Named Entity Recognition in Odia Language: A Rule-based Approach

Full text

Int. J. Reasoning-based Intelligent Systems, Vol. 15, No. 1, 2023 15
Copyright © 2023 Inderscience Enterprises Ltd.
Named entity recognition in Odia language:
a rule-based approach
Amrita Anandika, Sujata Chakravarty and
Bijay Kumar Paikaray*
Department of CSE,
Centurion University of Technology and Management,
Odisha, India
Email: amrita.anandika@gmail.com
Email: chakravartys69@gmail.com
Email: bijaypaikaray87@gmail.com
*Corresponding author
Abstract: NLP can be defined as a process of automatic handling of human spoken languages by
computers. NLP is a vast research field linked to multiple areas like artificial intelligence,
machine translation, information retrieval etc. NER is an information extraction (IE) process
concerned with extracting named entities (NE) from raw data and categorising these NEs into
some predefined classes. The process of recognising and extracting NEs from unstructured
corpus or data is an essential task for solving different complications in various research fields
such as, question answering, summarisation system, information extraction, machine learning,
semantic web search, bio-informatics, video annotation and many more. For this research work a
classical methodology, i.e., rule-based approach is used to construct a system for automatic
identification of NEs from tourism domain data. The system considers words with their repetition
and without their repetition and acquires 83% and 71% accuracy respectively.
Keywords: named entity recognition; NER; named entity extraction; information retrieval;
machine translation; natural language processing; NLP.
Reference to this paper should be made as follows: Anandika, A., Chakravarty, S. and
Paikaray, B.K. (2023) ‘Named entity recognition in Odia language: a rule-based approach’,
Int. J. Reasoning-based Intelligent Systems, Vol. 15, No. 1, pp.15–21.
Biographical notes: Amrita Anandika is currently pursuing her PhD in the Department of
Computer Science and Engineering, Centurion University of Technology and Management,
Odisha, India. Her areas of research include natural language processing and machine learning.
Sujata Chakravarty is currently working as a Professor in the Department of CSE, Centurion
University of Technology and Management, Bhubaneswar, India. Her research areas include
information security, computational intelligence and evolutionary computing, bio-medical data
classification, smart agriculture, natural language processing, intrusion detection system in
computer-network, and analysis and prediction of different financial time series data.
Bijay Kumar Paikaray is currently pursuing his PhD in the Department of Computer Science and
Engineering, Centurion University of Technology and Management, Odisha, India. His areas of
research include high-performance computing, information security and IoT.
1 Introduction
The term NE is currently used in various IE applications. In
1996, it was introduced by Grishman and Sundheim in the
Sixth Message Understanding Conference (MUC-6).
Named entity recognition (NER) can be defined as a process
of information extraction that is concerned with recognising
and classifying NEs in a given sentence or text. It is a
subpart of NLP (Mansouri et al., 2008; Anandika and
Mishra, 2019). Natural language can be defined as the
communication approach between humans specifically,
speech and text. NLP is the method of automatic
manipulation of human spoken language, i.e., natural
language by software.
NER is being an essential task in the field of NLP. It is a
part of information extraction process which handles
structured as well as unstructured text and identifies the
terminologies that indicates name of people, places,
organisations and companies. Generally NER has two steps.
First step identifies the appropriate names from the given
text, and the second step, categorises these names into set of
groups or classes such as person names, organisations like
committees, companies, government organisations,
locations such as cities, countries, rivers, date and time
expressions, etc. (Dhariya et al., 2017). These groups are
previously defined. NE hierarchy is classified into three
types where the first one is entity (ENAMEX). It consists of

16 A. Anandika et al.
person, organisation and location name. Second one is time
expression (TIMEX). It consists of date and time. Third one
is numeric expression (NUMEX). It consists of money,
percentage values.
