Enhancing the labelling technique of

International Journal of Data Mining & Knowledge Management Process (IJDKP) Vol.4, No.1, January 2014

ENHANCING THE LABELLING TECHNIQUE OF
SUFFIX TREE CLUSTERING ALGORITHM
R.Mahalakshmi1 and V.Lakshmi Praba2
1

Research Scholar, Manonmaniam Sundaranar University, Tirunelveli,Tamilnadu
2
Assistant Professor,Government arts College,Sivaganga,Madurai,India

ABSTRACT
Clustering the results of a search helps the user to overview the information returned. In this paper, we
look upon the clustering task as cataloguing the search results. By catalogue we mean a structured label
list that can help the user to realize the labels and search results. Labelling Cluster is crucial because
meaningless or confusing labels may mislead users to check wrong clusters for the query and lose extra
time. Additionally, labels should reflect the contents of documents within the cluster accurately. To be able
to label clusters effectively, a new cluster labelling method is introduced. More emphasis was given to
/produce comprehensible and accurate cluster labels in addition to the discovery of document clusters. We
also present a new metric that employs to assess the success of cluster labelling. We adopt a comparative
evaluation strategy to derive the relative performance of the proposed method with respect to the two
prominent search result clustering methods: Suffix Tree Clustering and Lingo.
we perform the experiments using the publicly available Datasets Ambient and ODP-239

KEYWORDS
Information retrieval, clustering, Search results clustering, Suffix Tree Clustering ,cluster labeling

1. INTRODUCTION
The overwhelming amount of textual documents available nowadays highlights the need for
information organization and discovery. Effectively organizing documents into a hierarchy of
topics and subtopics makes it easier for users to browse the documents. [1]
Search engines would retrieve all the documents based on the given key word and rank them
according to the priority and display the documents page wise. Lets take top 20 ranked
documents ,except the first 3 or 4 rest of the documents are not satisfying the user. More than 50
% of the documents are not relevant to the query since the retrieval is based on keyword and not
based on the semantic similarity.
The ideal solution is , forcing the user to input a larger number of highly accurate keywords, trim
down the number of results and yields somewhat improved ranking precision. This solution is,
unfortunately, not practical for the normal user since most users tend to input not more than 3
DOI : 10.5121/ijdkp.2014.4104

41


keywords[3] . To solve this problem, Cluster the search result approach has been proposed.
These methods have two main advantages. One is to make it easier for the user who has a clear
search target to locate the desired document because the user can easily select the most
appropriate cluster. The other is to assist the user, who would like to browse using just a few
keywords or who has no clear search target, in understanding the outline of the search result
through the labels of the clusters. The other important benefit is that the user often finds
interesting information apart from what he/she wants to retrieve. Realizing these benefits,
however, requires not only correct clustering but also labels clear enough to explain the
clusters[5]. A lot of research is being intended towards clustering ,not using conventional
clustering methods but regarding the clustering task as a task of selecting important key terms or
phrases [5][6][7][8][10].To extract the terms as label candidates from documents firstly we
consider that proper nouns are vital for characterizing documents. Second is a new label selecting
criterion that can select the labels from candidates .The main objective of this paper is to provide
more precise cluster label generation, superior group content discovery and incremental
processing.
This paper deals with the enhancement of generalized Suffix tree based clustering approach. In
general the most repeated phrase in the document tags is considered as cluster name. The general
labelling procedure of the Suffix Tree algorithm is enhanced to improve the cluster label quality.
This paper aims at organizing web search results into clusters facilitating quick browsing options
to the browser providing excellent interface momentous labels to clusters .Suffix tree clustering
produces comparatively more accurate and informative grouped results. The paper is organized as
follows, Section1 deals with Introduction and section II briefs the related work ,section III
describes the proposed methodology section IV details the experiments ,section V discusses the
results and analysis ,finally section VI gives the conclusion of the paper.

