Download Paper Title (use style: paper title)

DOI 10.4010/2016.1851
ISSN 2321 3361 © 2016 IJESC
Research Article
Volume 6 Issue No. 6
Automatic Discovery of Relevance Features from Text Documents
Using Text Mining
Vijay Ganpataro Ingawale1, Sunil Damodar Rathod2
PG Student1, Assistant Professor2
Department of Computer Engineering
Dr. D Y Patil SOE, Lohegaon, Pune, India
Abstract:
Currently, the huge task for group of Information Retrieval is to find the relevance feature in text documents which chooses
whether document is relevant or irrelevant. Most existing text mining techniques depend on term-based methodologies which
extract terms from a training set for combining relevant feature. In term-based methodology various meaning of same word in
different framework leads to produce polysemy and synonymy issue. The term based approach also experiences low level support
issue. Whereas pattern based text mining approach solve low level support issue but still this approach suffers from huge number
of noise pattern. In the propose work, a pattern based approach for text mining is discovered. This methodology finds frequent
sequential pattern and closed sequential patterns in text documents for distinguishing the most data substance of the documents
and extract valuable elements for text mining. It additionally characterizes extracted terms into three classifications: positive
terms, general terms, and negative terms. The different feature extracting technique finds positive and negative patterns in text
document as higher level features in order to weight low-level terms based on their specificity.
Keywords: Data mining, information retrieval, feature selection, text classification.
I.
INTRODUCTION
A web search engine contains number of web pages organized
by the page’s relevance to the user query. The problem with
web search relevance ranking is to form relevance of a page to
a query [12]. Nowadays, web search engines contain
numerous features to guess relevance [13].Information
Retrieval (IR) Systems is combination of Web and search
engines. These systems are considered to retrieve documents
from digital collections e.g. library abstracts, corporate
reports, news and so forth. Generally, IR relevance ranking
algorithms are used to gain high recall on medium sized
document collections using user query. Furthermore, textual
documents in these groups had practically no structure or
hyperlinks [12]. A web search engine contains many methods
for calculations of Information Retrieval Systems, however
required to adjust them to right their needs. Data mining
techniques are used to discovery important information from a
large amount of text documents on the Web. Several text
mining techniques have been introduced in order to get the
goal of retrieving valuable information for users [12]. Most of
them adapt the term based approach while others select the
pattern-based technique to produce a text representative for a
set of documents. Information Retrieval contains various
efficient term-based methods to solve this challenge [17]. The
advantage of term based technique includes effective
computational performance. In the existing work, many data
mining techniques have been introduced for feature discovery.
These tasks contain sequential pattern mining, frequent pattern
mining and closed pattern mining. The synonymy and
polysemy are the main problems related with term-based
methods [3], [9], and [11]. Polysemy denotes same word has
different meaning while synonymy denotes a different word
has the same meaning [3].Also pattern based techniques
surface from low frequency and misinterpretation problem [3].
In text document, the complex task is how to use discovered
patterns to accurately calculate the weights of useful feature
[3], [12].
International Journal of Engineering Science and Computing, June 2016
II. LITERATURE SURVEY
Nowadays web resources and its utilization are constantly
growing much over the time. User needs important
information quickly, while using web. There are various
documents in web and user need effective results while
searching the web. There are a few issues in web search [12],
for example, effective ranking and relevance. The IR group
challenges the test of dealing with a large amount of
hyperlinked information, however individuals from this group
can use demonstrating, document characterization and user
interfaces modifying to achieve their objectives [12] [13].
Information Retrieval models depend on ranking algorithm,
which is utilized as a part of web searchers to create the
ranked list of documents [6]. A ranking algorithm sorts an
arrangement of documents as indicated by their significance to
a give inquiry [8].
Feature selection is the system for selecting a subset of
relevant feature for use in model creation. In text document
feature can be term, design, sentence. However the
conventional component choice methods are not effective for
selecting text document for solving the relevance problem
because relevance is a single class issue [13]. The effective
way of feature selection method for relevance and methods is
based on a feature weighting function. A feature weighting
function determines feature occurrences in a document and
specifies the relevance of the feature.
The term-based Information Retrieval models contain the
Rocchio algorithm [13] [19], Probabilistic models, language
model and Okapi BM25 [19]. In a language model, the key
idea is the probabilities of word successions which incorporate
both sentences and words. They are usually approximated by
n-gram models [13], as Unigram, Bigram what's more, and
Trigram, for consider term dependence. In the late work
important issue for feature selection in a text document is to
recognize structure of the documents. Text feature can be a
single word or complex structure. It contains different
complex structures, for example, n-grams, pattern and term.
7892
http://ijesc.org/
The different effective algorithm, for example, Apriori
algorithm, FP-tree, SPADE, Prefix Span, GST and SLPMiner
[4], [5], [6], [7], and [8] have been proposed. Patterns can be
discovered by data mining techniques like sequential pattern
mining, closed pattern mining [2] and frequent item set
mining. To overcome the disadvantages of sequential patterns
And closed patterns have been established in pattern discovery
technique [18]. Feature classification is assigning different job
as per predefine gathering of documents. There is various
grouping classifications, for example, Rocchio, Naive Bayes,
KNN what's more, SVM have been utilized as a part of
Information Retrieval [14], [15], [16]. Support Vector
Machine (SVM) is one of the fundamental grouping
techniques utilized as a part of machine learning method [14].