Different types of methodologies exist in research
literature for NER extraction, such as: terminology-driven
NER, rule-based NER, machine learning-based NER,
hybrid NER. Presently, machine-learning (ML) approaches
are widely used for NER due to the nature of easily
trainable and adoptable to other domains and languages.
Moreover, in ML approaches maintenance of trained data is
less expensive. However such methodologies demand rich
corpuses which are not available as on date for many
languages where very poor research has been done like Odia
language. Hence, researchers need to look for more feasible
alternative like rule-based method.
In the following manner rest of the paper is arranged.
Section 2 presents literature review. In Section 3 NER in
Odia language is discussed. Section 4 represents different
challenges associated with Indian languages. Section 5
gives brief description of the proposed system. In Section 6,
experimental result of the proposed system is given and
finally, Section 7 concludes the research work and focuses
on future directions.
2 Literature review
Biswas et al. in 2009 have designed a hybrid model by
using maximum entropy model (MEM) and hidden Markov
model (HMM) along with some linguistic rules to identify
NEs (NE) in Odia language. Firstly, they have used MEM
to identify NEs in Odia corpus, and then they tagged them
temporarily as reference. After this to train the HMM they
have used the tagged corpus of MEM as training process.
This method can reach to high precision and high recall, if it
is being supplied with sufficient data for training and a
proper error correction mechanism. They have used some
grammatical features such as orthography feature, word
suffix and word prefix, part-of-speech feature,
morphological information and information regarding
surrounding words as well as their corresponding tags for
developing an Odia NER system based on MEM and HMM.
Gazetteers are used to identify the designation and title of a
person name. They have also developed some linguistic
rules in Odia language to identify time, number etc.
Abdallah et al. in 2012 have developed a NER system
for Arabic language (NERA) by integrating a machine
learning classifier with a rule-based system which is
previously developed. This system is a combination of
rule-based NERA system, and ML classification system. In
this system, two sets of features are extracted for each word
from unstructured text where the first set is based on
rule-based features and it contains the name entity tags that
are identified by rule-based approach by considering the
word in question and window of surrounding words and the
features of second set are general features which are based
on the experience of a developer. A parser, gazetteer and
filtration mechanism is required by this system. This
recognition system consists of following two steps: lookup
procedure also called as whitelist/gazetteer which contain
lists of known NEs. It performs the identification based on
the lists of NEs. A parser, that contains sets of different
grammatical rules. These rules are expressed as regular
expressions which are derived from local lexical content
analysis. These whitelists are fixed static dictionaries of
NEs which are matched with the target text. The words of
target text that exactly match with the whitelist entries are
NEs.
Petasis et al. in 2012 have presented a system which
supports both rule-based NER approach and classification
approach. This system follows an inventive approach for the
use of learning in NERC. This system does not use ML in
the construction of NERC system, rather it is used
autonomously. The system that is created by using ML is
used to monitor the performance of existing rule-based
NERC system. Feedback regarding the under control
rule-based approach like, whether the rule-based approach
is out of date and it requires an update is provided by the
new system. This system does not need any manual tagging
for training data which is the main advantage. An iterative
process is used in this approach. First to train the classifier
few labelled examples are supplied, then to test the
classifier some unlabeled examples are given. They have
implemented this approach in Greek and French language.
Day et al. in 2014 have developed a semi-hybrid
approach for the NER task. In development of such system
they have focused on development of stemming tool, POS
tagging tool and a NER detection tool by combining HMM
with look-up algorithm and rule-based approach. In the
proposed system most of the work is done by HMM (Dey
et al., 2014). The look-up algorithm and some rules are used
for handling ambiguous words. This system can also be
used for solving word sense disambiguation problem.
Mathur and Saxena in 2014 have developed a system
which transliterates NEs from English language to Hindi
language. The system contains two modules. To transliterate
the NEs, both the modules use phoneme-based approach. A
CMU pronouncing dictionary that has a collection of
133,270 words along with their respective pronunciation is
used by the module.1. Suppose a word which is to be
transliterated and it is not found in CMU pronouncing
dictionary then, module.2 will be used. Module.2 uses
5-gram approach where maximum of five letters are used
for generating transliterated target letter.