2. RELATED WORK
Clustering is a common unsupervised learning technique used to discover group structure in a set
of data. While there exist many algorithms for clustering, clustering is difficult because, unlike in
supervised learning, there are no class labels for the data and, thus, no obvious criteria to guide
the search. Another important problem in clustering is the determination of the number of
clusters, which clearly impacts and is influenced by the feature selection issue.
Scatter/Gather [11] is one of the first system that dealt with cluster labeling, in addition to the
cluster’s important terms, the titles of the document close to the centriod are also considered.
Filippo et al [3] demonstrated that labels extracted from titles provide better description that
those extracted from page’s content. There is a lot of research on linguistic-based summarization
techniques for multiple documents which are also related to the labeling task. Radev et al.[10]
[16]. However, multi-document summaries are usually too long to be utilized as short
comprehensive labels. Several labeling approaches attempt to enrich terms by exploiting external
resources for labeling, for example, the WordNet lexical database [4] was used to extract root
meanings of important terms and to determine semantic relationships among these terms. David
Carmel [10] utilized Wikipedia to represent the meaning of a text fragment as a weighted vector
of Wikipedia concepts. Wikipedia has recently become one of the major knowledge resource for
many information retrieval tasks, including text categorization and clustering [15, 16, 17],
computing semantic relatedness between concepts [18, 9], and predicting document topics[10]
[19]. Toda and Kataoka [21]also used named entities extracted from the text for labeling.
42


However, in many cases, a labeling approach that is solely based on the cluster content may have
difficulties in providing discriminative labels. Several labeling solutions look for alternative
resources.

3. SEARCH RESULT CLUSTERING ALGORITHM - STC
Zamir and Etzioni [4] introduced the suffix tree clustering algorithm (STC), which runs in O(n)
without computing O(n2) similarity values. In detail, STC is made up of three steps
Step 1. A suffix tree for all suffixes of each document in D = {d1, . . . , dn} is constructed, and
each suffix is associated with the set of documents wherein it is contained. In other words, using
the notation given above, for each edge e (each of which represents a certain suffix) the set S(e) is
computed. The sets S(e) with |S(e)| ≥ 2 are called “base clusters” and identify the documents di
with i ∈ S(e).
Step 2. Each base cluster is assigned a score f, which is a function of |S(e)| and the length of the
suffix that is represented by e. In [Zamir and Etzioni 1998] the authors propose f as the product of
|S(e)| and the length of the suffix that is represented by e.
Step 3. The k base clusters S1, . . . , Sk that score best under f are selected. A similarity graph in
which the base clusters form the node set is generated, and an edge between two nodes Si and Sj
is added if the Jaccard coefficient of Si and Sj is larger than 0.5, say, when |Si∩Sj | |Si∪Sj | > 0.5.
The connected components of this graph form the final clusters.
Step 4: Each base cluster is associated with a suffix, which can serve as a label for this cluster.
This method solves two basic problems in topic identification for document clusters [Stein and
Meyer zu Eißen 2004b]:word order preservation and topic length determination
STC [9] has proven to work well on document snippets that are returned by search engines , but
its properties have been analyzed by many researchers . As pointed out above STC is a heuristic
algorithm which is highly efficient, and has got few drawbacks.

4. IMPROVED STC ALGORITHM
After the execution of Step 3 in the normal STC ,that is before assigning base cluster suffixes to
the clusters ,all the cluster label phrases are reviewed by the following criterions
1. If we have two cluster labels as synonyms or with the simple difference like singular and
Plural merge the two into a single cluster ,discard the replicated documents and unique
documents could be retained by selecting a unique label from the existing key phrases.
2. If any verb is given as a label for the cluster, relabeling it by adding a meaningful noun from
the Key phrases or if not found, prefix the query phrase in to it.
3. On line databases are used to find the synonyms of the key phrases. Two online data
collections are increasingly used in all kinds of ways in IR. One is Wikipedia. The other is
WordNet. here we used Wordnet. Wordnet is used to find the parts of option which will list out
43


few Synonyms, then by applying term weight and term ranking we can select the most
descriptive phrase/label .
4. If a cluster label is part of the another cluster label then their relevance is checked. If more than
50% of the documents are replicated/overlapped in both the clusters, the rest are checked against
other cluster doc if all are overlapped with other cluster then remove the cluster, otherwise add
them in to other topics. Hence we effectively lessen shadowing in the generated clustering.

