COMPARATIVE ANALYSIS OF ARABIC STEMMING ALGORITHMS

International Journal of Managing Information Technology (IJMIT) Vol.5, No.2, May 2013
DOI : 10.5121/ijmit.2013.5201 1
COMPARATIVE ANALYSIS OF ARABIC
STEMMING ALGORITHMS
Dr. Mohammed A. Otair
Department of Computer Information Systems, Amman Arab University, Jordan
Otair@aau.edu.jo
ABSTRACT
In the context of Information Retrieval, Arabic stemming algorithms have become a most research area of
information retrieval. Many researchers have developed algorithms to solve the problem of stemming.
Each researcher proposed his own methodology and measurements to test the performance and compute
the accuracy of his algorithm. Thus, nobody can make accurate comparisons between these algorithms.
Many generic conflation techniques and stemming algorithms are theoretically analyzed in this paper.
Then, the main Arabic language characteristics that are necessary to be mentioned before discussing
Arabic stemmers are summarized. The evaluation of the algorithms in this paper shows that Arabic
stemming algorithm is still one of the most information retrieval challenges. This paper aims to compare
the most of the commonly used light stemmers in terms of affixes lists, algorithms, main ideas, and
information retrieval performance. The results show that the light10 stemmer outperformed the other
stemmers. Finally, recommendations for future research regarding the development of a standard Arabic
stemmer were presented.
KEYWORDS
Information Retrieval, Arabic stemmers, Morphological analysis, and Computational Linguistics.
1. INTRODUCTION
Information Retrieval is ultimately an issue of determining which documents in a corpus should
be retrieved to satisfy a user's information need which is represented by a query, and contains
search term(s), in addition to some information such as the relatively importance. Thus, the
retrieval decision is possessed by finding the similarity between the query terms with the index
terms appearing in the document. The decision of the retrieval process may be taking any shape
of the following result: binary: relevant, non-relevant, or partial. In the last case, it may involve
rating the degree of important relevancy that the document has to the submitted query. [15].
Unfortunately, the words that appear in the documents and in queries at the same time often have
many morphological variations (Morphology means the internal structure of words). For instance,
some terms such as "extracting" and "extraction" will not be recognized as similar or equivalent
without any kind of processing. Conflation techniques are needed to equate word variations
having similar semantic interpretations [15].
Arabic stemming is a technique that aims to find the stem or lexical root for words in Arabic
natural language, by eliminating affixes stuck to its root, because an Arabic word can have a more
complicated form than any other language with those affixes. Morphological variants of words
accept agnate semantic interpretations and can be advised as agnate for the purpose of advice

2
retrieval systems. Hence, a advanced ambit amount stemming Algorithms or stemmers accept
been developed to abate a chat to its axis or root. Many researches were conducted to compare
these algorithms. A study by Sawalha [23] gives a comparison between three stemming
algorithms: Khoja’s stemmer [17], Buckwalter’s morphological analyzer [7] and the Tri-literal
root extraction algorithm [5]. The Khoja stemmer obtains the accomplished accurateness
followed by the tri-literal basis abstraction algorithm, and assuredly the Buckwalter
morphological analyzer. Another study by Darwish [10] found that light stemming is one of the
most superior in morphological analysis. Similar study were done by Larkey [20] to compare
light stemming with several different stemming algorithms based on morphological analysis:
roots, stems, and some other criteria. Their results showed that the light stemmer passed the other
algorithms in terms of performance (precision and recall) [12].
As a summary, Arabic stemming algorithms can be classified, according to the desirable level of
analysis: root-based approach [17] and stem-based approach [20]. Root-Based approach uses
morphological analysis to find the root of a given Arabic word. Many algorithms have been
proposed for this approach [2, 6, 14]. The aim of the Stem-Based approach is to eliminate the
most frequent prefixes and suffixes [3, 8, 19, 20]. A lot of all those trials in this acreage were a set
of rules to abbreviate the set of suffixes and prefixes, as well there is no audible account of these
strippable affixes [24].
This paper is conducted to do a comparative analysis for the most of the existing light stemmers.
It compares stemmers in terms of the main ideas behind the development of the stemmers, the
prefixes and suffixes that can remove, and as well as the affixes. The stemmers also compared in
terms of their information retrieval performance; precision and recall. A lot of accepted and
acknowledged address acclimated for bearing stems of words is the ablaze stemming techniques.
This paper is going to compare the stemmers in terms of:
• The main idea behind the stemmer built,
• The prefixes and suffixes they remove, and
• The basis of choosing the affixes
• The algorithm they use to remove the affixes.
• Information retrieval performance; precision and recall.
• Limitation of the stemmer
2. LITERATURE REVIEW
Literature search showed the existence of many Arabic language stemmers. The strengths of
those stemmers are varied and stemming errors are produced for every stemmer. None of the
analyzed stemmers showed perfect performance, and none of them has been adopted as a
standard Arabic stemmer that fulfills the user’s information needs.
Arabic stemming approaches have been analyzed, weaknesses, and strengths have been pointed
out. Among the studied stemmers, there have been aggressive stemmers and weak stemmers.
The Larkey [21] has showed performance effectiveness for a group of stemmers that included
light 1, 2, 3, 8, and 10 as compared to normalization and raw or surface based. The result showed
that Light 10 is the best. In a previous research [20], Larkey showed the results of a comparison
of basic stemmers that included Light 8,3,2,1, Khoja, Khoja-u, normalized, and raw. The Light 8
showed the best performance results.
James [4] and Hull [16] both showed separately that Stemming performs better than no
stemming. In addition, Darwish [9] has experimented with Alstem, Umass, and Modified Umass
and his results showed that Alstem as the most effective.

