Innovative Study to the Graph-based Data Mining: Application of the Data Mining

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.

Share

Tags

Transcript

ISSN: 2277-3754
ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 2, August 2012
47
Abstract
—
Graph-based data mining represents a collection of techniques for mining the relational aspects of data represented as a graph. Two major approaches to graph based data mining are frequent sub graph mining and graph-based relational learning. This article will focus on one particular approach embodied in the Subdue system, along with recent advances in graph-based supervised learning, graph-based hierarchical conceptual clustering, and graph-grammar induction. The need for mining structured data has increased rapidly. One of the best studied data structures in computer science and discrete mathematics are graphs. Graph based data mining has become quite popular in the last few years. This paper introduces the theoretical basis of graph based data mining and surveys the state of the art of graph-based data mining. Brief descriptions of some representative approaches are provided as well. Index Terms
—
Graph, Tree, Path, Structured Data, Data Mining.
I
.
INTRODUCTION
During the past decade, the field of data mining has emerged as a novel field of research, investigating interesting research issues and developing challenging real-life applications. The objective data formats in the beginning of the field were limited to relational tables and transactions where each in-stance is represented by one row in a table or one transaction represented as a set. However, the studies within the last several years began to extend the classes of considered data to semi-structured data such as HTML and XML texts symbolic sequences, ordered trees and relations represented by advanced logics. Graph mining has a strong relation with the afore mentioned Multi-relational data mining. However, the main objective of graph mining is to provide new principles and efficient algorithms to mine topological substructures embedded in graph data, while the main objective of multi-relational data mining is to provide principles to mine and/or learn the relational patterns, represented by the expressive logical languages. The former is more geometry oriented and the latter more logic and relation oriented in this paper, the theoretical basis of graph-based data mining is explained in the following section. Second the approaches to graph-based data mining are reviewed and some representative approaches are briefly described.
A. Theoretical Approaches of Graph Based Data Mining
There are five theoretical based approaches of graph-based data mining. They are sub graph categories, sub graph isomorphism, graph invariants, mining measures and solution methods. The sub graphs are categorized into various classes, and the approaches of graph-based data mining strongly depend on the targeted class. Sub graph isomorphism is the mathematical basis of substructure matching and/or counting in graph-based data mining. Graph invariants provide an important mathematical criterion to efficiently reduce the search space of the targeted graph structures in some approaches. Furthermore, the mining measures define the characteristics of the patterns to be mined similarly to conventional data mining. In this paper, the theoretical basis is explained for only undirected graphs without labels but with/without cyclic edges and parallel edges due to space limitations. But, an almost identical discussion applies to directed graphs and/or labeled graphs. Most of the search algorithms used in graph-based data mining come from artificial intelligence, but some extra search algorithms founded in mathematics are also used
.
B.Recent Developments Carried Out On Graph Based Data Mining
Researchers have proposed a variety of unsupervised-discovery approaches for structural data. One approach is to use a knowledge base of concepts to classify the structural data. Systems using this approach learn concepts from examples and then categorize observed data. Such systems represent examples as distinct objects and process individual objects one at a time. In contrast, Subdue stores the entire database (with embedded objects) as one graph and processes the graph as a whole. Scientific discovery systems that use domain knowledge have also been developed, but they target a single application domain. An example is Mechem, which relies on domain knowledge to discover chemistry hypotheses. In contrast, Subdue performs general-purpose, automated discovery with or without domain knowledge and hence can be applied to many structural domains. Logic-based systems have dominated relational concept learning, especially inductive logic programming (ILP) systems. However, first-order logic can also be represented as a graph and, in fact, is a subset of what graphs can represent. Therefore, learning systems using graphical representations potentially can learn richer concepts if they can handle the larger hypothesis space. FOIL, the ILP system discussed in this article, executes a
Innovative Study to the Graph-based Data Mining: Application of the Data Mining
Amit Kr. Mishra, Pradeep Gupta, Ashutosh Bhatt, Jainendra Singh Rana
ISSN: 2277-3754
ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 2, August 2012
48 top-down approach to learning relational concepts (theories) represented as an ordered sequence of function-free definite clauses. Given extensional background knowledge including relations and examples of the target concept relation, FOIL begins with the most general theory. Then it follows a set-covering approach, repeatedly adding a clause that covers some positive examples and few negative examples. Then, FOIL removes the positive examples covered by the clause and iterates the process on the reduced set of positive examples and all negative examples until the theory covers all the positive examples. To avoid over complex clauses,
FOIL ensures that a clause’s
description length does not exceed the description length of the examples the clause covers. In addition to the applications discussed here, as well as applications in numerous recursive and no recursive logical domains, FOIL has been applied to learning search-control rules and patterns in hypertext.
