AN AUTOMATIC APPROACH TO FEATURE EXTRACTION
Manuela Angioni and Franco Tuveri
CRS4, Center of Advanced Studies, Research and Development in Sardinia
Parco Scientifico e Tecnologico, Ed. 1, 09010 Pula (CA), Italy
Keywords: Sentiment Analysis, Opinion Mining, NLP, Text Categorization, Semantic Disambiguation.
Abstract: The pervasive diffusion of social networks as common way to communicate and share information is
becoming a valuable resource for analysts and decision makers. Reviews are used every day by common
people or by companies who need to make decisions. It is evident that even the opinion monitoring is
essential for listening to and taking advantage of the conversations of possible customers in a decision
making process. Opinion Mining is a way to analyse opinions related to specific topics: products, services,
tourist locations, etc. In this paper we propose an automatic approach to the extraction of feature terms,
applying our experience in the semantic analysis of textual resources to Opinion Mining task and
performing a contextualisation by means of semantic categorisation, and by a set of qualities associated to
the sense expressed by adjectives and adverbs.
1 INTRODUCTION
Social networks are becoming a common way to
communicate and to share information, interests,
activities and opinions expressed in form of reviews
on blogs, forum, or discussion groups. The ability to
follow opinions gradually become less adequate and
new automatic tools are even more requested and
appreciated especially by large organizations that
track not only brands but even consumer preferences
and opinions. A Gartner analysis for the 2011-year
(Gartner, 2011) illustrates the expectations about
emerging technologies and how the need for
automated methods is growing. Social analytics
offers an answer (Crimson Hexagon, 2009), as one
of the key themes emerging in the near future.
The main goal of our work is the development of
an Opinion Mining system able to extract
automatically the features and the meaningful
information related and contained in opinions, in a
general and not clearly defined domain, from
multiple review sources. To achieve this purpose, we
propose a linguistic approach to extract and
contextualize features related to products or services.
We enriched the contents of WordNet, related to the
meanings expressed by adjectives and adverbs, with
a set of properties having a positive, negative or
objective value associated and other properties that
could add particular and important information in
semantic analysis. The term feature is here used with
the same sense given by (Ding et al., 2008) in their
approach to Opinion Mining.
As in (Benamara et al., 2007), we propose a
linguistic approach to Opinion Mining, based on a
combination of adverbs and adjectives. Our work
also introduces the use of the WordNet (Miller,
1998) synsets. In order to automatically extract the
features from the opinions, the approach is based on
the processing of textual resources, on the
information extraction and on the evaluation of a
semantic orientation. More in details, it performs a
syntactic analysis, a semantic disambiguation and a
contextualization phase and takes into account the
meaning express in conversations.
Moreover the properties related both to
adjectives and to adverbs are associated to each
synset, providing the result of the analysis grouped
into thematic categories.
The remainder of the paper is organized as
follows: Section 2 refers to the state of the art and
related works. Section 3 introduces the approach and
examines the work performed on adjectives and
adverbs’ structures. Section 4 explains the approach
to the feature extraction giving some details about a
modified version of the Leacock-Chodorow
(Leacock and Chodorow, 1998) algorithm and a
demonstration we realized in order to better analyse
the results. Finally, Section 5 draws conclusions.
473
Angioni M. and Tuveri F..
AN AUTOMATIC APPROACH TO FEATURE EXTRACTION.
DOI: 10.5220/0003714504730476
In Proceedings of the 4th International Conference on Agents and Artificial Intelligence (ICAART-2012), pages 473-476
ISBN: 978-989-8425-95-9
Copyright
c
 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 THE STATE OF THE ART
As asserted by (Lee et al., 2008), “Opinion Mining
can be roughly divided into three major tasks of
development of linguistic resources, sentiment
classification, and opinion extraction and
summarization”. Concerning the task of the
development of linguistic resources many works
perform WordNet exploring methods and gloss
classification methods. In opinion summarization
several approaches are based on the use of lexicons
of words able to express subjectivity. Akkaya et al.