5. EXPERIMENTS
ODP239, Ambient and PubMed datasets are used for our experiments. We run the two SRC’s
namely STC and LINGO with 10 different query key words and analyzed the same query with the
enhanced STC.The general observations are represented by graphs. Table 1 shows the basic
parameters and the attributes settings of the Suffix Tree algorithm ,Table 2 lists out the sample
resultant cluster labels produced by LINGO and STC. Table 3 shows the revised list of improved
cluster labels with no of documents. Number of Clusters are reduced from 16 to 12
Table 1 parameters setting of STC and Lingo algorithms
Dataset
Query Keyword

No of Doc
No of Clusters
Algorithms
Max Phrases Per Label
Optimal Label Length
Base Cluster Merge Threshold
Word Boost
Max Cluster Phrase Overlap

ODP239
Women’s health, jaguar
Education,Musiums,
Agriculture,environment,dic
tionary,entertainment
100
20
STC,LINGO
4
3
60%
60-80%
60%

Single Term Boost

Ambient
Montecarlo,Butterfly,
camel,

71
20
STC,LINGO
4
3
60%
60-80%
60%

80%

80%

Table 2 List of query terms, Datasets,Ambiguous lables and improved labels generated by STC, Lingo and
enhanced STC
S.No

Query Term

Ambiguous labels in
Lingo

Ambiguous
Labels in STC

Revised Labels
in Enhanced
STC

1

Cooking

Lemon,onion,includes

Lemon,uses,recipe, Soups,recipies
recipie

2

Children

Kids,Child,designs,A
vailable,sets

Dataset

ODP239

Baby,Babies,Offer Babies,Kids
s,offeringKids,Chil
dren
44


3

Entertainment

DVD,DVDS,speciali
zing,region,selecion,
merchandise,sales

selection,large
selection of CDs,
CDs,DVDs,CDs&
DVDS

Selection,CDs,D
VDs

4

Women’s
health

Forum,answer,topic
etc

Women,woman,fe Health centers,
male,topic,center,c diagnosis and
ondition,treatment. Treatment
.
uterus,prolapse,b
ladder

5

Beagle

adopting, search

Linux,Desktopsear Open
ch,Club,national
source,National
clubs,pages
beagle clubs,

6

Scorpian

7

Computer

increase,presence,selec
tive,changes,
improvement,efficient

Available,use,used, Compter
method,methods,si uses,methods,
gnificant,study
significant
performance

8

Operating
System

Score,organic,increase
d,water,biological,Para
meters, surgery,fields

performance,
performance
analysis,clinical,us analysis,clinical
ed, study
systems

9

Nobel prize

changed,number,
highly,
reprogramming,

reprogramming,cel Nobel prize
l,new,conjugated,s
shared
hared

10

Java

concentration,predictin indonesia.indonesi
g, outcome,etc
an,high,
compared,used

Ambient

Scorpian,Scorpians Scorpians,Scorpia
,Stung,Scorpian
n stings
stings,Known
PubMed

indonesia,outcom
e and analysis

The following are the observations
1. Few Label names are found to be duplicated then the cluster is removed and the unique
documents are merged with other cluster example “Treatment “ is removed and the
unique documents are merged with the “ Diagnosis &Treatment”
2. Verbs are given as label name for few clusters, they are further prefixed by a query
phrase and relabeled.(ex Label name CENTER is renamed as HEALTH CENTER to
give more clarity to the user.) Wordnet online dictionary is used.
3.

Some label names are synonyms, Ex label name FEMALE and WOMEN are the same
,the overlapping documents are removed and the remaining documents are merged with
the WOMEN cluster..
45


4. label names are improved by increasing the wordboost by 100 % for example there was
a lable called TOPIC and it was labeled by UTERUS,PROLAPSE,BLADDER. After
increasing the wordboost to 100% ,which gives more precision about the group to the
user.
5. Document overlapping is reduced by 12 % overall.

6. RESULTS AND ANALYSIS
6.1 Evaluation of Cluster Labels
Although clustering has been studied for several decades, the fundamental problem of a valid
evaluation has not yet been solved. Evaluating the quality of clustering results is still a challenge
in recent research. The sound evaluation of clustering results in particular on real data is
inherently difficult. In the literature, new clustering algorithms and their results are often
externally evaluated with respect to an existing class labeling.[14].The cluster labels could be
evaluated based on the following parameters Comprehensibility Descriptiveness, Discriminative
power Uniqueness, Non-redundancy.
1. Comprehensibility (f1):
A reader should have a clear imagination of the contents of a cluster. It can be formally defined
as the following.∀ܿ ∈ ‫ܮ ∈ ܲ :݈ܿ ∈ ݌∀ܥ‬ሺ‫ܩ‬ሻ௣ > 1 where lc is the cluster label of cluster c, p a
phrase of lc, and L(G) determines a formal language identifying noun phrases.

