News document analysis by using a proficient algorithm

K.Meena. Int. Journal of Engineering Research and Application www.ijera.com
ISSN : 2248-9622, Vol. 7, Issue 6, (Part -2) June 2017, pp.07-13
www.ijera.com DOI: 10.9790/9622-076020713 7 | P a g e
News document analysis by using a proficient algorithm
K.Meena*
, R.Lawrance**
*(Research Scholar, Research & Development Centre, Bharathiar University, Coimbatore, India)
** (Director, Department of Computer Applications, Ayya Nadar Janaki Ammal College, Sivakasi, India)
ABSTRACT
News articles analyzing is one of the emerging research topic in the past few years. News paper discusses
various types (political, education, employment, sports, agriculture, crime, medicine, business, etc) of news in
different levels such as International, National, state and district level. In this news articles, crime discussion
plays a major role because one crime leads to a many other crimes and also affect many other lives. In India,
Madurai is one of the important places which have many historical monuments. Madurai is a sensitive place.
This paper analyzes the crimes which occur in the year 2015 in and around Madurai. This analysis helps to
police department to reduce the occurrence of crime in the future. This proposed system used Support Vector
Machine (SVM) for effectively classify the document. News documents are preprocessed using pruning and
stemming. From the stemmed words, the informative words are selected and weighted using feature selection
methods such as Term-Frequency and Inverse Document Frequency (TF-IDF) and Chi-square. It returns the
high dimensional vector space. It is reduced to low dimension using Latent Semantic Analysis (LSA) method.
Compute the cosine similarity between the key document and news documents. Based on the value, the news
documents are labeled as crime and non-crime. Some of the documents are used to train the SVM classifier.
Some of the documents are used to test the performance of developed system. From the comparative study, it is
identified that the performance of the proposed approach improves the classification accuracy.
Keywords: News document, Chi-square, Latent Semantic Analysis, Cosine-similarity, Support Vector Machine
I. INTRODUCTION
News articles are extremely occasion
sensitive by nature. News articles are used to know
the events which are happen in the state, center and
international levels. It is used to express the
thoughts of people, president of people, ministers,
chief ministers and prime minister. It also makes
awareness about the schemes of government,
criticism of schemes and status of state, center,
country and inter-national level. News articles
consist of medical tips, business news and new
methodologies of agriculture used to people. It also
consist of sports news, cinema news which are
attracted by the people those who are interested in
sports and cini field. Even though nowadays
various mediums are available to spread the news,
news articles are usually read by the all people to
know the incidents of previous day. Employment
opportunity column in news paper is used for the
people those who are search for the jobs. Jeweler
and textile agencies used this media for advertise
their products. Two wheelers and four wheelers
Company also use this media to publish their new
products. General knowledge, political news, new
inventions, court judgments, crime details and
police department actions are also available in the
news paper.
The task of analyzing the news items is
both interesting and challenging. Past research has
dealt with forecasting the popularity of news items
on web. Crime analysis is a key step in the
sequence of activities aimed at conceiving,
implementing, and evaluating measures to prevent
crime. Crime Analysis is the qualitative and
quantitative study of crime and police related
information in combination with socio-
demographic and spatial factors to apprehend
criminals, prevent crime, reduce disorder and
evaluate organizational procedures.
This proposed system analyzes the crime
on Madurai region from the January 2015 to
December 2015. Police departments face a number
of crime problems because Madurai region is one
of the sensitive areas in Tamilnadu. The distinctive
crimes meet by the public, such as murder, theft,
alcohol-related problems such as underage drinking
and drunk driving, as well as assaults, sexual
assaults, and rape. Such problems often consume
many of their resources. This system can help to
police departments in preventing such crimes in or
around Madurai through the analysis of crime
occur in previous year. The previous year analysis
is used to determine why a problem was occurring,
who was responsible, who was affected, where the
problem was located, and what form the problem
takes. The implementation of an adapted set of
actions that address the most important findings of
an analysis phase. Responses typically focus on at
least two of the following: (1) preventing future
occurrences by deflecting offenders; (2) protecting
RESEARCH ARTICLE OPEN ACCESS

