Relation Extraction using Hybrid Approach and an Ensemble Algorithm

Integrated Intelligent Research (IIR) International Journal of Computing Algorithm
Volume: 05 Issue: 01 June 2016, Page No. 46- 49
ISSN: 2278-2397
46
Relation Extraction using Hybrid Approach and an
Ensemble Algorithm
R.Radha1
, S.Vijaya2
1
Associate Professor, Department of Computer Science, S.D.N.B.Vaishnav College for Women, Chromepet, Chennai.
2
Research Scholar, Department of Computer Science, S.D.N.B.Vaishnav College for Women, Chromepet, Chennai.
radhasundar1993@gmail.com,vijeshnisna@gmail.com
Abstract: The amount of information stored in the form of
digital has a gigantic growth in this electronic era. Extraction
of heterogeneous entities from biomedical documents is being
a big challenge as achieving high f-score is still a problem.
Text Mining plays a vital role for extracting different types of
entities and finding relationship between the entities from huge
collection of biomedical documents. We have proposed hybrid
approach which includes a Dictionary based approach and a
Rule based approach to reduce the number of false negatives
and the Random Forest Algorithm to improve f-score measure.
Keywords -- Biomedical documents, Heterogeneous entities,
Named Entity Recognition, Relation Extraction, Text Mining.
I. INTRODUCTION
Wide ranges of textual information are hidden in digital forms
in the data reservoirs. To extract required information and
discover new knowledge from these sources Text Mining plays
a vital role in this electronic era. Data Mining is used to
discover knowledge from structured data where as Text Mining
offers retrieval of text from the unstructured data collection,
extracts relation between the entities and leads to discover new,
hidden knowledge. In Biomedical field due to the increased
amount of data collection, covering all the information by a
researcher is tedious task. Text mining tool makes this tedious
task ease with its major activities such as Information
Retrieval, Information Extraction, Finding association,
Categorization and summarization etc.,[Hearst….][1].
Main focus of this paper is identifying relationship between
gene and disease because getting high performance for the
same is still a problem. The combination of Data and Text
mining is referred to as Duo-Mining[Creese.G,][2]. Many
researchers recommended duo-mining for good decision
making. The tasks involved in this work are i) Collecting
abstracts from PubMed database ii) Recognizing entities from
biomedical abstracts and Finding associations between
heterogeneous entities. PubMed abstracts are retrieved from
http://www.ncbi.nlm.nih.gov/pubmed , and parsed by
removing stop words. Biomedical Entities identified by
processing the parsed data with the proposed methods
Dictionary based approach, rule based approach and the
associations are classified using ensemble classifier Random
Forest Algorithm. This paper is organized as follows, Section 2
discusses related work based on Information Extraction,
Named Entity Recognition, Section 3 presents proposed work
and describes proposed methods. Section 4 discusses the
experimental results of the proposed algorithm and analyses
the performance of the proposed algorithm. Finally Section 5
includes the conclusion of this paper.
II. RELATED WORK
Alex Bravo et al.[6] developed BeFree system which is used to
identify relationships between heterogeneous biomedical
entities with a special focus on genes and the diseases
associated with them. The need for tools becomes necessary to
identify, gather relevant information and placing it in the
context of biomedical knowledge. In identifying complex
relationships between entities of biomedical domain several
challenges remain still. They have presented BeFree - a
supervised text mining system which composed of biomedical
NER(Named Entity Recognition) and a kernel based
RE(Relation Extraction) module , to identify relationships
between depression and the gene associated with depression,
from free text. They have developed a semi-automatic
annotation corpus. [J-D Kim et al][3] used supervised learning
approaches which require the annotated corpora for the
development and evaluation of Relation Extraction and their
results shown good performance . Maryam Khordad[4]
presented a system that provides an important as well as an
up-to-date data for database construction and updation, by
solving the problems such as Finding genotype names , finding
phenotype names and phenotype-genotype extraction.
A.S.Foulkes[5] had used population-based genetic association
studies to relate genetic sequence information derived from
unrelated individuals to a measure of disease progression or
disease status. Understanding the role of genetics in diseases is
one of the major goals of the post-genome era. Changqin Quan
et al[8] proposed Automatic gene-disease association approach
based on text mining and network analysis. They combined
information filtering , grammar parsing and network analysis
for gene disease association extraction. They have taken Breast
cancer as testing disease for system evaluation and checked
with 31 top ranked genes and closeness centralities through
NCBI database. The evaluation showed 83.9% accuracy for
testing genes and diseases, 74.2% accuracy for testing genes.
Hong-woo Chun et al.[9] described a system to extract disease-
gene relation from Medline , constructed a dictionary for
disease and gene names from six public databases and
extracted relation candidates by dictionary matching and used
machine learning based Named Entity Recognition to filter out
false recognitions of disease / gene names. Performance of
Relation Extraction was dependent upon the performance of
Named Entity Recognition filtering and improved the precision
of Relation Extraction by 26.7%.Their future work
encompassed increasing the size of the annotated corpus and
enriching annotation. Krauthammer et al[10] used Dictionary
based model and achieved 78.8% Precision and 71.1% Recall
value. Yi-Feng Lin et al[11] used Maximum Entropy model
based on features , combined with post processing and got the