II. APPROACHES OF GRAPH MINING
The approaches to graph-based data mining are categorized into five groups. They are greedy search based approach, inductive logic programming (ILP) based approach, inductive database based approach, mathematical graph theory based approach and kernel function based approach.
A. Greedy Search Based Approach
Two pioneering works appeared in around 1994, both of which were in the framework of greedy search based graph mining. Interestingly both were srcinated to discover concepts from graph representations of some structure, e.g. a conceptual graph similar to semantic network and a physical system such as electric circuits.
One is called “
SUBDUE
”.
SUBDUE deals with conceptual graphs which belong to a class of connected graph. The vertex set V (G) is R [ C where R and C are the sets of labeled vertices representing relations and concepts respectively. The edge set E(G) is U which is a set of labeled edges. Though the srcinal SUBDUE targeted the discovery of repeatedly appearing connected subgraphs in this specie type of graph data, i.e., concept graph data, the principle can be applied to generic connected graphs. SUBDUE starts looking for a subgraph which can best com press an input graph G based on Minimum Description Length (MDL) principle. The found subgraph can be considered a concept. This algorithm is based on a computationally constrained beam search. It begins with a subgraph comprising only a single vertex in the input graph G, and grows it incrementally expanding a node in it. At each expansion it evaluates the total description length (DL), I(Gs)+I (GjGs), of the input graph G which is defined as the sum of the two: DL of the subgraph, I(Gs), and DL of the input graph, I (GjGs), in which all the instances of the subgraph are replaced by single nodes. It stops when the subgraph that minimizes the total description length is found.
B. ILP Based Approach
The first system to search for the wider class of frequent substructure in graphs named WARMR was proposed in 1998. They combined ILP method with Apriori-like level wise search to a problem of carcinogenesis prediction of chemical compounds. The structures of chemical compounds are represented by the first order predicates such as atomel(C;A1; c), bond(C;A1;A2;BT), aromatic ring(C; S1) and alcohol(C; S2). The first two state that A1 which is a carbon atom bond to A2 where the bond type is BT in a chemical compound C. The third represents that substructure S1 is an aromatic ring in a chemical compound C, and the last represents that S2 is an alcohol base in C. Because this approach allows variables to be introduced in the arguments of the predicates, the class of structures which can be searched is more general than graphs. However, this approach easily faces the high computational complexity due to the equivalence checking under µ-subsumption (an NP-complete operation) on clauses and the generality of the problem class to be solved. To alleviate this difficulty, a new system called FARMAR has recently been proposed. It also uses the level wise search, but applied less strict equivalence relation under substitution to reduced atom sets. FARMAR runs two orders of magnitudes faster. However, its result includes some propositions having different forms but equivalent in the sense of the µ-subsumption due to the weaker equivalence criterion. A major advantage of these two systems is that they can discover frequent structures in high level descriptions. These approaches are expected to address many problems, because many context dependent data in the real-world can be represented as a set of grounded first order predicates which is represented by graphs.
C. Inductive Database Based Approach
A work in the framework of
inductive database
ha
ving
practical
computational
efficiency is MolFea
system
based on the level-wise version space
algorithm.
This
method performs the complete search of
the
paths
embedded
in a graph
data
set where the
paths
satisfy
monotonic
and
anti-monotonic
measures
in
the
version space. The version space is a
searc
h
subspace in a lattice
structure.