likely victims; or (3) making crime locations less
conducive to problem behaviors. The focus of this
system reflects the fact that this type of analysis is
used for police departments and that there is an
increasing demand for the application of crime
analysis techniques that utilize data available from
new papers.
The news articles for this study collected
from English news papers in and around Madurai
region. Multi dimensional features are selected
from these articles. This paper uses TF-IDF, chi-
square and Latent Semantic Analysis (LSA) to
select the informative features. The primary
objectives of the proposed system are 1) to select
enlightening features related to crime 2) to train
and test the SVM classifier model using the
selected features with different kernel settings 3) to
provide information to the police department about
the analysis of crime in and around Madurai. This
paper is arranged as follows. Section 2 provides
details of related works. Section 3 illustrate about
the crime analysis of proposed system. Section 4
discusses the results and discussions. Section 5
concludes the proposed work.
II. RELATED WORKS
Roja bandari et al.[1] analyze the news
articles and predict the online popularity of that
articles using support vector machine. Steven Pires
and Ronald Belance[2] discusses the crime
problems faced by the police departments in and
around the college campus. Geographic
Information System is used to identify the spots for
criminal behavior has been mostly useful for police
to reduce problem.
Masoumeh Zareapoor and Seeja K. R.[3]
proposed various feature selection and feature
extraction method used for email classification. It
also shows that Latent Semantic Analysis(LSA)
outperforms other methods used for feature
extraction. Monica Rogati and Yiming Yang [4]
performed an wide study of the performance of
over 100 variants of 5 filter feature selection
methods using two benchmark collections (Reuters
21578 and part of RCV1) and four classifiers
(Naive Bayes, Rocchio, K-Nearest Neighbor and
Support Vector Machines). The result of the study
shows that the methods which include chi-square
attain high performance in classification. Tehseen
Zia, Qaiser Abbas and Muhammad Pervez Akhtar
[5] conducted a study to analyze the performance
of five feature selection method such as
information gain, gain ratio, Chi statistics,
symmetric uncertain and OneR using six classifiers
(naive Bayes, KNN, support vector machine with
linear, polynomial and radial basis kernels and
decision tree) on two Urdu test collections: naive
collection and EMILLE collection. It shows that
Linear SVM with feature selection methods IG and
Chi is outperformed other combinations of
classifiers.
III. CRIME ANALYSIS
3.1 Preprocessing
Methods used for preprocessing is used to reduce
the size of the file.
3.1.1 Pruning
This process removes the stop words used such as
'a', 'an', 'and' etc. A Newspaper article has various
stop words. Pruning used to increase the predictive
accuracy of the articles by removing stop words.
3.1.2 Stemming
This process produces the stem of the
word. In this news articles, a word is occur in
various formats such as word rob is appear as
‘robbery’ in one document, ‘robbed’ in another
document. Stemming is used to avoid counting
'robbery' and ' robbed ' as two separate words.
Stemdocument function in r is used to remove the
suffixes attached to the words.
3.2 Term-document matrix
Construct term-document matrix after
preprocessing is completed. This matrix is
constructed with the term frequency only. In this
matrix, row represents the words and column
represents the articles. Each cell contains number
which tells number of times that word appears in
the particular article.
3.3 Feature selection
Feature selection methods used to remove
the features that are irrelevant and also used to
strengthen the weight of the relevant words.
Methods used for feature selection use an
estimation function applied to a single word [6].
Various feature selection methods such as term
frequency, document frequency, information gain,
inverse document frequency, probability ratio and
gain ratio etc. are available [7]. These methods
commonly used in information retrieval to find the
individual words and their occurrence or co-
occurrence.
3.3.1 Term Frequency
Number of occurrences of a term in a document is
called term frequency. TF is calculated by add 1
with the value of taking log for the frequency of
term t occurs in the document cd.
3.3.2 TF-IDF (Term Frequency – Inverse
Document Frequency)
IDF is calculated by dividing the total number of
answer by the number of document containing the

