} g {1, *} {1, **} e {1, *} {1, **} h {1, *} f d {1, **} b k {2, **} r {2, *} {1, **} {1, *} {2, *} g k ?k ?  [ 0 (g),  0 (g)) = [8, 9);  0 (k),  0 (k)) = [3, 4); [ 1 (g),  1 (g)) = [6, 8);  1 (k),  1 (k)) = [5, 6). In G 2, k is reachable from c, which shows that k is reachable from g.

Summary Transitive closure compression based on graph deduction - DAG decomposition: a spanning and a subgraph -Reachability checking: tree labels and reachability of anchor nodes in the subgraph Transitive closure compression based on recursive graph deduction -DAG decomposition: a series of spanning trees and a subgraph -Reachability checking: interval sequences and anchor node sequences

Summary Computational complexities - labeling time: O(ke + b k 1.5 n k ) -space overhead: O(kn + b k n k ) -query time: O(k) where n – number of the nodes of G, e - number of the nodes of G, n k – number of the nodes of G k, and b k – width of G k.

Thank you.