The monotonic
and
anti-
monotonic
mining measures described in define borders in the version
space.
T
o define the borders
,
the minimal a maximal elements of a set in terms of generality are
introduced.
D. Mathematical Graph Theory Based Approach
The
mathematical
graph
theory
based approach mines a complete set of
subgraphs
under mainly support
measure.
The initial work is
AGM
(Apriori-based Graph
Mining) system.
The
basic
principle of AGM is similar to the Apri
ori
algorithm
for basket analysis
. Starting from
frequen
t
graphs where each graph is a single
v
ertex,
the
frequen
t
graphs having larger sizes
are
searched in
bottom
up manner by
generating
candidates
having an extra
v
ertex.
ISSN: 2277-3754
ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 2, August 2012
49 An edge should be added between the extra vertex and some of the vertices in the smaller frequent
graph
when
searching for the connected
graphs.
One
graph
constitutes
one
trans-
action. The graph
structured
data
is
transformed without
much computational effort into an adjacency
matrix
mentioned. Let the number of vertices con
tained
in a graph be its
“size”,
an
adjacency
matrix
of a graph whose size is k be
Xk
,
the
ij
th
elemen
t
of
Xk
,
xij
and its
graph, G(Xk
).
AGM
can handle the graphs
consisting
of labeled
vertices
and labeled edges.
E. Kernel Function Based Approach
As explained in the last paragraph in subsection 2.3, a kernel function K defines a similarity between two graphs Gx and Gy. For the application to graph-based data mining, the key issue is to find the good combinations of the feature vector XG and the mapping defines appropriate similarity under abstracted inner product. A recent study proposed a composition of a kernel function characterizing the similarity between two graphs Gx and Gy based on the feature vectors consisting of graph invariants of vertex labels and edge labels in the certain neighbor area of each vertex. This is used to classify the graphs into binary classes by SVM mentioned in subsection. Given training data consisting of graphs having binary class, the SVM is trained to classify each graph. Though the similarity is not complete and sound in terms of the graph isomorphism, the graphs are classified properly based on the similarity defined by the kernel function. Another framework of kernel function related with
graph structures is called “diff
usion kernel". Though this is not dedicated to graph-based data mining, each instance is assigned to a vertex in a graph structure, and the similarity between instances is evaluated under the diffusion process along the edges of the graph. Some experiments report that the similarity evaluation in the structure characterizing the relations among the instances provides better performance in classification and clustering tasks than the distance based similarity evaluation. This type of work is supposed to have some theoretical relation with graph-based data mining.
III. OUR STUDY RESULTS
Our study results indicate that Subdue CL, the graph-based relational concept learner, is competitive with logic-based relational concept learners on a variety of domains. This comparison has identified a number of avenues for enhancements. Subdue CL would benefit from the ability to identify ranges of numbers. We could
accomplish this by utilizing the system’s existing capability
to find similar but not exact matches of a substructure in the input graph. Numeric values within the instances could be generalized to the encompassing range. A graph-based learner also needs the ability to represent recursion, which plays a central part in many logic-based concepts. More research is needed to identify representational enhancements for describing recursive structures
—
for example, graph grammars. Our future work will also focus on extending Subdue to handle other forms of learning, such as clustering. We are continuing our testing of Subdue in real-world applications. In biochemistry, for example, we are applying Subdue to data from the Human Genome Project to find patterns in the DNA sequences that indicate the presence of a gene-transcription-factor site. Unlike other approaches to finding patterns in gene data, Subdue uses a graph to represent structural information in the sequence. We hope that the discovered patterns will point to genes in uncharted areas of the DNA sequence. In another area of chemistry, we are applying Subdue CL to the Predictive Toxicology Challenge data. This data contains the structural descriptions of more than 300 chemical compounds that have been analyzed for carcinogenicity. Each compound (except for about 30 held out for future testing) is labeled as either cancer-causing or not. Our goal is to find a pattern in the cancerous compounds that does not occur in the noncancerous compounds. So far, Subdue CL has found several promising patterns, which are currently under
evaluation in the University of Texas at Arlington’s
Department of Chemistry. In addition, we are applying Subdue to a number of other databases, including the Aviation Safety Reporting System database, US Geological Survey earthquake data, and software call graphs. Subdue has discovered several interesting patterns in the ASRS
database. Burke Burkart of UTA’s Department of Geology evaluated Subdue’s results on the geology data and found
that Subdue correctly identified patterns dependent on earthquake depth, often the distinguishing factor among earthquake types. These and other results show that Subdue discovers relevant knowledge in structural data and that it scales to large databases.