ISSN: 2278-2397
47
Precision value 72.1%, Recall value 71.1% and 78.5% f-score.
D.Hanisch et al.[13] used Maximum Entropy method on
protein data set and the precision, recall and f-score they
achieved are 49.1%,62.1% and 54.8% respectively. Zhenfei Ju
et al[12] developed a biomedical named entity recognition
system with GENIA corpus data set and scored 84.24%
Precision and 80.76% Recall. For efficient retrieval of data we
have combined dictionary based approach and rule based
approaches.To improve the f-score measure we have proposed
an ensemble approach called Random Forest Algorithm in our
work.
III. PROPOSED WORK
Alex Bravo et al.[6] developed BeFree system which
discovered only a small proportion of the gene-disease
associations from expert curated databases . Our proposed
work aims to increase a large proportion of gene-disease
associations with high performance. The block diagram of our
proposed system is given below.
Block Diagram of Proposed work
3.1. Problem solving methodologies
The major task involved in the proposed work are as follows:
1.Information Extraction
2.Named Entity Recognition
The first task is concerned about the retrieval of abstracts from
PubMed database, the second task dealt with the identification
of Biomedical named entities and finding associations between
entities.
3.1.1.Information Extraction
Wide range of information is available in the form of electronic
documents. If the data is in tabular form it would be easier to
answer queries straight forward. So, the first task to be done is
converting unstructured data to structured data as extraction of
structured data from electronically available biological
literature has become important area of current research. In this
task we extracted abstracts from PubMed database and pre-
processed the abstract collections by segmenting words and
removing stop words.
3.1.2. Named Entity Recognition(NER)
Named Entity Recognition refers to the ability of recognizing
entities such as name of the person, location, date and time. In
the field of biomedical it recognizes entities such as genes ,
diseases, protein and drug. It mines useful knowledge. The
primary goal is identifying all textual mentions of the named
entities. There are two major tasks involved in NER , they are
i) Identifying the boundaries of the Named Entity ii)
Identifying its type.
3.2. Proposed Method
3.2.1 Dictionary Based Approach
The first step in extracting information from the biomedical
documents is Dictionary based entity name recognition.
Dictionary based protein name recognition is used in
extracting information from biomedical documents as it can
provide ID information on recognized terms. To overcome
false recognition mainly caused by short names Tsuruoka et
al[14] have used machine learning to filter out false positives ,
to improve recall rate they have used appropriate string
searching techniques and expanded the dictionary in advance
with a probabilistic variant generator. As Dictionary based
approaches have limitations such as false positive recognition
and lack of a unified resource that covers newly published
names they had addressed an approach with two-phase method
where the first method scans text for protein name candidates
and the second phase method filters irrelevant candidates by
utilizing a Naïve Bayes Classifier. In our work we have
proposed Rule based method to reduce the rate of false
recognition caused by Dictionary based method and to increase
the recall rate.
3.2.2.Rule Based Approach
Rule based approaches are used in Information Extraction to
deal with a large database, broader range of variations and
cover the word order variations. They utilize handmade rules
and patterns. Fukuda et al [17]presented a method to identify
core terms , feature terms and concatenates the terms by
utilizing handcrafted rules, patterns and extend the boundaries
to adjacent nouns and adjectives. Instead of Rule based
approach Proux et al.[20] used a tagger with a nondeterministic
finite automaton and scored 91.4% Precision, 94.4% recall and
92.9 % F-score.
3.2.3. Random Forest Algorithm
The Random Forest(RF) algorithm is an ensemble approach
and popular machine learning algorithm with in statistical
genetics. The main characteristics of RF are,i)Well adopted for
both prediction and Variable Importance(VI), ii) Robust to the
setting of tuning parameters. It has the capability of handling
large amount of observations and predictors, and this method
combines the “bagging” idea of Brieman and the random
feature selection .[Benjamin A et al][16]. RF develops lots of
decision tree based on random selection of data and random
selection of variables. It provides the class of dependent
variable based on many trees. As the trees are based on random
selection of data as well as variables, these are called as
random trees. Random Forest is an ensemble classifier that
uses many decision tree models. Ensemble models combine the
results from different models, the result from an ensemble
model is usually better than the result from one of the
individual models.
3.2.3.1: Steps involved in RF Algorithm
The following algorithm is used to construct each tree :
1. The number of training cases can be taken as N, and the
number of variables in the classifier as M.