2.0 Descriptiveness (f2):
Every document of a cluster should contain the associated cluster label
∀ܿ߳‫݂ܿ݀ :ܿܲ ∈ ′݌∀݈ܿ ∈ ݌∃ܥ‬ሺ‫ ′′݌‬ሻ ≪ ݂݀ܿሺ‫݌‬ሻ where Pc is the set of phrases in the cluster c.
ଵ

݂2ሺܿ, ‫݌‬ሻ = 1 − ௉௖/௅௖ ∙ ∑௣′ఢ௉௖ ݂݀ܿሺ‫ ′′݌‬ሻ/݂݀ܿሺ‫݌‬ሻ

2.0

3.0 Discriminative Power(f3)A cluster label should only be present in documents of its own
cluster, could be formally defined as

46


4.0 Uniqueness (f4):Cluster labels should be unique. , formally defined as

5.0 Non-redundancy (f5)Cluster labels should not be synonymous, formally,
∀ܿ ∈ ‫ݏݑ݋݉ݕ݊݋݊ݕݏ ݐ݋݊ ݁ݎܽ ݅݌ ݀݊ܽ ݌ :݈ܿ ∈ ′݌ ,݌∀ܥ‬

6.0 Relevance of a phrase with respect to a cluster:All constraints can be combined into a single
criterion:

where wi is a weighting factor and F = {f|1 . . . 5}, namely,f1 Comprehensibility,f2
Descriptiveness,f3 Discriminative Power,f4 Uniqueness,f5 Non-redundancy .Note, that the effect
of every constraint on the quality of a phrase is so far unevaluated.
Table 3 Label evaluation

47


Fig 1. Comparative analysis on label evaluation of STC,LINGO and Enhanced STC
Table.4 Label relevance

Fig 2. Comparitive analysis of Label
relevance of STC,Lingo and Enhanced STC

48


7. CONCLUSION
Search results clustering is one of many methods that can be used to improve user experience
while searching collections of text documents. We must accurately and concisely describe the
contents of the cluster, so that the user can quickly decide if the cluster is interesting or not. This
aspect of document clustering is sometimes neglected. cluster labelling is not less important than
clustering. In fact a good cluster with a poor descriptive label is likely to be ignored by the user.
As discussed that evaluating the quality of clustering results is still a challenge in recent research
The cluster labels are evaluated based on the following parameters
Comprehensibility
Descriptiveness, Discriminative power Uniqueness, Non-redundancy. It was proved that the new
proposed method of enhanced STC produced semantically meaningful, comprehensible and
compact text labels to the document clusters. This paper addressed the effect of repeated
acquisition of labels for search results clusters when the labelling was imperfect. We examined
the improvement in label quality via avoiding repeated labelling, and focus especially on the
improvement of descriptive and unique labels. The result shows that the Overlapping is reduced
by 12 % and the labels relevancy is also improved by 8.4 percentage when compared to other two
SRC’s STC and LINGO. We are able to increase the label quality with respect to different
parameters described in the evaluation criteria. If the key phrases are verbs that could be
improved by finding appropriate noun in the key phrases or by prefixing the query term with the
verb.In our work we didn’t concentrate much on the label name as adjectives.

REFERENCES
[1]