term, and then taking the logarithm of that quotient.
Here N represents the total number of documents,
CD represents the documents and t is the term in
the document.
These two methods are used to improve the
strength of words and it given as input to the
feature transformation
3.3.3 Chi-square
Chi-square chooses words that have the more
distribution across categories. If a term has a
higher chi-squared score, that term is more useful
[3]. This method used for measuring the
independence between occurrences of two words. It
is calculated by using
dchisq(list, df)
where list means weight vector values and df
represents the number of values that are free to
vary.
3.4 Feature transformation
Methods used for feature transformation is used to
reduce the feature set size. There are various
methods like Hyperspace Analog to Language
(HAL), Principle Component Analysis (PCA) and
Latent Semantic Analysis(LSA) for feature
transformation.
3.4.1 Latent Semantic Analysis
Latent Semantic Analysis is a natural
language processing technique used for analyzing
the relationship between the set of documents and
the terms. LSA literally means analyzing
documents to find the underlying meaning or
concepts of those documents [8]. It is a fully
automatic mathematical technique for extracting
and inferring relations of expected contextual usage
of words in the passages of discourse which returns
a matrix. In this matrix, rows represent the unique
words and columns represent each paragraph. The
mathematical technique of Single Value
Decomposition (SVD) is used to reduce the matrix.
Latent Semantic Analysis constructs a rectangular
matrix of words by passages, with each cell
containing a transform of the number of times that
a given word appears in a given passage. The
matrix is then decomposed in such a way that every
passage is represented as a vector whose value is
the sum of vectors standing for its component
words. Similarities between words and words,
passages and words are then computed using cosine
similarity function. LSA works well on dataset
with diverse topics. This method is reliable and
faster than other dimensionality reduction methods.
The weighted matrix(m) is converted into semantic
space using the function lsa
The format of semantic space is converted into an
answer term matrix using the function
3.4.2 Principal component analysis
Principal component analysis is a method
used for extracting vital variable from a large set of
variables from the data set. It is used to capture
essential information and also used to reduce the
size of the data set. PCA is performed on a
symmetric correlation matrix. It should be numeric
and have standardized data. In PCA, first principle
components are identified. A principle component
is a standardized linear combination of the original
predictors in a data set. Executing PCA on un-
normalized data will lead to large loadings for
variables with high variance. Before finding the
principle component, data should be normalized.
For given m*n dimensional data, min (m-1, n)
principle components can be constructed where m
represents the number of documents n represents
the number of terms. The correlation between these
components should be zero. It is an unsupervised
approach because the response variable is not used
to determine the direction of these components.
prcomp(matrix) used to compute the principle
component analysis on the given input and return a
list. In that list, the value of rotated data is used for
calculation.
Using these procedures, construct the high
dimensional vector space for the articles and the
key factors. The key factor consists of words which
are considered as crime. With the articles vector
and the key factor vector, compute the cosine
similarities. Resultant score is stored in a separate
file. According to the score, label (Crime, Non
crime) is given to the article. This resultant article
name, score and label is given as input to the
classification process.
3.5 Classification
Many text classifications have been
proposed in data mining. Some of them are
decision trees, naïve bayes, rule induction, neural
network and Support Vector Machine (SVM).
Naive bayes is often used in text classification
because of its simplicity and effectiveness [9].
3.5.1 Support Vector Machine
Support Vector Machine is the supervised
learning methods used for classification and
regression tasks that are originated from the
statistical learning theory. It is used in many fields
such as bio informatics, text mining and image
recognition. SVM is a machine learning technique
used as to minimize the error based on the
Structural Risk Minimization (SRM) principal [10].