3
Aljlay [3] has performed experiments on surface-based, root-based, and light stemmer. He arrived
to the point that stemming via a root-based algorithm creates invalid conflation classes. He also
showed that light stemming approach outperforms the root-based stemmer.
Mayfield et al. [20] have developed a system that merge surface-based and 6-gram based retrieval
which performed remarkably well for Arabic.
Which stemmer is the best? Which one should we use? And what is the standard Arabic stemmer
to adopt? Many questions need to be answered regarding Arabic stemmers. Hence, in the
following sections some elaborations on conflation techniques and the stemming theory,
performance, strength, errors, and advantages will be noted.
3. CONFLATION TECHNIQUES
Conflation is an accepted appellation for all processes of amalgamation calm none identical
words which accredit to the aforementioned arch concept [11]. In this context, conflation refers to
morphological conflation and morphological query expansion, where the related word forms
resemble each other as character strings. Conflation algorithms can be disconnected into two
capital classes as Leah S. Larkey [21] states:
1. Affix removal algorithms (stemming algorithms): which are language dependent; and
designed to handle morphological word variants.
2. Statistical techniques: which are (usually) language independent; and designed to
manipulate all types of variants involving n-grams, string similarity, morphological
analysis, and co-occurrence analysis.
The domain of morphology can be classified into two subclasses, derivational and inflectional
[18]. Derivational morphology could or could not impact meaning of a word. Although English
language has a comparatively weak morphological, other languages such as Arabic have stronger
morphology (for example, a lot number of variants maybe given for a word word). In analyze
with the changes of Inflectional assay which describes accepted changes a chat undergoes as an
after effect of syntax (the plural and control anatomy for nouns, and the accomplished close and
accelerating anatomy for verbs are a lot of accepted in English) [18]. There is no effect on a
word’s ‘part-of-speech’ for these changes (a noun still remains a noun after pluralisation) [3].
4. STEMMING
Stemming is a very essential technique for processing strong morphological languages such as
Arabic. Thus, the definition of stemming and related issues is required to create the necessary
basic knowledge before proceeding further with Arabic stemmers.
Stemming is mainly affected by means of suffix lists which are lists of possible word
terminations, and this way were successfully applied on many different languages. However, it is
less applicable to complexly morphological language such as Arabic, which require a further
effort of morphological analysis, where absolutely morphological techniques are required that
eliminate suffixes from words according to their internal structure.
Thus, conflation is concerned with attempting to 'reverse' the inflexion process by performing the
inverse operation related to the basic inflexion rules [15]. Conflating English words faces several
problems due to the existence of strong verbs that follow no set pattern for inflexion[15], e.g.
throw, threw, thrown, and irregular verbs, e.g. go, went, and gone. The use of a lexicon
(dictionary) is a must to avoid errors.

4
A. Stemming Advantages and Stemming Errors
Stemming simplifies the searchers’ job by making the IR system satisfy their information need. In
increase in recall is gained. However, conflation – basic word form generation with dictionary
lookup – can improve precision. Stemming reduces the size of index terms and thus reduces the
size of the index (inverted file), too. As the size of index terms is reduced, the storage space and
processing time are reduced as well.
By the use of stemmers, Words in the collection must be organized into groups, multiple errors
are produced and may be used to compare and evaluate stemmers.
• If the two words accord to the aforementioned semantic category, and are adapted to the
aforementioned stem, again the conflation is correct. If they are adapted to altered stems,
this is an understemming absurdity [15] (in added words, too abundant of a appellation is
removed).
• If the two words accord to altered category, and abide audible afterwards stemming,
again the stemmer has proceeded correctly. If they are adapted to the aforementioned
stem, this is advised as an over-stemming [15] absurdity in added words, too little of a
appellation is removed).
B. Stemmer Performance and Strength measurement
When stemming algorithms are used, the effectiveness of the retrieval system is enhanced if the
size of the retrieval set is taken into account as Hull (1996) noted [16]. There are many measures
to evaluate stemmer effectiveness and performance such as: Direct assessment, Precision and
Recall, and counting both Stemming Errors.
When the stemmer merges a few of the most highly related words together, it is called a ‘weak’
or ‘light’ stemmer. A 'strong' or 'heavy' stemmer combines a much wider variety of forms. The set
of metrics that measure stemmer strength as follows [13]:
• Number of words per conflation class
• Index Compression: the ad-measurements to which a accumulating of altered words is
bargain (compressed) by stemming.
• The Word Change Factor: This is an artlessly the ad-measurements of the words in a
sample that accept been afflicted in any way by the stemming process.
• Mean Number of Characters Removed
• Hamming Distance: The Hamming Distance takes two strings of according breadth
and counts the amount of agnate positions area the characters are different. If the
strings are of altered lengths, we can use the Modified Hamming Distance.
That was stemming theory and stemmers’ characteristics. In the next section, the generic
stemming algorithms for English will be analyzed in order to check if they may fit for Arabic or
not.
5. GENERIC STEMMING ALGORITHMS
Stemming algorithms are numerous; in this section, a review of the generic stemming algorithms
will be summarized as stated in [15] which they were developed mainly to English language and
not to Arabic. However, they are summarized here for the purpose of proving that those stemmers
are not fit for Arabic and should not be considered in the final comparative analysis.

5
A. Lovins Stemmer
The Lovins Stemmer [15] is a single pass, context-sensitive, and longest-match Stemmer
developed in 1968. The approach is not complex enough to stem many. Lovins' aphorism account
was acquired by processing and belief a chat sample. The capital botheration with this action is
that it has been begin to be awful capricious and frequently fails to anatomy words from the
stems, or matches the stems of like acceptation words.
The Lovins Stemmer removes a best of one suffix from a word, consistent to its attributes as
individual canyon algorithm. Lovins Stemmer uses a almost abbreviate account of about 250
altered suffixes, and eliminates the longest suffix affiliated to the word, ensuring that the axis
afterwards the suffix has been removed is consistently at atomic 3 characters long. Then the
terminating of the stem may be reformed (e.g., by un-doubling a final consonant if applicable), by
indicating to a list of recoding transformations.
B. Dawson Stemmer
The Dawson developed the Stemmer in 1974; it is strongly based upon the Lovins Stemmer,
prolongation the suffix rule list to about 1200 suffixes. It acquire the longest bout and individual
canyon attributes of Lovins, and exchanges the recoding rules, which were begin to be wildcat,
application instead a constancy of the fractional analogous action as well authentic aural the
Lovins Stemmer.
The agnate affair amid the Lovins and Dawson stemmers is that every catastrophe independent
aural the account is associated with an amount that is acclimated as a basis to seek an account of
exceptions that accomplish assertive altitude aloft the abatement of the associated ending.
The above aberration amid the Dawson and Lovins stemmers is the address acclimated to break
the botheration of spelling exclusions. The Lovins stemmer employs the technique known as
recoding. This action is advised as allotment of the capital algorithm and performs n amount of
transformations based on the belletrist aural the stem. In comparing with the Dawson stemmer
employs fractional analogous which attempts to bout stems that are according aural assertive
limits.
C. Paice/Husk Stemmer
The Paice/Husk Stemmer was developed in the backward 1980s; the Stemmer has been
implemented in Pascal, C, PERL and Java. When operating with its accepted rule-set, it is a rather
'strong' or 'heavy' stemmer. It is a simple accepted Stemmer; it removes the endings (suffixes)
from a chat in a broad amount of steps.
D. Porter Stemmer
The Porter stemmer was first presented in 1980. The stemmer is an ambience acute suffix
abatement algorithm. It is based on the abstraction that the suffixes in the English accent
(approximately 1200) are mostly fabricated up of an aggregate of abate and simpler suffixes. It is
a lot of broadly acclimated of all the stemmers and implementations in abounding languages are
available. The stemmer is divided into a number of linear steps, five or six; a linear step Stemmer.
Porter himself implemented the algorithm in Java, C and PERL. Porter developed an Improved
Porter stemmer as well.
E. Krovetz Stemmer
The Krovetz Stemmer [18] was developed in 1993 as a 'light' stemmer. The Krovetz Stemmer
finer and accurately removes inflectional suffixes in three steps:

6
1. The about-face of a plural to its individual anatomy (e.g. `-ies', `-es', `-s'), the about-face
of accomplished to present tense (e.g. ‘-ed’), and the removal of ‘-ing’.
2. The about-face action firstly removes the suffix, and again admitting a action of
analytical in a concordance for any recoding (also getting acquainted of exceptions to the
accustomed recoding rules), allotment the axis to a word. The concordance lookup as
well performs any transformations that are appropriate due to spelling barring and as well
converts any axis produced into an absolute chat that acceptation can be grasping.
3. Due to the high accuracy of the stemmer, but weak strength, it is used as a form of pre-
processing performed before the main stemming algorithm (such as the Paice/Husk or
Porter Stemmer). This would provide partly stemmed ascribe for the stemmer that deals
with accepted situations accurately and effectively, and accordingly could abate
stemming errors.
F.Truncate (n) Stemmer
This stemmer simply retains the first n letters of the word, where n is a appropriate integer,
such as 4, 5 or 6. If the chat has beneath than n belletrist to alpha with, it is alternate unchanged.
After truncation, words are compared to each other. If the retained parts are similar, they are
conflated to the same group, otherwise they are not. There are some problems with this approach;
manual construction of the grouped word collection is time-consuming and conflation
groups depend on topic of original text.
G. N-grams (String Similarity)
String-similarity approaches to conflation absorb the arrangement artful admeasurements of
affinity amid an ascribe concern appellation and anniversary of the audible agreement in the
database. Those database agreements that accept an animated affinity to a concern appellation are
again displayed to the user for accessible admittance in the query. N-gram matching technique is
one of the most common of these approaches [14]. An n-gram is a set of n successive characters
taken away from a word. The main idea of this approach is that, similar words will have an
elevated proportion of n-grams in common. Idealistic values for n are 2 or 3, these
corresponding to the use of diagrams or trigrams, respectively. For example, the word (computer)
results in the generation of n-grams as shown in table (1).
Table 1. Different Digrams and Trigrams for ‘computer’
Where ‘*’ denotes a padding space.
We approved to administer anniversary of the antecedent stemmers to Arabic but abominably
none of them seems to be acceptable candidate. Hence, it is required to elaborate further on the
Arabic language characteristics in order to understand and analyze the Arabic stemmers.

7
6. CHARACTERISTICS OF ARABIC LANGUAGE
Arabic accent is announced by over 300 actor people; as compared to English language, Arabic
characteristics are assorted in abounding aspects. In [22] Nizar summarized most of the Arabic
language characteristics.
A. Arabic script or alphabets
Arabic is accounting from appropriate to left, consists of 28 belletrist and can be continued to 90
by added shapes, marks, and vowels [3]. Arabic script and alphabets differ greatly when
compared to other language scripts and alphabets in the following areas:
(1) Shape, marks, diacritics, Tatweel (Kashida), Style (font), and numerals
(2) Distinctive letters ( ‫ب‬ ‫ت‬ ‫ث‬ ‫س‬ ‫ش‬ ), and none distinctive letters ( (‫اإأآىء‬ ‫)ؤ‬
B. Arabic Phonology and spelling
(1) 28 Consonants, 3 short vowels , 3 long vowels (‫ا‬ ‫و‬ ‫,)ي‬ and 2 diphthongs ( ‫علة‬ ‫حرفا‬ ‫إدغام‬
‫معا‬ ‫)متصالن‬
(2) Encoding is in CP1256 and Unicode
C. Morphology
(1) Consists from bare root verb form that is triliteral, quadriliteral, or pentaliteral.
(2) Derivational Morphology (lexeme = Root + Pattern)
(3) Inflectional morphology (word = Lexeme + Features); features are
(4) Noun specific: (conjunction, preposition, article, possession, plural, noun)
Number: collective, plural, dual, singular.
Gender: feminine, masculine, Neutral.
Definiteness: definite, indefinite.
Case: accusative, genitive, nominative.
Possessive clitic.
(5) Verb specific: (conjunction, tense, verb, subject, object)
Aspect: imperative, perfective, imperfective.
Voice: active, passive.
Tense: past, present, future.
Mood: jussive, indicative, subjunctive.
Subject: person, number, gender.
Object clitic.
(6) Others: Single letter conjunctions and single letter prepositions
D. Morphological ambiguity
Derivational ( ‫قاعدة‬ ) base or rule?, Inflectional ( ‫ﺗكتب‬ ) you write or she writes?, spelling
ambiguity due to missing diacritics and miss spelling ( ،‫ا‬ ،‫ة‬ ‫ي‬ ), and Combined ambiguity.
For the purpose of advice retrieval this affluence of forms, lexical variability, and orthographic air
headedness after effect in a greater likelihood of conflict amid the anatomy of a chat in a concern
and the forms begin in abstracts accordant to the query. Stemming is a tool that is used to combat
this vocabulary mismatch problem.
7. ARABIC STEMMERS
Variation in morphological properties among world's languages is high, and stemmers are
language dependent as stated earlier. Hence, it is expected to see distinct stemmers for the Arabic

8
language that are different form the English ones. The factors described in section (4) make
Arabic very difficult to stem. Abu-Salem [1] documented that stems or roots are useful index
terms for Arabic. Their benefits were clearer than for English, as the Arabic language is a root
based language.
The affair of whether roots or stems are the adapted akin of assay for IR has been one aggravation
that has accustomed acceleration to added approaches to stemming for Arabic accent besides
Affix abatement and Statistical Stemming approaches as declared in section(3). Added
approaches include manual dictionary construction, morphological analysis, and new statistical
methods involving alongside corpora [3].
However, this paper concentrates on the analysis of Arabic stemming algorithms only. In this
section, a summary of the major Arabic stemming techniques are analyzed and compared.
A. Affix Removal
(1) Normalization which functions as follows [20]:
• Converting to windows Arabic encoding (CP1256)
• Remove punctuations
• Remove diacritics
• Remove none letters
• Replace ‫أ،إ،آ‬ with ‫ا‬
• Replace ‫ى‬ with ‫ي‬
• Replace ‫ة‬ with ‫ه‬
This process is usually conducted as a pre-processing step before stemming. The main issue is to
know whether these steps are necessary or not are debatable. The authors strongly agree on
necessity of some of those steps but not all of them. Diacritics removal produces inflectional
ambiguity. The three replacements steps affect the spelling ambiguity. As a result, precision and
recall are lowered.
(2) Surface-based stemmers that comprise from at least two morphemes as stated in [3]:
• A three consonantal root conveying semantics
• A word pattern ‫الوزن‬ carrying syntactic information
Conflation is based on the surface words that exist in the user query and the corpora documents.
(3) Root-based stemmers the ultimate goal of a root-based stemmer is to extract the root of a
given surface word. The root extraction is performed after the suffixes and prefixes are removed.
The remaining stem is then compared against matching patterns of the same length to extract the
root as shown in table (2) by dropping similar associated letters [3]. Weaknesses of root-based
stemmers are:
• Increases word ambiguity
• All possible patterns are not included
• Weak double triliteral verbs conflation
• Weak Irregular triliteral verbs conflation
(4) Algorithmic Light Stemmers which remove a small set of prefixes and suffixes without
dealing with infixes or recognize patterns, and find roots as noted in [3] and [20]. Their
drawbacks are as follows:
• No absolute abundant lists of strippable prefixes and/ or suffixes or algorithm had been
published.
• It fails to conflate broken plurals for nouns
• Adjective generally does not provide conflated with its singular form

9
• Past tense do not get conflated with their present forms
Many versions exit for the light stemmer approach; all of which follow the same steps:
Remove ‫,و‬ remove any of the definite article from Appendix A, and remove suffixes from
Appendix B.
Among the light versions shown in Appendix A and B, it can be easily concluded that
Alstem is the strongest light stemmer and the Light 1 is weakest light stemmer.
(5) Simple Stemmers are Light stemmers and removals of infixed vowels ‫ا،و،ي،ء‬ from variant
patterns as Larkey noted in [20]. Many Simple stemmer versions do exist and are shown in the
table (3).
Affix removal algorithms are scaled from weak to strong stemmers. Mean average precision and
some other attributes for the major stemmers currently used are analyzed in Appendix C.
Investigation of the Appendices A, B, and C shows that those stemmers are very difficult to be
compared unless those measured are conducted on the same document collection.
B. Manually Constructed Dictionaries
They are manually congenital dictionaries of roots and stems for anniversary chat to be indexed.
C. Morphological Analyzers (Stemmers)
Which attempt to find roots or any number of possible roots for each word automatically by a
software program.
D. Statistical Methods involving Parallel Corpora
Statistical stemmers, which accumulation chat variants application absorption techniques and Co-
occurrence assay methods that are activated to automated morphological assay software systems.
However, those techniques cannot be expected to perform well on the Arabic language due its
strong morphology.
Table 2. Extracting root from stem by comparing patterns
Table 3. Versions of Simple Stemmers

10
8. CONCLUSIONS AND RECOMMENDATIONS
In this work, the definitions concerning stemming approaches, analysis of the generic stemmers,
the morphological structure of the Arabic language, and the published Arabic stemmers have
been discussed. The previous work by many authors proved greatly that their work was
autonomous, and no signee towards a standard Arabic stemmer was shown. Details of stemmers’
algorithms were not available, suffixes and prefixes lists were not clear and some were not
available as well. Every author develops his own algorithm and named it after himself or as he/
she likes.
Every columnist called an accumulation of stemmers for appraisal purposes, and the after-effects
of the appraisal are referred to that accumulation only. Diverse evaluation samples were used and
many stemmers were developed. This can be easily seen from the previous work mentioned in
section (2).
As a result of this study, the authors suggest the followings:
A. A research should be conducted to analyze most of the effective published Arabic stemmers in
order to decide on a standard Arabic stemmer.
B. The standard Arabic stemmer must take the Arabic diacritics into considerations, because this
affects the semantics of the language. Hopefully, this will reduce stemming ambiguities and
errors.
C. The future Arabic stemmer has to be very intelligent in order to deal with all kinds of word
variants. This will probably use what the authors of this work call, a Hybrid Intelligent Arabic
Stemmer (HIAS).
D. An agreed upon test collection has to be chosen and standardized so as all stemmers are tested
against it.
The author proposes that such a collection must be the Holy Quran.
REFERENCES
[1] Abu-Salem H., Al-Omari M. & Evens M., "Stemming methodologies over individual query words for
an Arabic IR system". Journal of the American Society for Information Science, 50: 524 – 529, 1999.
[2] Al-Fedaghi S. and Al-Anzi F., “A new algorithm to generate Arabic root-pattern forms”. In
proceedings of the 11th national Computer Conference and Exhibition. PP 391-400. March 1989.
[3] Aljlayl M. and Frieder, O., "On Arabic Search: Improving the Retrieval Effectiveness via a Light
Stemming Approach". Proceedings of the eleventh international conference on Information and
knowledge management, 340-347, 2002.
[4] Allan J. and Kumaran G., "Details on stemming in the language modelling framework". In UMass
Amherst CIIR Tech. Report, IR-289, 2003.
[5] Al-Shalabi R., Kanaan G., & Al-Serhan H., "New approach for extracting Arabic roots". In
proceedings of The International Arab Conference on Information Technology, 2003.
[6] Al-Shalabi R. and M. Evens. “A computational morphology system for Arabic”. In Workshop on
Computational Approaches to Semitic Languages, COLING-ACL98. 1998.
[7] Buckwalter T., "Buckwalter Arabic Morphological Analyzer Version 2.0". Linguistic Data
Consortium (LDC) catalogue number LDC2004L02, ISBN 1-58563-324-0, 2004.
[8] Chen A. and F. Gey. “Building an Arabic Stemmer for Information Retrieval”. In Proceedings of the
11th Text Retrieval Conference (TREC 2002), National Institute of Standards and Technology, 2002.
[9] Darwish K. , Douglas W. Oard, "CLIR Experiments at Maryland for TREC 2002: Evidence
Combination for Arabic-English Retrieval". In TREC, 2002.
[10] Darwish K. and Oard D., “Term Selection for Searching Printed Arabic”, in Proceedings of the 25th
ACM SIGIR Conference, pp. 261–268, 2002.

11
[11] Farag A., Andreas N., "N-Grams Conflation Approach for Arabic Text", SIGIR'07 iNEWS07
workshop, 2007.
[12] Fouzi H., Aboubekeur H., and Abdulmalik S., "Comparative study of topic segmentation Algorithms
based on lexical cohesion: Experimental results on Arabic language", The Arabian Journal for
Science and Engineering, Volume 35, Number 2C, 2010.
[13] Frakes W., Fox C. “Strength and similarity of affix removal stemming algorithms”. ACM SIGIR
Forum, Volume 37, No. 1., 26-30, 2003.
[14] Freund, G. and Willett P., "Online Identification of word variants and arbitrary truncation searching
using a string similarity measure". Information Technology: Research and Development 1: 177-187,
1982.
[15] Hoopr R. & Paice C., "The Lancaster Stemming Algorithm", 2005.
http://www.comp.lancs.ac.uk/computing/research/stemming/index.htm.
[16] Hull D., "Stemming algorithms: a case study for detailed evaluation", Journal of the American
Society for Information Science, 47(1), 70-84, 1996.
[17] Khoja S. and Garside R., "Stemming Arabic text". Technical report, Computing Department,
Lancaster University, Lancaster, 1999.
[18] Krovetz R., "Viewing Morphology as an Inference Process", ACM Press, p.p. 191202, 1993.
[19] Larkey L., and M. E. Connell. “Arabic information retrieval at UMass in TREC-10”. Proceedings of
TREC 2001, Gaithersburg: NIST. 2001.
[20] Larkey S., Ballesteros L., Margaret E. Connell, “Improving Stemming for Arabic Information
Retrieval: Light Stemming and Occurrence Analysis”, in Proc. of the 25th ACM International
Conference on Research and Development in Information Retrieval (SIGIR’02), Tampere, Finland,
pp.275–282, 2002.
[21] Larkey S., Ballesteros L., Margaret E. Connell, "Light Stemming for Arabic Information Retrieval,
Arabic Computational Morphology Text", Speech and Language Technology Volume 38, 2007, pp
221-243.
[22] Nizar H., "Introduction to Arabic Natural Language Processing", ACL’05 Tutorial University of
Michigan - 2005.
[23] Sawalha M. and Atwell E., “Comparative Evaluation of Arabic Language Morphological Analyzers
and Stemmers”, in Proceedings of COLING-ACL, 2008.
[24] Syiam M., Fayed Z. & Habib M., "An Intelligent System for Arabic Text Categorization", IJICIS,
Vol.6, No. 1, 2006.
- Appendix A –
Versions of light stemmers with prefixes

12
- Appendix B -
Versions of light stemmers with suffixes
- Appendix C -
Comparison of Arabic Light Stemmers
Author
Mohammed A. Otair is an Associate Professor in Computer Information Systems,
at Amman Arab University-Jordan. He received his B. Sc. in Computer Science
from IU-Jordan and his M.Sc. and Ph.D in 2000, 2004, respectively, from the
Department of Computer Information Systems-Arab Academy. His major
interests are Mobile Computing, Databases, ANN.He has more than 30
publications.

COMPARATIVE ANALYSIS OF ARABIC STEMMING ALGORITHMS

More Related Content

What's hot (20)

Similar to COMPARATIVE ANALYSIS OF ARABIC STEMMING ALGORITHMS (20)

More from IJMIT JOURNAL (20)

Recently uploaded (20)

COMPARATIVE ANALYSIS OF ARABIC STEMMING ALGORITHMS