The gathering issues join the single and multi-marked issue.
Term based model document having semantic importance and
documents are broke down on the basis of the term. The
regular strategy [13] to the various named issues is to separate
it into a couple of classifiers, where a classifier distributes two
predefined classifications. The two classifications are positive
or negative classifier. Term based strategy experience the ill
effects of the issue of polysemy and synonymy [10]. Polysemy
supposes a word has various meaning’s and synonymy
suggests various words having the same meaning. Further
patterns in the same groups are into a master pattern that
comprises of an arrangement of terms which are made into a
term-weight distribution. It is still a challenging issue for
pattern-based method to manage low frequency patterns.
In summary, the recent techniques for discovering relevance
features are separated into three methodologies. The first
methodology considers feature terms that turn out in both
positive and negative patterns that are Rocchio-based models
[19] and SVM [14]. The second approach is created on
probabilistic based models [15] in which terms appear or don't
appear in positive and negative documents which describes
their significance. The third approach considers just positive
patterns from the documents [11].
extraction information are examine on pattern basis [2].
Pattern can be found by data mining routines like sequential
pattern mining, closed pattern mining and frequent item set
mining.
With the continued use of web information, it has turned out
to be more critical to give improved mechanism to discover
data quickly. Information Retrieval frameworks rank the
documents based on maximizing relevance to the user query
[12], [13]. A document ranking method is one where every
document in the positioned by importance and data.
Relevance Feature contains three specific features positive,
negative and general features. Thus, the most important
research question is how to find the most exact relevant
classifier for positive documents and negative documents, for
a given set of features. In this propose work, term
classification strategy contains two exact components as per
testing sets. In this propose work focus on approximation
approach to deal with discover the relevance feature.
B. Problem Definition
To design and implement relevant and irrelevant text
documents using text mining. The text mining approach has
problems of polysemy and synonymy. This approach
discovers closed and frequent sequential pattern in text
documents to identify the most important content from the
documents and extract useful features for text mining. It also
helps to classify extracted terms into three categories: positive
terms, general terms, and negative terms.
C. System Architecture
System architecture serves as the blueprint of relevant feature
discovery in text document using text mining, defining the
work assignments that must be carried out by design and
execution. The architecture is contains different phases for
calculating which document are more relevant or irrelevant.
Sample References Section
A. Taxonomy chart
The taxonomy chart given below shows the comparison of
various existing methods. The constraints used here to for
comparison gives us clear idea where in a concrete work can
be done.
Figure 1. Taxonomey chart for various existing methods.
III. PROPOSED APPROCH FRAMEWORK AND DESIGN
In proposed approach most text analysis for document
classification incorporates a stage of text extraction to decide
the words or terms that occur in every documents. In pattern
International Journal of Engineering Science and Computing, June 2016
Figure 2. System Architecture for relevant feature discovery
for text documents using text mining.
7893
http://ijesc.org/
D. Algorithm
In proposed work an Efficient Fclustering is used.
Input: Set of DP+, DP-, General terms and function spe.
Output: There are three categories of terms T+, G and T-.
1. Start.
2. Select the folder contain all documents (T+, T-, and G).
3. User decides the term extraction with minimum value.
4. Starting merging process in three clusters.
5. Perform term support weight calculation for all documents.
6. Ranking the document according to term support.
7. Assign term class specification using clustering algorithm.
8. Stop.
E. System Modules
i. Term extraction module
The module provides text analysis for example document
classification incorporates a phase of text extraction to decide
the words or terms that occur in each document. Text feature
extraction relies on upon some meaning of which characters
are to be deal with as word characters versus non-word
characters.
2.
The pattern cleaning method is useful for reducing
the noise in discovered patterns from positive
feedback documents.
Figure2. The MAP performance of different patterns with
min_sup
ii. Pattern extraction module
In this module pattern extraction documents are broken down
on pattern basis. Patterns can be created by data mining
techniques like frequent item set mining, sequential pattern
mining and closed pattern mining.
iii. Document ranking module
In document ranking module systems rank the documents
according their relevance. A document ranking technique is
one where each document in the ranked according to query
relevance and information.
Figure3. The comparison of closed sequential pattern and
sequential pattern
iv. Term classification module
In relevance Feature discovery is done in this model and the
features are classified as three specific term, a positive,
negative and general terms. In this propose work, term
classification method requires two empirical factors following:
1. Training method.
2. Testing method.
IV.
DATASET
In proposed work Reuters-21578 corpus dataset are used for
test the model. Reuters-21578 corpus is a typically used
collection for text mining. The data was originally collected
and labeled by Carnegie Group, Inc. and Reuters, Ltd. in the
course of developing the CONSTRUE text categorization
system. In proposed work we also used a TREC dataset for
test the model. TREC data supplier also provide reliable
assessor topics for RCV1, aiming at testing robust information
filtering and information retrieval systems.