Bajwa and Kaur in 2015 developed a combination of
rule-based and supervised learning (HMM)-based approach.
Two different interpretations are used in this paper. First
interpretation is HMM-based and second one is based on the
combination of rule-based and HMM. But the disadvantage
with this approach is that proper nouns are not automatically
tagged which are directed to the generation of training and
testing dataset as no dataset is accessible.
Ahmad and Satyaraj in 2015 have implemented MEM
for retrieval of NEs from the database. To train the system
they have used gazetteer list so that the system will retrieve
those words that have the maximum entropy amongst all

Named entity recognition in Odia language: a rule-based approach 17
others. It is also proved to be the fastest method in
retrieving and classifying the entity sets from database.
Advantage of this method is, it has increased the freedom of
choosing features to represent observations and sequence
tagging. It is also observed that when the MEM is used to
retrieve the information from the gazetteer list it gives better
result.
Chopra et al. in 2016 developed a HMM for the NER in
Hindi language. Advantage of this approach is that,
difficulty of managing the list of open class words while
adopting the gazetteers is handled effectively. But this
methodology cannot handle unknown names in a precised
way.
Boros et al. in 2017 have developed an optimised
decision tree computation algorithm that follows the
guidelines of ID3 algorithm. This algorithm calculates
entropy and information gain by using single pass over to
the training data. They have also developed a tree-pruning
algorithm to solve the issue of over fitting of the training
set. They implemented a result caching method in order to
increase the speed of the system. They have seen that the
tree-pruning achieves satisfactory accuracy.
Wang et al. in 2008 have designed and implemented
classifiers ensemble approaches for biomedical NER by
using four different learning algorithms. They are
generalised winnow, conditional random field, support
vector machine and maximum entropy. They also compared
the performances of three different classifiers ensemble
strategies, i.e., arbitration rules, stacked generalisation and
cascade generalisation. The stacked generalisation involves
class-stacking and class-attribute-stacking. They have also
discovered different features for biomedical NER like local
features, full text features and external resource features.
3 Named entity recognition in Odia
Odia is known as the official language of Odisha state and
the first language of more than 35 million of people. It is
also used as second and third language by many people of
India. The internal linguistic composition of state Odisha
has many tribal groups that linguistically belong to two
different individual language family that are Dravidian and
Munda (Swain and Pati, 2013). Through borrowing in Odia
the NEs in these languages are nativised.
Odia language is recognised as a classical language
because it is rich in literature and also has a history of more
than thousand and twenty years. Odia script consists of
64 letters (14 vowels, 50 consonants) and 10 digits. In Odia
language there is no upper case or lowercase letters. Odia
language is a free word order language meaning that in Odia
one single sentence can be written in different ways
(Balabantaray et al., 2013), e.g.:
• ଭୁ ବେନର ଓଡିଶାର ରାଜଧାନୀ (Bhubaneswar Odishara
rajadhanee)
This sentence can also be written as,
• ଓଡିଶାର ରାଜଧାନୀ ଭୁ ବେନର (Odishara rajadhanee
Bhubaneswar)
One major problem in recognising NE in Odia is that in this
language no morphological and punctuation marks are
present which can help in identifying the NEs. It can be
concluded that, in Odia NEs are based on the semantic
features of this language. NEs of Odia language have
similarity with the NEs of other languages to some extent
and it is also different from NEs of other languages
(Abdallah et al., 2012; Biswas et al., 2010). It is different
according to the area that is associated with tourism, history
and culture of the state, geographical diversity.