Text Document Topical Recursive Clustering and Automatic Labeling of a Hierarchy of Document
Clusters Xiaoxiao Li1, Jiyang Chen, and Osmar Zaiane
[2] “Cluster Generation and Cluster Labelling for Web Snippets Filippo Geraci 1 ,2, Marco Pellegrini,
Marco Maggini 2, and Fabrizio ebastiani
[3] A Search Result Clustering Method using InformativelyNamed Entities” Hiroyuki Toda NTT Cyber
Solutions Laboratories, NTTCorporation WIDM’05, November 5, 2005, Bremen,
Germany.Copyright 2005 ACM 1-59593-194-5/05/0011 ...$5.00.
[4] Web Document Clustering: A Feasibility Demonstration” Oren Zamir and Oren EtzioniDepartment of
Computer Science nd Engineering University of Washington Seattle, WA 98195-2350 U.S.A.
[5] Zamir, O., Etzioni, O. and Grouper, A.: “Grouper: A Dynamic Clustering Interface to Web Search
Results.” roceedings of WWW8, pp.1361-1374, 1999.
[6] Zeng, H. J., He, Q. C., Chen, Z., Ma, W. Y. and Ma, J.: “Learning to Cluster Web Search Results.”
Proceedings of SIGIR’04, pp.210-217, 2004.
[7] Kummamuru, K., Lotlikar, R., Roy, S., Signal, K. and Krishnapuram, R.: “A hierarchical monothetic
document clustering algorithm for summarization and browsing search results.” Proceedings of
WWW’04, pp.658-665, 2004.
[8] Ohta, M., Narita, H. and Ohno, S.: “Overlapping Clustering Method Using Local and Global
Importance of Feature Terms at NTCIR-4 Web Task.” Working Notes of NTCIR(NII-NACSIS Test
Collection for IR Systems)-4 Vol.Supl. 1, pp.37-44, 2004.
[9] The Suffix Tree Document Model Revisited Sven Meyer zu Eissen (Paderborn University, Germany
smze@upb.de) Benno Stein (Bauhaus University Weimar, Germany benno.stein@medien.uniweimar.de) Martin Potthast (Paderborn University, Germany beebop@upb.de
[10] Enhancing Cluster Labeling Using Wikipedia David Carmel, Haggai Roitman, Naama Zwerdling
IBM Research Lab Haifa 31905, Israel{carmel,haggai,naamaz}@il.ibm.com SIGIR’09, July 19–23,
2009, Boston, Massachusetts, USA. Copyright 2009 ACM 978-1-60558-483-6/09/07 ...$5.00.
[11] C. Carpineto, S. Osi ´nski, G. Romano, D. Weiss. A Survey of Web Clustering Engines. ACM
Computing Surveys (CSUR), 41(3):Article 17, 2009.
49


[12] C. Clifton, R. Cooley, and J. Rennie. TopCat: Data Mining for Topic Identification in a Text Corpus.
IEEE Trans. Knowl. Data Eng., 16(8):949–964, 2004.
[13] W. de Winter and M. de Rijke. Identifying Facets in Query-Biased Sets of Blog Posts. In Proceedings
of ICWSM 2007, pages 251–254.
[14] On Using Class-Labels in Evaluation of Clusterings Ines Färber1, Stephan Günnemann1, Hans-Peter
Kriegel2, Peer Kröger2, Ahornstrasse 55, 52056 Aachen, Germany, 2Ludwig-MaximiliansUniversität München Oettingenstrasse 67, 80538 München, Germany.
[15] F. Geraci, M. Pellegrini, M. Maggini, and F. Sebastiani. Cluster Generation and Cluster Labelling for
Web Snippets: A Fast and Accurate Hierarchical Solution. In Proceedings of SPIRE 2006, pages 25–
36.
[16] S. Osi ´nski, J. Stefanowski, and D. Weiss. Lingo: Search Results Clustering Algorithm Based on
Singular Value Decomposition. In Proceedings of IIPWM 2004, pages 359–368.
[17] A. Popescul and L.H. Ungar. “Automatic labeling of document clusters”.
http://www.cis.upenn.edu/~popescul/Publications/popescul00labeling.pdf, 2000.
[18] B. Stein and S. Meyer zu Eißen. Topic Identification: Framework and Application. In Proceedings of
I-Know 2004, pages 353–360.
[19] H. Toda and R. Kataoka. A Clustering Method for News Articles Retrieval System. In Proceedings of
WWW 2005, pages 988–989.
[20] D. Weiss. “Descriptive clustering as a method for exploring text collections”. Ph.D. dissertation.
Pozna´n University of Technology, Poland, 2006.

50

Enhancing the labelling technique of

More Related Content

What's hot (17)

Viewers also liked (7)

Similar to Enhancing the labelling technique of (20)

Recently uploaded (20)

Enhancing the labelling technique of