ISSN: 2278-2397
48
2. ‘m’ should be much less than ‘M’ as the number of input
variables(m) to be used to determine the decision at a node
of the tree.
3. From all ‘N’ available training cases , choose a training set
for this tree by choosing ‘n’ times. The rest of the cases
can be used to estimate the error of the tree, by predicting
their classes.
4. Randomly choose ‘m’ variables for each node of the tree
on which to base the decision at that node. The best split
can be calculated based on those ‘m’ variables in the
training set.
5. 5.Each tree is not pruned but fully grown.
6. A new sample is pushed down the tree for prediction and
is assigned the label of the training sample in the terminal
node it ends up in. The average vote of all trees is reported
as random forest prediction by iterating over all trees in
the ensemble.
Random Selection of data
3.2.4 Implementation Steps
The steps involved in our work are as follows:
1.Dataset Collection.
2.Input Document.
3.Preprocessing the Document.
4.Comparing with genes and diseases.
5.Applying proposed algorithm.
3.2.4.1 Dataset Collection
We have used PubMed abtracts as dataset. The abstracts are
retrieved from http://www.ncbi.nlm.nih.gov/pubmed and
stored in a text file.
3.2.4.2 Input Document
The input document (text file in which retrieved abstracts are
stored) is read using file reading commands.
3.2.4.3 Pre processing the Document
The file read is pre processed by segmenting and removing
stop words.
3.2.4.4 Comparing genes and diseases
The terms (genes and diseases) are recognized by comparing
each term with the respective genes file and disease file
retrieved from genome database
(www.genome.jp/kegg/genes.html) and Disease database
(www.genome.jp/kegg/disease.html).
3.2.4.5 Applying proposed algorithm
Applied proposed methods ( dictionary-based approach, Rule
based approach and Random Forest Algorithm) on the pre
processed data.
IV. EXPERIMENTAL RESULTS
The f-score value of biomedical named entity recognition
system reached more than 85% but is less than general named
entity recognition that can reach about 90%[Zhenfei Ju et
al][12]. With the GENIA corpus data set they got 84.24%
Precision and 80.76% recall , where on the same data set using
HiddenMorkovModel[Zhou et al.][15]got 66.5% Precision and
66.6 Recall. With our proposed hybrid approaches
(Combination of Dictionary based approach and Rule based
approach ) with Random Forest ensemble classifier we
achieved 95.15% Precision and 92.64% Recall with 93.89%
F-score on gene names and 96.75% Precision and 96.70%
Recall with 96.73% F-score on disease names. Precision, an
F-Score are calculated as,
TP/
Precision(P) =
TP + FP
TP
Recall(R) =
TP + FN
(β2
+1)+(P*R)
F- Score =
β2
(R + P)
(As we weighted Precision and Recall equally, β = 1) . So we
calculated the F-Score as 2*(P*R)/(P+R).
V. CONCLUSION
The need of Text Mining tool to discover and extract relation
between entities is in a great need in the field of biomedical
due to the rapid growth of biomedical documents reservoirs.
To improve the performance in relation extraction between
gene and disease we used a hybrid approach with Dictionary
based approach , Rule based approach and used Random
Forest algorithm to improve f-score. Our experimental results
shown good results with 95% Precision, 92.64% Recall and
93.89% F-score on gene names . With disease names 96.75%
Precision, 96.70% Recall and 96.73% F-score.
References
[1] Hearst,M.A., “Untangling Text Data Mining”, Proc. 37th
Annual Meeting of the Association for Computational
Liguistics , College Park, MD,1999, pp.3-10.
[2] Creese, G.Duo-Mining:Combining data and text mining ,
DM Review,September,(2004).
[3] Kim J-D, Ohta T, Pyysala S, Kano Y, Tsujii, “Overview of
BioNLP’09 shared task on event extraction “, In
BioNLP’09 Proc Work Curr. Trends Biomed Nat Lang
Process Shar Task. Association for Computational
Liguistics;2009:1-9.
[4] Maryam Khordad, “Investing Genotype-Phenotype
Relationship Extraction from BioMedical Text”, The
University of Western Ontario, London, Ontrio, Canada.
[5] A.S.Foulkes,”Applied Statistical Genetics with R : For
population-based Association Studies, Use R”,