Although lots of work has been done in NER for many
Indian as well as non-Indian languages but, a very little
work has been done for Odia language. Hence, enough
resources and appropriate corpus are not available for Odia
language. In this study a rule-based model has been adopted
for identifying NEs from any document. After identification
of NEs it has been categorised into person, location and
organisation names as per rules. To implement this system
an Odia corpus from tourism domain has been adopted.
Hence, detection of NEs from a less annotated corpus is the
main challenge.
4 Challenges for Indian languages
For many European languages NER system works correctly
especially for English language, whereas, many difficulties
are still there for Indian languages (Anandika and Mishra,
2019; Shah et al., 2016). Some of the difficulties are
discussed here:
• No capitalisation: in English language capitalisation
plays vital role for recognising proper nouns. But, in
Indian languages capitalisation cannot be applicable.
• Morphologically rich: Indian languages are
morphologically very rich that’s why it is too difficult
to identify the root word, e.g., in Hindi, words Pyasa
and Pyasi both represents same meaning that is Thristy.
But the original word is Pyas (Thrist).
• Ambiguity between common noun and proper noun: for
Indian languages the most common issue is ambiguity
between common noun and proper noun because most
of the names of the people are dictionary words (like
Gagan, Aakash) and without capitalisation unlike
western names.
a Company vs. fruit: some name entity which
represents name of an organisation can be used as
name of a fruit, e.g., Apple.
b Person vs. month: sometimes a month name is
used as name of any person, e.g., June.
c Date vs. time: in some cases date expression
represents both date and time expression.

18 A. Anandika et al.
d Person vs. location: some name entity which
indicates a location name can also be used as any
person’s name, e.g., Let us consider the common
word Rose. It means Rose is a flower but at the
same time it can also be name of a person which
creates ambiguity between common noun and
proper noun.
• Ambiguity in suffixes: Indian languages may have a
number of post positions attached to the root word to
form a single word. For example: Manipur refers to
name of a location but Manipuree means anyone who
lives in Manipur. When ‘ree’ is added to the word
meaning of the root word gets changed.
• Standardisation deficiency and spell variations: one of
the biggest problems in many Indian languages is the
spelling variation, i.e., same word is spelled differently
by different people, e.g., a word ‘you’ in Hindi it is
spelled as ‘aap’, ‘tum’, ‘tu’. In Odia it is spelled
‘apana’, ‘tume’. In Bengali, it is spelled as ‘apani’,
‘tumi’. In Assamese, it is spelled as ‘aponar’, ‘tomar’.
In Telugu it is spelled as ‘miru’.
• Less resources and less labelled data: to work in NER
for any Indian languages a person faces many
difficulties like, very less availability of resources as
well as a proper annotated data because very less
amount of work has done in NER for different Indian
languages. The tools required for preprocessing the data
such as chunking and speech tagging gives very poor
performance (Chopra et al., 2016). In Indian languages
it is too difficult to find large amount of corpus or
training data. Basic resources like good morphological
analyser, part of speech (POS) tagger, name lists are
not available for many Indian languages or are at
research stage. Whereas a large amount of resources are
available in English.
• Agglutinative nature: when some additional features
are added to a word in order to make its meaning more
complex then it is called as agglutinative nature which
is very common for Indian languages, e.g., in
Assamese, Guwahati refers to a name entity of type
location but Guwahatiya is not a name entity. It refers
anyone who stays in Guwahati. Hence, the name entity
can be found which may have appeared as compound
word or with any suffix. In this case the root word of
the name entity has to be identified.
• Foreign words: some NEs are often language specific
like State Bank of India. When such entity is used in
any other language text, either it has to be translated to
that language or transliterated. Transliteration is
defined as a process of writing source language
expression in target language characters based on
phonetic similarity, e.g., ‘Aapnaar naam kee?’ (What is
your name?) And ‘Morbhal’ (I am fine) are the
transliterations of Bengali and Assamese language in
Roman. NE phrases are most difficult to translate
because they are domain specific and not found in
bilingual dictionaries. Translation of some NEs does
not make any sense like All India Radio, Air India etc.