(2009) build and evaluate a supervised system to
disambiguate members of a subjectivity lexicon
while Rentoumi et al. (2009) propose a methodology
for assign a polarity to word senses applying a Word
Sense Disambiguation (WSD) process.
Some authors (Lee et al., 2008) asserted that the
systems that adopt syntactic analysis techniques on
extracting opinion expressions tend to show higher
precision and lower recall. One relevant task in
opinion summarization step regards the features
extraction. Some works about features are based on
the identification of nouns through the part of speech
tagging (pos-tagging) and provide an evaluation of
the frequency of words in the review based on tf-idf
criterions and its variation. Others researchers (Zhai
et al., 2010) proposed a constrained semi-supervised
learning method based on the contextualization of
reviews grouped on specific domains and on a
characterization based on features defined by users.
Other two works are (Popescu and Etzioni, 2005),
that worked on the explicit features in noun phrases,
and SentiWordNet (Esuli and Sebastiani, 2007) in
which WordNet is expanded thanks to a semi-
automatic acquisition of the polarity of WordNet
terms, evaluating the polarity of each synset.
3 THE APPROACH
Our approach to the feature extraction of terms takes
advantage of some tools developed in a previous
work on semantic: the Semantic Classifier and the
SemanticNet as extension of the WordNet semantic
net. Moreover we propose a linguistic approach to
extract and contextualize features related to products
or services, and a brief description of the new
resources we developed: a lexical database of
adjectives and adverbs and the semantic distance
algorithm.
3.1 A Subjective Lexical Database
Due to the lack of freely available resources having
the qualities our approach needs in order to
contextualize and group features, a lexical database
of synsets has been defined as extension of the
properties of WordNet. The resource contains a
subset of adjectives and adverbs of WordNet. Each
synset has been enriched with a set of properties
about the polarity and other properties that could add
particular and important information in semantic
analysis. We classified about 2.300 pairs of
adjectives/synsets and about 480 pairs of
adverbs/synsets according to a set of attributes
identified by their association with nouns and verbs
and chosen on the basis of their frequency of use in
the language. The identified characteristics provide
additional information about the content of the
sentences, regarding for instance personal, moral,
aesthetical aspects or related to geography, to time
or to weather for the adjectives. We identified about
15 typologies of adjectives and about 7 typologies of
adverbs. Some of these properties allow a
polarization that can be used by Opinion Mining
algorithms, such as the adverbs of manner.
3.2 The SemanticNet
The SemanticNet has been developed starting from
the synsets of the terms contained in WordNet and
mapped on the contents included in Wikipedia. It
has been defined by adding new nodes, attributes,
and new relations named of Common-Sense. defined
by means of the conceptual associations defined by
the authors in the pages of Wikipedia. Moreover,
each page of Wikipedia is associated to the WordNet
synset having the most correct meaning by means of
a similarity algorithm. In the SemanticNet the nodes
are the “senses” (identified by the synset of
WordNet or by a unique key defined as
“term+category”) and links are given both by the
WordNet semantic relations and by the conceptual
associations built by the Wikipedia authors.
3.3 Feature Extraction
The aspect we want to investigate is related to
opinions about events or facts even in change where
it is difficult to work on a set of reviews well defined
or clearly related to a specific topic. An example
could be related to politics where it could be very
relevant investigate how electorate reacts to
politicians statements. Certainly in this case features
could not be determined a priori, but should be
extracted from text automatically. We integrated the
approach of (Yi, J. et al., 2003) with some tools in
the phase of the extraction of the features, filtering
ICAART 2012 - International Conference on Agents and Artificial Intelligence
474

the list of features through a contextualization phase
by means of the semantic classification. In order to
extract the features, we essentially make use of
syntactic and semantic analysis, the semantic
classifier, the subjective lexical database of
adjectives and adverbs and the semantic net of
WordNet. The syntactic analysis, composed by a
pos-tagging and phrase-chunking phase, has been
performed in order to identify and extract nouns and,
by means of the lexical database, the set of
properties related to the adjectives and adverbs.