IV. CONCLUSION
There are many other studies related to graph mining. An approach is proposed to derive induced subgraphs of graph data and to use the induced subgraphs as attributes on decision tree approaches. The method can be used to find frequent induced subgraphs in the set of graph data. A method to completely search homomorphically equivalent subgraphs which are the least general over a given set of graphs and do not include any identical triplet of the labels of two vertices and the edge direction between the vertices within each subgraph. It show that the computational complexity to find this class of subgraphs is polynomial for 1/2 locally injective graphs where the labels of any two vertices pointing to another common node or pointed from another common vertex are not identical. However, many graphs appearing in real-world problems such as chemical compound analysis are more general, and hence the polynomial characteristics of this approach do not apply in real cases. In addition, this approach may miss many interesting and/or useful subgraph patterns since the
ISSN: 2277-3754
ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 2, August 2012
50 homomorphically equivalent subgraph is a small subclass of the general subgraph. In this paper, the theoretical basis of the graph-based data mining was explained from multiple points of views such as subgraph types, subgraph isomorphism problem, graph invariants, mining measures and search algorithms. Thus, representative graph-based data mining approaches were shown in the latter half of this article. Even from theoretical perspective, many open questions on the graph characteristics and the isomorphism complexity remain.
ACKNOWLEDGEMENT
We would like to thank Prof. Yudhveer Singh Moudgil from Uttaranchal Institute of Technology, Dehradun for his continuous support and guidance throughout reviewing this research paper. He specialize in Cryptography & network Security and Advance Computer Architecture. He is author of many technical books and published a plenty of research papers in national and international journals.
REFERENCES
[1]
MRDM'01: Workshop multi-relational data mining. In conjunction with PKDD'01 and ECML'01, 2002. http://www.kiminkii.com/mrdm/. [2]
R. Agrawal and R. Srikant. Fast algorithms for mining association rules. In VLDB'94: Twentieth Very Large Dada Base Conference, pages 487{499, 1994. [3]
J. Cook and L. Holder. Substructure discovery using minimum description length and background knowledge. J. Artificial Intel. Research, 1:231{255, 1994. [4]
L. De Raedt and S. Kramer. The levelwise version space algorithm and its application to molecular fragment finding. In IJCAI'01: Seventeenth International Joint Conference on Artificial Intelligence, volume 2, pages 853{859, 2001. [5]
A. Debnath, R. De Compadre, G. Debnath, A. Schusterman, and C. Hansch. Structure-activity relationship of mutagenic aromatic and heteroaromatic nitro compounds. Correlation with molecular orbital energies and hydrophobicity. J. Medicinal Chemistry, 34, 1991. [6]
L. Dehaspe and H. Toivonen. Discovery of frequent data log patterns. Data Mining and Knowledge Discovery, 3(1):7{36, 1999. [7]
T. Gaertner. A survey of kernels for structured data. SIGKDD Explorations, 5(1), 2003. [8]
W. Geamsakul, T. Matsuda, T. Yoshida, H. Motoda, and T. Washio. Classifier construction by graph-based induction for graph-structured data. In PAKDD'03: Proc. of 7th Pacific-Asia Conference on Knowledge Discovery and Data Mining, LNAI2637, pages 52{62, 2003. [9]
P. Geibel and F. Wysotzki. Learning relational concepts with decision trees. In ICML'96: 13th Int. Conf. Machine Learning, pages 166{174, 1996. [10]
T. Imielinski and H. Mannila. A database perspective on knowledge discovery. Communications of the ACM, 39(11):58{64, 1996. [11]
A. Inokuchi, T.Washio, and H. Motoda. Complete mining of frequent patterns from graphs: Mining graph data. Machine Learning, 50:321{354, 2003. [12]