Hence transliteration of NEs is more important even
when they can be translated.
Moreover Odia language is highly inflectional and
morphologically rich in nature. Odia language has relatively
free word order meaning that a name entity can appear for
both subject and object positions. At the same time there is
no subject-verb agreement and due to free word order,
named entities can appear at any position. In English
language, web sources for name lists are available. But there
are no such lists available for Odia language (Swain and
Pati, 2013). Lack of labelled data. Larger gazetteers are not
available in Odia language. To overcome above stated
problems associated with Odia language a rule-based
system is proposed for NER.
5 Proposed rule-based approach
Initially, NER systems were constructed on the basis of
some hand-crafted rules. Basically, the rules that are created
by humans form the background of a rule-based NER
system. A rule-based approach is a method which uses
human made rules for storing, sorting and manipulating
data. It is a classical approach of NER used by many
systems (Dhariya et al., 2017). A rule-based system needs a
set of information or data source along with a set of rules to
manipulate that data. Most of the time, these rules are
referred to be as if statements, i.e., if A happens then do B.
Generally, it focuses on retrieving names on the basis of
some rules (Wakao et al., 1996). Such as: grammatical (part
or speech), syntactic (word precedence), orthographic
feature (capitalisation), dictionary. These rules are manually
written by a linguistic expert. Normally, name identification
system executes the text in three different phases (Gali
et al., 2008). They are: recognising phrases, recognising
patterns, and merging.
5.1 Reasons behind using rule-based system
In rule-based system, the main advantage is, it does not
require any training data. Precision is high in rule-based
system. The system is cost efficient. This system provides
better accuracy. The system can be available to the user
easily. The system operates at high speed. Due to predefined
rules, error rate is very less in rule-based system. Hence, the
system possesses high accuracy. There is reduced amount of
risk in terms of accuracy. Output of the system is stable as
they are generated according to the predefined rules. Hence
the output cannot be indeterminate. The rule-based system
gives result in the same way as a human does.
In this work a tagged Odia corpus is used which is
collected from tourism sector. The corpus is based on
description of different tourist places, present in different
states of India. The dataset consists of 1,000 lines. Each line
starts with a line header and each word in the lines is
attached with its respective tag by the symbol ‘\’. Tag

Named entity recognition in Odia language: a rule-based approach 19
defines how a particular word is used in the sentence, i.e.,
either it is used as noun or proper noun or verb etc. The
below example describes about the structure of a line from
the dataset.
• E.g., htd9001 େମରିନ\N_NNP ରାୀୟ\JJ ଉଦାନ\N_NN
ଆାମାନ\N_NNP ଜିାେର\N_NN ଅବିତ\JJ
\RD_PUNC
5.2 Features considered in the proposed model
• Context word feature: the term context represents the
immediate linguistic environment of a word. However
it is not always explicit. Sometime it may be hidden
within the neighbouring words of the keyword that are
used in the same piece of text. Two types of context are
there. Such as: local context and topical context. Local
context represents the surrounding members of the key
word, i.e., one or two immediately before and after
words of the key word. Topical context represents the
topic of the text where the key word is utilised. We
have considered local context in our research work. For
example: If a key word has Shrijukta, Shriman,
Shrimati, Kumari as its previous word then the key
word will be a NE of person name. Similarly if the key
word has Sahara, Nagara, Grama as it is after word then
it will be a location name.
• Word suffix: word suffix refers to the end letters of a
word up to a fixed length, i.e., the last two or last three
letters of the word. The feature of a fixed length word
suffix can be applied to the current word or to any
surrounding word. This feature is considered as one of
the most powerful and helpful approach for identifying
a NE. For example: If a key word has Natha, Kanta as
suffix then that word is a NE of person type. If a key
word has Gada, Pur as suffix then that word is a NE of
location type.