A distinction between objective and subjective
sentences is performed considering the combination
of adjectives and adverbs having a polarity
associated. Given a collection of reviews related to a
specific topic, the semantic classifier categorizes the
sentences, extracts as result a set of categories and
weights as measures of their relevance in order to
define the context for the features in the review. The
list of categories with their weights describes the
domain referred to the collection of reviews.
For instance, starting from a collection of
reviews about tourism and especially linked to
hotels, the classifier categorizing the sentences
containing opinions produces categories about
geographical locations, buildings, staff, food or
gastronomy and their weights. These categories are
useful in the WSD of the candidate features
previously extracted by means of a tf-idf function.
The mapping between synsets related to each
feature and the categories, defines a first level of
contextualization of the features. The mapping also
reduces the list of candidate features to a restricted
set having synsets mapped to the top categories. The
result is obtained using as reference the categories
defined in WordNet Domains (Magnini et al., 2002),
a lexical resource, based on the Dewey Decimal
Classification system, representing domain
associations among the word-senses of WordNet.
In order to perform the WSD activities a
semantic distance algorithm, based on the Leacock-
Chodorow algorithm, has been applied. The
algorithm calculates the distance measure between
feature terms based on information related to the
number of synsets referred to each term and
according to the mapping of the synsets to the set of
categories extracted by the semantic classifier. The
algorithm also defines a matrix of the relations of the
features having as rows and columns the names of
the extracted features and as values the above
distance measure. The matrix defines in this way a
map of the existing relations between features and,
with the introduction of a threshold on the values,
puts in evidence different areas of features in order
to group them in more specific thematic set.
4 AN INFORMAL TEST CASE
The test is performed on a collection of about 100
reviews of the Alma Hotel of Alghero (Sardinia,
Italy) extracted from TripAdvisor, a site of advices
and opinions from a huge community of travellers.
About 950 sentences have been extracted from the
above reviews by the tools for the syntactic analysis.
Tf-idf criterions allows to define an initial set of
about 450 candidate features, based on the frequency
of their use in the collection.
Once identified, the semantic classifier
categorizes the subjective sentences and gives back
as result a small set of categories with their
associated weights. Then a reduction of candidate
features based on the synset mapping on categories
has been performed. As final result we obtain a set
of about 80 features. As said the semantic distance
of synsets and the assignment of weight to the
features defines the matrix of the relations between
the features extracted. Figure 1 shows the features
related to the feature breakfast, one of the features
extracted by the system and related to the topic
tourism. In the upper side there are two definitions:
the first from Wikipedia and the second from
WordNet. The SemanticNet in this specific case is
Figure 1: The feature breakfast and its related data.
used in order to show the existing mapping between
the concepts as defined in the two different
resources and a list of related images. Follows the
list of the features with the weights measuring their
relation to the breakfast feature. Figure 2 shows a
AN AUTOMATIC APPROACH TO FEATURE EXTRACTION
475

Figure 2: A sample of opinion summarization.
sample of result of the opinion summarization task.
The figure illustrates a simple polarity evaluation
comparing the preferences about the represented
features. In the phase of data aggregation, it is
important to notice relations between features, e.g.
breakfast and buffet, as showed in the first picture in
order to give a correct interpretation of data. The
system currently does not analyse this kind of
relations between data, but is able to deduce only
that users like the breakfast of the hotel but less the
buffet organization.
5 FUTURE WORKS
Several Opinion Mining methods and techniques
have been developed in order to analyse contents
and reviews. In this paper an automatic approach to
the extraction of feature terms has been proposed.