SVM provides only global minima by using the
concept of convex quadratic programming [11].
SVM was developed for binary classification. The
fundamental feature of SVM is the operating
maximum margin hyper plane whose position is
determined by maximizing distance the distance
from support vectors. The training point which is
closer to the optimal hyper plane is called support
vector. If the decision surface is obtained, it can be
used for classifying new data. With the help of
training data, SVM maps the input space into a
high dimensional feature space. Consider a
training data set of label pairs (pi,qi) where piεR,
qiε{1,0} and i=1,2…. m. In the data set, p is an
input vector space that contains the important
words. The class label 1 denotes the crime
document and the label 0 denotes the non-crime
document. SVM classification finds an optimal
separating hyper plane that can distinguish the two
classes from the training data set. The SVM
contains many arguments like formula, data type,
kernel, scale degree, gamma and cost. The kernel
attribute is used in training and predicting such as
linear, sigmoid and radial. SVM is used in this
proposed system due to its high performance,
capacity to deal with the high dimensional data.
3.6 Flow graph and algorithm of proposed
approach
The proposed document based SVM
classification system takes the documents and
generates a model that classifies the new
documents into different predefined class.
Preprocessing work is used to reduce the size of the
file. Feature selection and feature transformation is
important because it is used to improve the
performance of the classification process. Figure 1
shows the flow process of the proposed approach.
The documents are split into two group: Training
documents, Test documents. Training documents
are used to train the classifier. Test documents ate
used to evaluate the action of the developed
system.
Fig. 1 Flow graph of proposed work
Algorithm:
Algorithm I CCLS
Objective: To classify the documents
Input : Documents and Key document
Output: Documents are classified
Procedure:
1. Prune the documents
2. Stem the documents
3. Term-document matrix is constructed
4. TI-IDF and chi-square weighting scheme is used for feature selection
5. LSA is used for feature transformation
6. Compute cosine similarity is between the document vector returned from the LSA and key
document vector
7. According to the cosine similarity values label is given to the documents
8. Some of the documents are considered for training and some of the documents used for testing
9. SVM classification is used to classify the documents as crime and non-crime.

IV. RESULTS AND DISCUSSION
This section presents the result of the
proposed approach using news articles in the year
2015. It is implemented in R software. Collected
articles are stored as separate documents and
keywords related to the crime are stored in a
separate document named as key document.
Documents are classified into two types: crime
document and non-crime. The documents and key
document are preprocessed using pruning and
stemming process.
Feature selection is the important process
used to select small subset of features that diminish
redundancy and make the most of relevance to the
target [12]. In this proposed system, some of the
feature selection techniques such as term
frequency, TF-IDF and chi-square are used to
select the features that are highly discriminative. It
leads to enhanced learning performance of
classification, lower computational cost and leads
to develop improved model. Table I shows the
classification result after using feature selection
method.
Table I Classification rate after using feature selection methods
The stemmed words are weighted using
TF-IDF and chi-square methods. After feature
selection, input space contains some set of features.
If the number of data in feature selection is too
large, input space will be transformed into a
reduced set of features. So subset of features
selected from feature selection method is given as
input to feature transformation techniques such as
PCA and LSA. It maps the original feature space
into a new feature space with lower dimensions. In
PCA, best eigenvectors are selected as new features
and rests are discarded [3]. LSA is a technique
mainly used for text classification. It examines an
association between a term and concepts in a free
format text. It also returns the features which are
mostly related to terms and documents in a lower
dimension. Table III shows the classification result
after using feature selection and feature
transformation methods.
Table III Classification rate after using feature selection and feature transformation methods
Compute the cosine similarity between the
vector for the document and vector for key
document. Based on the value, the documents are
labeled as crime and non-crime. These documents
are given as input to the SVM classifier. Some of
documents are used to train the classifier. The
performance of the SVM classification is tested
with test documents. The test set can be classified
in one of four ways: True positives (the number of
crime documents classified correctly), True
negatives (the number of non-crime documents
classified correctly), False positives (the number
non-crime documents misclassified) and False
negatives (the number of crime documents
misclassified). Table III represents the results of
classification using proposed algorithm CCLS with
other methods. Figure 2 shows the classification
accuracy of proposed method and various other
methods.
This crime data set is also analyzed in
different ways which are useful to the crime
department. Figure 3 shows the number of crime
occurred in the months of 2015 in and around
Madurai. From that figure, crime department able
to analyze the crime rate of months. June month
has the highest crime rate. March month has the
lowest crime rate.
Feature selection
method
% of crime documents
correctly identified
misidentified
TF 63 37
TF-IDF 74 23
Chi-square 72 24
TF-IDF + chi-square 77 23
Feature selection + Feature
transformation method
correctly identified
misidentified
TF-IDF + PCA 49 51
TF-IDF + LSA 88 12
TF-IDF + PCA+ LSA 58 42
TF-IDF + chi-square+ LSA 96 4
TF-IDF + chi-square+ PCA 64 46