• Word prefix: prefix information of a word is also
needed for NE detection. Prefix refers to the starting
letter of a key word up to some fixed length like
starting two or three letters of the key word. A fixed
length word prefix feature can be applied to the current
word or to the surrounding words of the keyword.
• Part of speech (POS) tag: the POS tag of a key word or
its surrounding words can also be considered as a
useful feature for identifying NE.
E.g., in most of the cases a word which is tagged as NNP is
a named entity apart from cases like:
• େମୗଲି ରାୀୟ ଉଦାନ ଅରୁଣାଳେଦଶର ପୁବସିୟା
ଜିାେର ଅବିତ 
In the above sentence େମୗଲି is tagged as NNP but it is
not a NE. The proposed rule-based NER system for Odia
language is implemented in this work by using python. The
platform used for the implementation of the language is
Natural Language Tool Kit (NLTK). NLTK is an open
source library or platform for developing Python programs
to work with human language data for applying NLP.
Figure 1 shows the flow diagram of the proposed system.
Figure 1 shows the flowchart for the proposed system.
In the first step of the flow chart, raw text data is given as
input. Here the input raw text data is a tagged Odia dataset.
Next step represents data pre-processing where line
segmentation is done first followed by word segmentation.
After the word segmentation process, noun phrase
extraction has been carried out. In the next step a single
word is considered at a time to check whether the word has
an N-NNP tag or not. N-NNP tag represents that the
associated word is used as a proper noun. Generally proper
nouns are name specific like people, places, things etc. If
the word does not have an N-NNP tag then, go to the next
word and again check for the N-NNP tag. If the word has
N-NNP tag then implement the proposed rule-based
algorithm, which results in identifying whether the word is a
named entity or not.
Figure 1 Flow diagram of rule-based approach for NER
Start
Input raw text
Data preprocessing
Line segmentation
Word segmentation
Noun phrase extraction
Implementation of
proposed R-B algorithm
Go to next
word
Non-NE
NE Stop
Yes
Yes No
No
If the word has
N-NNP tag
If the rule
satisfies
6 Experimental result
The proposed rule-based approach has been applied on a
dataset which is based on tourism and it consists of
1,000 lines. In order to calculate the performance of the
proposed system, confusion matrix has been considered.
• true positive (TP): observation is true and predicted as
true
• true negative (TN): observation is false and predicted as
false.
• false positive (FP): observation is false and predicted as
true.
• false negative (FN): observation is true and predicted as
false.

20 A. Anandika et al.
• class 0: non-N-NNP tagged word
• class 1: N-NNP tagged word.
Accuracy of the proposed system is calculated by
considering the words present in the dataset with their
repetition and without their repetition (Kaur and Gupta,
2010). Hence two different confusion matrices and two
different accuracies are obtained (Guo et al., 2019; Parai
et al., 2009; Das et al., 2020). Performance of the proposed
system is calculated in terms of precision, recall, F1-score
and accuracy. Equations for calculation of these measures
are given below:
• Precision: precision is calculated as the ratio of total
number of correctly classified true values to the total
number of predicted true values. High precision points
to low FP. The formula for precision is given in
equation (1).
TP
Precision TP FP
=
+
(1)
• Recall: recall is calculated by dividing the total number
accurately classified true values with the total number
of true values. High recall points to low FN.
Equation (2) shows the formula for recall.
TP
Recall TP FN
=
+
(2)
• F1-score: F1-score is calculated by using Harmonic
mean in place of Arithmetic mean. The formula for
F1-score is shown in equation (3).
2
1- Recall Precision
Fscore Recall Precsion
∗∗
=
+
(3)
• Accuracy: below equation represents the formula for
calculating accuracy.