With the introduction of the synsets and the
semantic categorization, we aim to define a method
of extraction of more accurate meanings and features
from textual resources. We propose the
identification of the context of features by means of
the semantic net of WordNet in order to reach a
more complete list of features and attributes related
to an object. Future works will provide the
development of the opinion summarization task, the
definition of a tool for the navigation of features and
related opinions and the generalization of the
approach in order to apply it to general contexts.
The tool will perform opinion monitoring activities,
an essential task in listening to and taking advantage
from consumer preferences and opinions. A
validation to support the value of the expressed ideas
will be one of the goals of the above mentioned
approach and experimental results will be product.
REFERENCES
Akkaya, C., Wiebe, J., Mihalcea, R., 2009. Subjectivity
word sense disambiguation. In Conference on Empirical
Methods in Natural Language Processing, Singapore,
The Association for Computational Linguistics.
Benamara, F., Cesarano, C., Picariello, A., Reforgiato, D.,
Subrahmanian, V. S., 2007. Sentiment Analysis:
Adjectives and Adverbs are better than Adjectives
Alone. In Proceedings of ICWSM 07 International
Conference on Weblogs and Social Media.
Crimson Hexagon, 2009. Listen, Understand, Act. How a
listening platform provides actionable insight. http://
www.crimsonhexagon.com/PDFs/Crimson_Hexagon_
Listen_Understand_Feb_2009.pdf
Ding, X., Liu, B., Yu, P. S., 2008. A Holistic Lexicon-
Based Approach to Opinion Mining. WSDM '08 Pro-
ceedings of the international conference on Web
search and web data mining, ACM, New York, USA.
Esuli, A. Sebastiani, F., 2007. PageRanking WordNet
synsets: An application to Opinion Mining.
Proceedings of the 45th Annual Meeting of the
Association of Computational Linguistics Volume: 45,
Issue: June, Publisher: Association for Computational
Linguistics, Pages: 424-431
Gartner, 2011. Gartner's 2011 Hype Cycle Special Report
Evaluates the Maturity of 1,900 Technologies. http://
www.gartner.com/it/page.jsp?id=1763814
Leacock C. and Chodorow M. 1998. Combining local
context and WordNet similarity for word sense
identification. In Fellbaum 1998, pp. 265-283.
Lee, D., Jeong, O. R., Lee, S., 2008. Opinion Mining of
customer feedback data on the web. In ICUIMC '08
Proceedings of the 2nd international conference on
Ubiquitous information management communication
Magnini, B., Strapparava, C., Pezzulo, G., Gliozzo, A.,
2002. The Role of Domain Information in Word Sense
Disambiguation. Natural Language Engineering,
special issue on Word Sense Disambiguation, 8(4), pp.
359-373, Cambridge University Press
Miller, G., 1998. WordNet: An Electronic Lexical
Database, Bradford Books
Popescu, A. M., Etzioni, O, 2005. Extracting Product
Features and Opinions from Reviews. Proceedings of
the 2005 Conference on Empirical Methods in Natural
Language Processing.(EMNLP’05).
Rentoumi, V., Giannakopoulos, G., 2009. Sentiment
analysis of figurative language using a word sense
disambiguation approach. In International Conference
on Recent Advances in Natural Language Processing
(RANLP 2009), Borovets, Bulgaria, The Association
for Computational Linguistics
Yi, J., Nasukawa, T., Bunescu, R., Niblack, W., 2003.
Sentiment analyzer: Extracting sentiments about a
given topic using natural language processing
techniques. In Proceedings of the IEEE International
Conference on Data Mining.
Zhai, Z., Liu, B., Xu, H., Jia, P., 2010. Grouping Product
Features Using Semi-Supervised Learning with Soft-
Constraints. In Proceedings of the 23rd International
Conference on Computational Linguistics (COLING-
2010), Beijing, China.
ICAART 2012 - International Conference on Agents and Artificial Intelligence
476