ISSN: 2278-2397
49
DOI:10.1007/978-0-387-89554-3-1.@Springer Science +
Business Media LLC 2009.
[6] Alex Bravo,Janet Pinero, Nuria Queralt, Michael
Rautschka and Laura l.Furlong.,”Extraction of relations
between genes and diseases from text and large-scale data
analysis: implications for translationl research”, bioRxiv
preprint first posted online July 24, 2014;
doi;http://dx.doi.org/10.1101/007443.
[7] Buyko, E..Beisswanger,E., & Hahn, U,(2012), “The
extraction of pharmcogenetic and pharmacogenomic
relations- a case study using PharmGKB, Pacific
Symposium on Biocomputing, 376-87. Retrieved from
http://www.ncbi.nlm.nih.gov/pubmed/22174293.
[8] Changqin Quan and Fujiren, “Gene-disease association
extraction by text mining and network analysis “,
Proceedings of the 5th
International workshop on Health
Text Mining and Information Analysis(Louhi)@ EACL
2014, pages 54-63, Gothenburg, Sweden, April 26-30
2014.(C)2014 Association for Computational Linguistics.
[9] Hong-woo Chun, Yoshimasa Tsoruoka, Jin-Dongkim,
RieShiba, Noiki Nagata,TeruyoshiHishiki and Jun’ichi
Tsujii,”Extraction of Gene-disease relation from Medline
using domain dictionaries and machine learning”,Pacific
Symposium on Biocomputing 11:4-15(2006) October 12
,2005,15:5 Proceedings.
[10]M.Krauthammer, A.Rzhetsky, P.Morozov, C.Friedman,
“Using BLAST for Identifying Gene and Protein Names in
Journal articles”, Gene 259(2000) 245-252.
[11]Yi-Feng Lin, Tzong-Han Tsai et a., “A Maximum Entropy
Approach to Biomedical Named Entity Recognition “,
BIOKDD04: 4th
Workshop on Data Mining in
Bioinformatics(With SIGKDD Conference)
[12]Zhenfei Ju,Jain Wang , FeiZhu, “Named Entity
Recognition from Biomedical Text using SVM”, 978-1-
4244-5089-3/11/ ©2011 IEEE.
[13]D.Hanisch, J.Fluck,H.Mevissen and R.Zimmer, “Playing
Biology’s Name Game: Identifying Protein Names in
Scientific Text” , In PSB ’03, 2003.
[14]Tsuruoka ,Y., Tsujii J, “Improving the Performance of
Dictionary-based Approaches in Protein Name
Recognition”, J Biomed Inform 2004,37:461-70.
[15]Zhou G, et al., “Recognizing Names in Biomedical Texts :
A Machine Learning Approach”, Bioinformatics,
Vol.20,No.4,2004,pp.1178-1190.
[16]Benjamin A,Goldstein, Eric C.Polley and Farren B.S.
Briggs(2011),” Random Forests for Genetic Association
Studies “, Statistical Applications in Genetics and
Molecular Biology: Vol. 10 : ISS.1. Article 32.
[17]Fukuda K., et al., “Toward Information Extraction:
Identifying Protein Names from Biomedical Papers”, Proc.
Pacific Symp.on Biocomputing, 1998, pp.707-718.
[18]Jong C.Park and Jung-jae Kim,”Named Entity
Recognition”, Text Mining for Biology and Biomedicine.
[19]www.youtube.com
[20]Proux D., et al., “Detecting Gene Symbols and Names in
Biomedical Texts: A First Step Toward Pertinent
Information.

Relation Extraction using Hybrid Approach and an Ensemble Algorithm

Recommended

More Related Content

Similar to Relation Extraction using Hybrid Approach and an Ensemble Algorithm (20)

More from MangaiK4 (20)

Recently uploaded (20)

Relation Extraction using Hybrid Approach and an Ensemble Algorithm