TP TN
Accuracy TP TN FP FN
+
=
+++
(4)
As discussed in Table 3, the proposed system got an
accuracy of 83% with word repetition and 71% of accuracy
without word repetition. Word repetition represents
considering multiple occurrence of same word in different
places in the dataset. Without word repetition represents
considering only one occurrence of any word in the whole
dataset. From this research work it can be concluded that, if
only one occurrence of any word is considered then the
proposed system gives an average performance. Whereas if
any word considered with its multiple occurrences (as many
times it is being used in different places in the dataset) then,
the proposed system gives better performance. Hence, word
repetition can be used as a key to enhance the performance
of any rule-based system.
Table 1 Confusion matrix with word repetition
n = 942 Predicted false Predicted true
Actual false TN = 1,526 FP = 210
Actual true FN = 169 TP = 303
Table 2 Confusion matrix without word repetition
n = 942 Predicted false Predicted true
Actual false TN = 625 FP = 210
Actual true FN = 169 TP = 303
Table 3 Performance measures with word repetition and
without word repetition
Performance measures with word repetition
Precision Recall F1-score Accuracy
0.59 0.64 0.61 0.83
Performance measures without word repetition
Precision Recall F1-score Accuracy
0.59 0.64 0.61 0.71
7 Conclusions and future directions
In this study, a rule-based NER system for Odia language
has been presented. The proposed system is implemented on
tourism dataset and got accuracy of 83% and 71% by
considering words with their repetition and without their
repetition respectively. In this system, different features like
context word feature, word suffix, word prefix and POS tag
are considered. Precision, recall and F1-score have been
considered to measure the efficacy of the system. Although
rule-based approach gives promising results, but it has one
major challenge. It is highly domain dependant. If the
research domain is changed then changes has to be made in
the system structure.
In future research, more features can be added to
rule-based system in order to enhance the system
performance. This system can also be implemented in other
domain as well as in other languages that are similar to Odia
language like Bengali and Asami. Moreover, different high
level rules or mechanism can also be incorporated in order
to reduce the ambiguity between the named entities present
in Odia language. A hybrid model can also be developed by
combining the proposed rule-based system with any
machine learning approaches.

Named entity recognition in Odia language: a rule-based approach 21
References
Abdallah, S., Shaalan, K. and Shoaib, M. (2012) ‘Integrating
rule-based system with classification for Arabic named entity
recognition’, International Conference on Intelligent Text
Processing and Computational Linguistics, Springer,
pp.311–322.
Ahmed, I. and Sathyaraj, R. (2015) ‘Named entity recognition by
using maximum entropy’, International Journal of Database
Theory and Application, Vol. 8, No. 2, pp.43–50.
Anandika, A. and Mishra, S.P. (2019) ‘A study on machine
learning approaches for named entity recognition’,
International Conference on Applied Machine Learning
(ICAML), pp.153–159.
Bajwa, K.S. and Kaur, A. (2015) ‘Hybrid approach for named
entity recognition’, International Journal of Computer
Applications, Vol. 118, No. 1, p.3641.
Balabantaray, R.C., Lenka, S.K. and Sahoo, D. (2013) ‘Name
entity recognizer for Odia using conditional random fields’,
Indian Journal of Science and Technology, Vol. 6, No. 4,
pp.4290–4293.
Biswas, S., Mishra, S.P., Acharya, S. and Mohanty, S. (2010)
‘A hybrid Oriya name entity recognizer: harnessing the power
of rule’, International Journal of Artificial Intelligence
and Expert Systems (IJEA), Vol. 1, No. 1, pp.1–6,
ISSN: 2180-1282.
Biswas, S., Mohanty, S. and Mishra, S.P. (2009) ‘A hybrid Oriya
named entity recognition system: integrating HMM with
MaxEnt’, Second International Conference on Emerging
Trends in Engineering & Technology, IEEE, pp.639–643.
Boros, T., Dumitrescu, S.D. and Pipa, S. (2017) ‘Fast and accurate
decision trees for natural language processing tasks’,
Proceedings of Recent Advances in Natural Language
Processing, pp.103–110.