I. Jonyer, L. Holder, and D. Cook. Concept formation using graph grammars. In Workshop Notes: MRDM 2002 Workshop on Multi-Relational Data Mining, pages 71{792, 2002. [13]
H. Kashima and A. Inokuchi. Kernels for graph classification. In AM2002: Proc. of Int. Workshop on Active Mining, pages 31{35, 2002. [14]
R. Kondor and J. Laferty. Difusion kernels on graphs and other discrete input space. In ICML'02: Nineteenth International Joint Conference on Machine Learning, pages 315{322, 2002. [15]
M. Kuramochi and G. Karypis. Frequent subgraph discovery. In ICDM'01: 1st IEEE Conf. Data Mining, pages 313{320, 2001. [16]
M. Liquiere and J. Sallantin. Structural machine learning with galois lattice and graphs. In ICML'98: 15
th
Int. Conf. Machine Learning, pages 305{313, 1998. [17]
H. Mannila and H. Toivonen. Discovering generalized episodes using minimal occurrences. In 2nd Intl. Conf. Knowledge Discovery and Data Mining, pages 146{151, 1996. [18]
B. Mckay. Nauty users guide (version 1.5). Technical Report Technical Report, TR-CS-90-02, Department of computer Science, Australian National University, 1990. [19]
A. Mendelzon, A. Mihaila, and T. Milo. Querying the world wide web. Int. J. Digit. Libr., 1:54{67, 1997. [20]
S. Muggleton and L. De Raedt. Inductive logic programming: Theory and methods. J. Logic Programming, 19(20):629{679, 1994. [21]
S. Nijssen and J. Kok. Faster association rules for multiple relations. In IJCAI'01: Seventeenth International Joint Conference on Artificial Intelligence, volume 2, pages 891{896, 2001. [22]
A. Srinivasan, R. King, and D. Bristol. An assessment of submissions made to the predictive toxicology evaluation challenge. In IJCAI'99: Proc. of 16th International Joint Conference on Artificial Intelligence, pages 270{275, 1999. [23]
V. Vapnik. The Nature of Statistical Learning Theory. Springer Verlag, New York. 1995. [24]
X. Yan and J. Han. gspan: Graph-based substructure pattern mining. In ICDM'02: 2nd IEEE Conf. Data Mining, pages 721{724, 2002. [25]
K. Yoshida, H. Motoda, and N. Indurkhya. Graph- based induction as a unified learning framework. J. of Applied Intel, 4:297{328, 1994.
Author’s Profile
Amit kumar Mishra is Asst. Prof. in Uttaranchal Institute of Technology Dehradun. He has done his M.Tech from Uttarakhand Technical University. He specializes in core areas of computer science and holds experience of more than 4 years. He is an innovative person with deep knowledge of Software Engineering, Data base management system, C Programming. He is involved in various other activities in developments and various advance industrial trends.
ISSN: 2277-3754
ISO 9001:2008 Certified International Journal of Engineering and Innovative Technology (IJEIT) Volume 2, Issue 2, August 2012
51
Pradeep Gupta is Asst. Prof. in Uttaranchal Institute of Technology Dehradun. He has done his M.Tech from Uttarakhand Technical University. He has total academic experience of four years. His area of research is computer graphics, distributed system, and operating system. He is a man of high self esteem. Ashutosh Bhatt is Asst. Prof. in Uttaranchal Institute of Technology Dehradun. He has done his M.Tech from Graphic Era University Dehradun. His area of research is computer organization and data structure. He has a successful career in teaching with 3 years of experience. Jainendra singh Rana is Asst. Prof. in Uttaranchal Institute of Technology Dehradun. He has done his M.Tech from Uttarakhand Technical University. His area of research is data warehouse & data mining and mobile computing. Presently his area of research is graph based data mining. He started his career in Software Industry as a programmer, and then he moved to academics. There he attained great heights in research and training aspects of the field of computer science.

We Need Your Support

Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks