An evaluation and overview of indices

International Journal of Computer Science, Engineering and Applications (IJCSEA) Vol.4, No.4, August 2014
AN EVALUATION AND OVERVIEW OF INDICES
BASED ON ARABIC DOCUMENTS
Hanandeh
This research is funded by the Deanship of Research and Graduate Studies in Zarqa
University Jordan
ABSTRACT
The paper aims at giving an overview about inverted files , signature files, suffix array and suffix tree
based on Arabic documents collection. The paper also aims at giving the comparison points between all
these techniques and the performance of this techniques on each of the comparison points. Any information
retrieval System is usually evaluated through efficiency and effectiveness of this system. Moreover, there
are two aspects of efficiency: Time and Space. The time measure represents the time needed to retrieve a
document relevant to a specified query, while space represents the capacity of memory needed to create
the two indices.
In this paper, four indices will be built: inverted-file , signature-file, suffix array and suffix tree. However,
to measure the performance of each one, a retrieval system must be built to compare the results of using
these indices.
A collection of 242 Arabic Abstracts from the proceeding of the Saudi Arabian National Computer
Conferences have been used in these systems, and a collection of 60 Arabic queries have been run on the
there systems. We found out that the retrieval result for inverted files is better than the retrieval result for
other indices.
KEY WORDS:
Information retrieval, Arabic document, Indices, Recall, Precision
1. INTRODUCTION
Information Retrieval (IR), refers to the processing of user requests, commonly referred to as
queries to obtain relevant information from collection of documents [6]. Due to historical reasons,
documents in a collection are frequently represented through a set of indexing terms or keywords;
these keywords could be extracted directly from the text of the document or might be specified by
human specialist. To increase the search speed among index terms, several data structures are
proposed to store these terms (called indices) [1].
The most commonly used structures for information retrieval can be classified as lexicographical
indices (indices that are sorted), and indices based on hashing. One type of lexicographical index
is the inverted file. On the other hand, an example of hashing index is the signature file [2].
DOI : 10.5121/ijcsea.2014.4402 9

We used here tow mechanisms in inverted file, the first is word-oriented mechanism for indexing
a text collection and the second is block-oriented mechanism in order to see what is better for
searching, while signature-files are word-oriented index structures based-on hashing [1]. The
suffix array and suffix tree are word-oriented index structures .We built all indices one time by
root and the second by full word.
An inverted file index consists of a record, or inverted list for each term that appears in the
document. A term’s record contains an entry for every occurrence of the term in the document
collection identifies the documents and, possibly, gives the location of the occurrences or a
weight associated with the occurrences [7]. The set of all different terms in the document is
referred to as (the vocabulary) [1].
A signature file uses hash function (or ‘signature’) that maps words to bit masks (signatures) [1].
The most common way for generating signatures from a text is the Super-imposed coding. In
super-imposed coding, a text is divided into text blocks containing the same number of unique,
non trivial words. Each word in a text block is hashed into word signature. A block signature is
generated by super-imposing all word signatures generated from the block [8].
A qualitative comparison of the signature-base method versus inverted method is as follows:
signature-based method requires less space overhead (typically 10-20 %)[1], as apposed to (50-
300%)[2] that inversion requires, on the other hand, inverted mechanism has the advantage of
better search speed than signature file mechanism.
2. GENERAL PROBLEM
Information retrieval deals with representation, storage, organizational and access to information
items. Information retrieval can be problematic for some factors. We have huge amount of data,
the desire to keep related items together; documents have unstructured domain, data. The
information is distributed and interlinked without forgetting the performance criteria.
As the volume of data and documents have reached a limit which is statistically unsuitability, and
as the search processes, whether in the Internet or in the libraries, have become more complicated,
it is necessary to improve the search process.
The efforts was done by researchers in other natural languages such as English language. but the
efforts is still limited in our language, I mean the Arabic language and not satisfy our ambition .
This was the motivation behind this research hoping this work will enhance information retrieval
system degree.
3. OBJECTIVES OF THE STUDY
We summarize the objectives as follows:
1. Investigating and evaluating different indexing method such as Inverted file, Signature
10
file, Suffix tree and suffix array to carry out the candidate indexing term.
2. Applying the work on Arabic information retrieval system using Arabic Root and Arabic
full word as index terms.
3. Evaluating the retrieval effectiveness using the recall, precision measures factors.
4. Comparing between Different indexing method based on Arabic Documents

11
4. EXPERIMENT AND TESTING
To compare the four indices structures, inverted file, signature file, suffix array and suffix tree we
run our system over 242 Arabic documents and pre-defined 60 Arabic natural language queries.
However, the comparison will be based totally on search-speed, storage overhead and average
recall, precision.
Our system was implemented in C# NET language, and Runs on IBM/PCs and compatible
microcomputer. However, running the system each time with different structure will allow us to
measure the search-speed for each one. By search-speed, we mean the time needed to retrieve
relevant documents associated with a specified query.
After we have built the files, the storage space of each one has been identified and, at this point,
we can compare between them to determine which mechanism requires less storage overhead. To
measure the efficiency of retrieval, a retrieval evaluation must be used. Recall, precision
evaluation measurement are used to determine the goodness of retrieved documents. In other
words, high recall means high number of relevant documents retrieved while high precision
means high number of relevant retrieved documents, where the optimal is to have a system with
high recall and high precision .
To compare the efficiency of the all structures, we have computed the average recall and
precision over the 60 queries, the steps followed to compute average recall and precision are:
1. Defining manually the relevant documents for each query.
2. defining a thresh-hold
3. for i = 1 to number of quires
Retrieve all documents that have similarity greater than the thresh-hold Compute the precision at
the 11 standard recall level (0.1,0.2,…..,1)
End
4. for r = 0 to 10
Nq
P(r) = Σ Pi(r) /Nq………………..[1]
Results
i=1
Figure 1 illustrates the space requirement of the original documents for files (inverted, signature,
suffix array, suffix tree) relative to the number of documents in the database.

12
Space Requirments
2000
1500
1000
500
0
10
40
70
95
120
140
170
200
220
230
N o . Of D o cumen ts
S p ace(S IZE )
S ignature file S uffix tree S uffix array Inverted file
Figure 1. space requirement
Figure 2 illustrates the search time of files (inverted, signature, suffix array, suffix tree) relative
to the number of documents in the database. The search time represents the time taken to run the
60 queries in the system.
Time Requirments
60
50
40
30
20
10
0
10 40 70 95 120 140 170 200 220 230
No. Of Docume nts
Time(SIZE)
Inverted file Suffix array Suffix tree Signature file
Figure2. Time requirement
Figure 3. Illustrates the Average Recall Precision of files (inverted, signature, suffix array, suffix
tree) for the 60 query
0 . 8
0 . 6
0 . 4
0 . 2
0
0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6 0 . 7 0 . 7 5 0 . 8 0 . 9
R e c a l l
P r e c i s i o n
S ig n a t u re In ve rt e d S u ffix -a rra y S u ffix -t re e
Figure 3. Average Recall Precision for the 60 queries
Figure 4 illustrates the average recall/precision for the 60 queries when using the all files after
minimizing the false drops for the signature file.