V. EXPERIMENTS AND RESULTS
This section describes experimental evaluations of the
proposed framework.
These two hypotheses are:
1. A post-processing method for frequent patterns in
text is necessary to improve the quality of extracted
features for describing user information needs or
preferences.
International Journal of Engineering Science and Computing, June 2016
Figure4. Comparison of PTmining with text mining based
methods
VI. CONCLUSION AND FUTURE ENHANCEMENT
Methodology is proposed for extracting relevance feature
discovery in text documents. Different data mining techniques
have been proposed. Frequent item set mining, closed pattern
mining, sequential pattern mining, and closed pattern mining
these all techniques are used in data mining techniques. The
pattern deploying and pattern evolving techniques are used in
proposed Method. In this research work problems of low
frequency and misinterpretation in pattern mining techniques
are discussed. The paper also defines different approaches for
relevance feature discovery.
ACKNOWLEDGMENTS
It gives me a great pleasure and immense satisfaction to
present this paper of topic “Relevant feature discovery from
text documents using text mining” which is the result of
unwavering support, expert guidance and focused direction of
7894
http://ijesc.org/
my guide Mr. Sunil. D. Rathod to whom I express my deep
sense of gratitude and humble thanks, for his valuable
guidance.
[14] D. D. Lewis, Y. Yang, T. G. Rose, and F. Li. Rcv1: A
new benchmark collection for text categorization
research. J. Mach. Learn. Res., 5:361–397, December
2004.
REFERENCES
[1] Jaillet, S., Laurent, A., Teisseire, and M.: Sequential
patterns for text categorization. Intelligent Data Analysis
10 (3), 199–214 (2006).
[2] Wu, S., Li, Y., Xu, Y., Pham, B., Chen, and P.:
“Automatic pattern-taxonomy extraction for web mining”.
In: 3th IEEE/WIC/ACM WI International Conf. In Web
Intelligence, pp. 242–248 (2004).
[3] Zhong, N., Li, Y., Wu, S.: Effective pattern discovery for
text mining. IEEE Transactions on Knowledge and Data
Engineering,
doi:
http://doi.ieeecomputersociety.org/10.1109/TKDE,2011.
[4] D.B. Liu. Web data mining: exploring hyperlinks,
contents, and usage data. Data-centric systems and
applications. Springer, Berlin, 2007.
[15] X. Li and B. Liu. Learning to classify texts using positive
and unlabeled data. In Proceedings of IJCAI’03, pages
587–592, 2003.
[16] X. -L. Li, B. Liu, and S. -K. Ng. Learning to classify
documents with only a small positive training set. In
Proceedings of ECML’07, pages 201–213, Berlin,
Heidelberg, 2007.
[17] S. E. Robertson and I. Soboroff. The trek 2002 filtering
track report. In Proceedings of TREC’02, 2002.
[18] Y. Li, X. Zhou, P. Bruce, Y. Xu, and R. Y. Lau. Twostage Decision Model for Information Filtering. Decision
Support Systems, 52 (3): 706-716, 2012.
[19] T. Joachims. A probabilistic analysis of the rich algorithm
with tfidf for text categorization. In Proc. On ICML’97,
pages 143–151, 1997.
[5] A. Rakesh and R. Srikant. Mining sequential patterns. In
proceedings of the 11th International Conference on Data
Engineering, pages 3.14, 1995.
[6] X. Yan, J. Han, and R. Afshar. Clospan: Mining closed
sequential patterns in large data sets. In Data Mining
(SDM03), pages 166.177, 2003.
[7] J. Han and K. Chang. Data mining for web intelligence.
IEEE Computer, 35 (11): 64:70, 2002.
[8] M. J. Zaki. Spade: an efficient algorithm for mining
frequent sequences. In Machine Learning Journal, special
issue on Unsupervised Learning, pages 31-60, 2001.
[9] S. -T. Wu, Y. Li, and Y. Xu, “Deploying Approaches for
Pattern Refinement in Text
Mining,” Proc. IEEE Sixth
Int’l Conf. Data Mining (ICDM ’06), pp. 1157-1161,
2006.
[10] G. Salton and C. Buckley, “Term-Weighting Approaches
in Automatic Text Retrieval,” Information Processing and
Management: An Int’l J., vol. 24, no. 5, pp. 513-523,
1988.
[11] Y. Li, A. Agony, and N. Zhong. Mining positive and
negative patterns for relevance feature discovery. In
Proceedings of KDD’10pages 753–762, 2010.
[12] C. C. Yang. Search engine information retrieval in
practice. J.Am. Soc. Inf. SCI. Technol., 61:430–430,
2010.
[13] C. D. Manning, P. Raghavan, and H. Sects. Introduction
to Information Retrieval. Cambridge University Press,
2009.
International Journal of Engineering Science and Computing, June 2016
7895
http://ijesc.org/