Table III Comparison of proposed algorithm with other methods
Method Classification rate (%) Error rate (%)
TF-IDF + chi-square+ LSA +Naïve bayes 91 9
TF-IDF + chi-square+ LSA +SVM radial 87 13
TF-IDF + chi-square+ LSA +SVM sigmoid 82 18
CCLS 100 0
Fig. 2. Comparison of proposed with other methods
Fig. 3. Crime rate of months in the year 2015 in and around Madurai
0
20
40
60
80
100
120
TF-IDF + chi-
square+ LSA
+Naïve bayes
TF-IDF + chi-
square+ LSA
+SVM radial
TF-IDF + chi-
square+ LSA
+SVM sigmoid
CCLS
Accuracy
Classification rate (%)
Error rate (%)
0
20
40
60
80
100
120
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
NumberofCrimes
Month
Crime Report

V. CONCLUSION
News document has been classified by
using SVM. Sizes of the documents are reduced
using pruning and stemming. TF-IDF and Chi-
square are used to select the didactic words from
the document. The high dimensional vector space
is reduced using LSA. These methods improve the
classification accuracy of the SVM. The proposed
algorithm produces better classification result when
compared with other methods. This analysis is
mostly helpful to the law enforcement department
to categorize the crime and also used to reduce the
crime rate of the month.
REFERENCES
[1]. Bandari, Roja, Sitaram Asur, and Bernardo
A. Huberman. "The pulse of news in social
media: Forecasting popularity." arXiv
preprint arXiv:1202.0332 (2012).
[2]. Steven Pires and Ronald Belance. “Mapping
Smoking Violations on a College Campus:
Implications for Prevention”. Crime
Mapping & Analysis News, Issue 4, 2015
[3]. Zareapoor, Masoumeh, and K. R. Seeja.
"Feature Extraction or Feature Selection for
Text Classification: A Case Study on
Phishing Email Detection." International
Journal of Information Engineering and
Electronic Business 7.2 (2015): 60.
[4]. Rogati, Monica, and Yiming Yang. "High-
performing feature selection for text
classification." Proceedings of the eleventh
international conference on Information and
knowledge management. ACM, 2002.
[5]. Zia, Tehseen, Qaiser Abbas, and
Muhammad Pervez Akhtar. "Evaluation of
Feature Selection Approaches for Urdu Text
Categorization." International Journal of
Intelligent Systems and Applications 7.6
(2015): 33.
[6]. Soucy, Pascal, and Guy W. Mineau.
"Feature selection strategies for text
categorization." Conference of the Canadian
Society for Computational Studies of
Intelligence. Springer Berlin Heidelberg,
2003.
[7]. Ikonomakis, M., S. Kotsiantis, and V.
Tampakas. "Text classification using
machine learning techniques." WSEAS
transactions on computers 4.8 (2005): 966-
974.
[8]. Landauer, Thomas K., Peter W. Foltz, and
Darrell Laham. "An introduction to latent
semantic analysis." Discourse processes
25.2-3 (1998): 259-284.
[9]. Kim, Sang-Bum, et al. "Effective methods
for improving naive bayes text classifiers."
Pacific Rim International Conference on
Artificial Intelligence. Springer Berlin
Heidelberg, 2002.
[10]. Vapnik, Vladimir. The nature of statistical
learning theory. Springer science & business
media, 2013.
[11]. Chen, Hui-Ling, et al. "A support vector
machine classifier with rough set-based
feature selection for breast cancer
diagnosis." Expert Systems with
Applications 38.7 (2011): 9014-9022.
[12]. Tang, Jiliang, Salem Alelyani, and Huan
Liu. "Feature selection for classification: A
review." Data Classification: Algorithms and
Applications (2014): 37.

News document analysis by using a proficient algorithm

Recommended

More Related Content

Similar to News document analysis by using a proficient algorithm (20)

Recently uploaded (20)

News document analysis by using a proficient algorithm