Chopra, D., Joshi, N. and Mathur, I. (2016) ‘Named entity
recognition in Hindi using hidden Markov model’, IEEE
Second International Conference on Computational
Intelligence Communication Technology, pp.581–586.
Chopra, D., Joshi, N. and Mathur, I. (2016) ‘Named entity
recognition in Hindi using hidden Markov model’, IEEE
Second International Conference on Computational
Intelligence Communication Technology, pp.581–586.
Das, D., Singh, M., Mohanty, S.S. and Chakravarty, S. (2020)
‘Leaf disease detection using support vector machine’, in
2020 International Conference on Communication and Signal
Processing (ICCSP), IEEE, July, pp.1036–1040.
Dey, A., Paul, A. and Purkayastha, B. (2014) ‘Named entity
recognition for Nepali language: a semi hybrid approach’,
International Journal of Engineering and Innovative
Technology, Vol. 3, No. 8, pp.21–25.
Dhariya, O., Malviya, S. and Tiwary, U.S. (2017) ‘A hybrid
approach for Hindi- English machine translation’,
International Conference on Information Networking
(ICOIN), IEEE, pp.389–399.
Gali, K., Sharma, H., Vaidya, A., Shisthla, P. and Sharma, D.M.
(2008) ‘Aggregating machine learning and rule-based
heuristics for named entity recognition’, IJCNLP-08.
Workshop on NER for South and South East Asian
Languages, pp.25–32.
Guo, J., Han, Y. and Ke, Y. (2019) ‘A neural-based re-ranking
model for Chinese named entity recognition’, International
Journal of Reasoning-based Intelligent Systems (IJRIS),
Vol. 11, No. 3, pp.265–272.
Kaur, D. and Gupta, V. (2010) ‘A survey of named entity
recognition in English and other Indian languages’,
International Journal of Computer Science Issues (IJCSI),
Vol. 7, No. 6, pp.239–245.
Mansouri, A., Affendey, L.S. and Mamat, A. (2008) ‘Named entity
recognition approaches’, International Journal of Computer
Science and Network Security, Vol. 8, No. 2, pp.339–344.
Mathur, S. and Saxena, V.P. (2014) ‘Hybrid approach to
English-Hindi name entity Transliteration’, Proceedings of
IEEE Students Conference on Electrical, Electronics and
Computer Science, pp.1–5.
Parai, G.K., Tenneti, T., Borah, P.K., Shah, S. and Sanyal, S.
(2009) ‘Document summarisation using combination and
reduction of extracted sentences’, International Journal of
Reasoning-based Intelligent Systems, Vol. 1, Nos. 3–4,
pp.191–199.
Petasis, G., Vichot, F., Wolinski, F., Paliouras, G., Karkaletsis, V.
and Spyropoulos, C.D. (2012) ‘Using machine learning to
maintain rule-based named-entity recognition and
classification systems’, Proceedings of the 39th Annual
Meeting of the Association for Computational Linguistics,
pp.426–433.
Shah, H., Bhandari, P., Mistry, K., Thakor, S., Patel, M. and
Ahir, K. (2016) ‘Study of named entity recognition on indian
languages’, International Journal of Information Science and
Techniques (IJIST), Vol. 6, Nos. 1–2, pp.11–25.
Swain, D. and Pati, C. (2013) ‘Named entity disambiguation in
Odia’, International Journal on Advanced Computer Theory
and Engineering (IJACTE), Vol. 2, No. 4, pp.137–143.
Wakao, T., Gaizauskas, R. and Wilks, Y. (1996) ‘Evaluation of an
algorithm for recognition and classification of proper names’,
Proceedings of COLING-96.
Wang, H., Zhao, T., Tan, H. and Zhang, S. (2008) ‘Biomedical
named entity recognition based on classifiers ensemble’,
International Journal of Computer Science and Applications,
Vol. 5, pp.1–11.