13
1
0 . 8
0 . 6
0 . 4
0 . 2
0
0
0 . 1
0 . 2
0 . 3
0 . 4
0 . 5
0 . 6
0 . 7 5
0 . 8
0 . 8 5
0 . 9
0 . 9 5
1
R e c a l l
P r e c i s i o n
In v e r t e d S u f f i x - a r r a y
S u f f i x - t r e e S ig n a t u r e
Figure 4. Average Recall Precision for the 60 queries after minimizing the false drops
5. CONCLUSION
After running the system and getting the results, a conclusion can be made about the four
indexing mechanisms. However, the comparison between such mechanisms is based on three
factors: Time, Space, and average recall/precision evaluation.
From figure 1 we can conclude that signature file is much smaller than the original size of the
collection. It takes around 16% from the original collection size like almost suffix tree while
inverted file takes around 173% from the original collection size but the suffix array is take
around 154% from the original collection size. Baeza-Yates and Ribeiro-Neto [1] refer in their
book that the signature needs around (10-20%) space while Faloutsos [2] refers that inverted file
needs space from 50-300% of the original size.
Figure 2 illustrates that the search time needed by inverted file is less than the time needed by
other files. Since the signature file has to be searched exhaustively if signatures are organized in a
single sequential file [8]. Donna Harman et al [2], refer that the use of inverted file will improve
search efficiency by several orders of magnitude. Figure 3, we conclude that inverted file has
better average recall/precision than signature file, suffix tree and array tree,. Figure 4 shows that
the average recall/precision for the inverted file is better from the average recall/precision for the
signature file but the difference between inverted and signature less than the Figure 3, this
happens due to the effects of the false drops in the signature file [4]. Lastly, Figure 4 shows that
when reducing the number of words per block in the signature file, the average recall/precision
for the signature file has improved. This is due to Equation 1 [4].
p=(1-(1-w/m)s )w ………………………………. Equation 1[4]
Where
p : false drops probability
w: number of bit set by a word
m: signature length
s : number of words hashed into a block signature

14
REFERENCES
[1] Kanaan,G., Comparing Automatic Statistical and Syntactic Phrase Indexing for Arabic Information
Retrieval, PhD. Thesis, University of Illinois ,Chicago USA,1997
[2] Baeza-Yates, R., and Ribeiro-Neto, B. (1999). "Modern Information Retrieval". Addison Wesley
[3] Darwish,K., Building a Shallow Arabic Morphological Analyzer in one Day, Acl Workshop on
Computational Approaches to Semitic Language,2002,PP.47-57.
[4] Salton, G.,Automatic Text Processing: the Translation Analysis and Retrieval of Information by
Computer, Addison-wesley publishing, USA,1988.
[5] Gname, Mohm’d, The Computing Systems for retrieval by Natural Language, Ph.D. Thesis,
University of Cairo, Egypt, 2003
[6] Khaib, Ahmad. the specification term and their application in Arabic language.Amman,1995. the fifth
teen season for Jordanian Arabic language meeting.P 177-213.
[7] Al-Yaseen, M., Kanaan, G., Al-Shalabi R., Kanaan T. and Bsoul A (2007). Applying Partition
Around Medoids like Clustering Algorithm for Enhancing Retrieval Processes in Arabic Text using
different Similarity Measures. Proceedings of the Information and Communication Technologies
International Symposium ICTIS'07. fez-Morocco.
[8] Kanaan, G., Al-Shalabi R., Bsoul A., and Kanaan T., (2007). Arabic-English Cross-Language
Information Retrieval using Latent Semantic Indexing with Singular-Value Matrix Decomposition.
Proceedings of the Information and Communication Technologies International Symposium
ICTIS'07. fez-Morocco.
[9] Kanaan, G. Riyad, Al-Shalabi, Abd-Allah Al-Akhras, (2006). KNN Arabic Text Categorization
Using IG Feature Selection, The 4th International Multiconference on Computer and Information
Technology, CSIT 2006, Amman, Jordan.
[10] Riyad Al-Shalabi, Ghassan Kanaan, and Manaf H. Gharaibeh, (2006). Arabic Text Categorization
Using KNN Algorithm, The 4th International Multiconference on Computer and Information
Technology, CSIT 2006, Amman, Jordan.
[11] Al-Shalabi R., Kanaan G., Jahjah L. (2005). Question Answering Arabic System based on Passage
Selection. Proceedings of the 5th International Business Information Management Conference
(IBIMA). Cairo, Egypt.
[12] Al-Shalabi, R., Kanaan, G., and Muaidi, H. (2003). New Approach for Extracting Arabic Roots.
Proceeding of the International Arab Conference on Information Technology. Alexandria, Egypt.
[13] Ghwanmeh S., Kanaan G., Al-Shalabi R., Gharaibeh J. and Samara S. (2005). Comparison between
Inverted and Signature Files based on Arabic Documents, International Journal of Applied Science
and Computations, , 12(3): 174-193.

An evaluation and overview of indices

More Related Content

What's hot (19)

Similar to An evaluation and overview of indices (20)

An evaluation and overview of indices