Opinion-Ontologies
Short and Sharp
Iaakov Exman
Software Engineering Dept., The Jerusalem College of Engineering – JCE Azrieli, Jerusalem, Israel
Keywords: Opinion-Ontologies, Subjective, Sharp, Tweetable, Recommendation.
Abstract: Opinion-Ontology is a short and sharp tweetable recommendation conceptualization which can be actually
sent in a message. But, one should ask in which sense is this a true ontology? On the one hand, it does not
represent the common vocabulary or the shared meanings of a domain, as it is subjective. On the other hand,
it does contain a semantic structure, which in spite of being subjective enables making inferences and taking
rational decisions in practical situations. These are demonstrated by case studies with several examples of
booking a table in a restaurant or a room in a hotel in previously unvisited places. The proposed
characteristics of opinion-ontologies – efficient information transmission and integration with similar
opinion-ontologies without expanding their sizes – can be and we actually intend to implement in a software
system, to enable testing in practice, the whole approach.
1 INTRODUCTION
In view of the rapid changes in the personal way of
decision making, caused by:
Web usage – for booking hotels and flights,
scheduling meetings in restaurants, etc. in
previously unvisited places;
Fast messages – in SMS format, in tweets,
smartphone applications or location-based
social networks;
the booking individual is increasingly relying upon
personal opinions and recommendations. Therefore
new methods to evaluate reliability of opinions and
recommendations are needed.
This work proposes a way to increase reliability
while keeping the overall semantic information – the
basis of judgment evaluation – contained in a limited
size.
Opinion-ontologies are short information pieces
that can be transmitted by fast messages and
integrated with previous messages, without
increasing overall size. This is possible, and
distinguishes them from free-form messages, due to
their structured semantic content.
1.1 Related Work
Opinion and ontologies have been dealt
fromdifferent perspectives. Chang et al. (Chang,
2005) deal with reputation ontologies. They refer
among other things to “Trustworthiness of Opinion
Ontology”. Li and Du (Li, 2011) investigate
ontology-based opinion leader identification for
marketing in online social blogs.
Cambria et al. (Cambria, 2010) describe a public
semantic resource for Opinion Mining, called
SenticNet. Sentic Computing enables analysis of
even very short documents – say one sentence.
Opinion-ontologies are a recent example of short
flexible ontologies for specific purposes. Previous
examples involved micro-ontologies – discussed in
(Biagioli, 1997), nano-ontologies – discussed in the
context of misbehaviour (Exman, 2013), – and
pluggable ontologies – discussed in the context of
non-concepts (Exman, 2012).
The concept of tweetable events and/or abstracts
appears in several contexts in the literature. For
instance, Kiciman (Kiciman, 2012) refers to events
that are interesting and tweetable. People have
thought about tweetable abstracts of scientific papers
as a means to force information compaction.
In a more general vein, there are works dealing
with the interplay of semantics and communication,
in particular for touristic services. See e.g. the
bibliography at the STI web site (STI, 2014).
Akbar et al. (Akbar, 2014) deal with semantic-
aware rules for online communication, among others
with social networks such as Twitter. Toma et al.
(Toma, November 2013) aim at scalable multi-
454
Exman I..
Opinion-Ontologies - Short and Sharp.
DOI: 10.5220/0005166104540458
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2014), pages 454-458
ISBN: 978-989-758-049-9
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
channel communication by means of semantic
technologies.
Toma et al. (Toma, 2014) refer to touristic
services with semantic annotations, using relevant
concepts mapped to types in schema.org
(schema.org, 2014). Toma et al. (Toma, June 2013)
specifically refer to the booking problem in the
tourism domain.
1.2 Paper Organization
The remaining of the paper is organized as follows.
Opinion-ontologies are first given a motivation in
section 2; syntactic, semantic and operational
properties of opinion-ontologies are described in
section 3; merging and inference operations on
opinion-ontologies are considered in section 4; case
studies are presented in section 5; the paper is
concluded with a discussion in section 6.
2 OPINION-ONTOLOGIES
MOTIVATED
The motivation behind opinion-ontologies is
twofold: to obtain condensed information and be
able to evaluate their reliability. Compact
information should be usable to make fast decisions.
2.1 Introductory Example:
A Restaurant Review
Figure 1: Schematic graphical representation of an opinion
ontology (opon) about a restaurant – Ellipses (in blue) are
classes, each of them linked by an association to the main
opon class rest (standing for restaurant). Rectangles (in
yellow) are property names, linked by a thin arrow to the
respective class.
Often there are in newspapers columns dedicated to
restaurant reviewing and ranking. Summarizing one
such specific column about an Italian style restaurant
could look as in the next opinion-ontology (from
now on abbreviated as opon):
This opinion-ontology conveys the opinion about
a restaurant named “Don Giovanni”. The restaurant
is located in downtown. It specializes in Italian food.
In general the meal is stingy, but the tiramisu is
perfect. The restaurant decoration is kitsch with
standard furniture. This opon does not explain
whether kitsch is intentional or just a derogatory
judgment. The service is extremely slow.
A schematic graphical representation of the same
opinion ontology is seen in Fig. 1.
2.2 Sharp Recommendations
The current usage of recommendation web-sites has
several disadvantages:
Long texts – one must read a considerable
amount of text in order to get an overall,
frequently vague, idea about the review;
Low keyword density – one needs to manually
search to eventually extract a too small
amount of important keywords;
Bias and irrelevance – opinions often focus
on arbitrary or low probability events, such as
the specific direction or smaller size of a
particular room in a big hotel.
In contrast, short opinion ontologies intend to
enable rapid winnowing of the undesirable features
listed above and to provide a sharp view of the
expressed opinion.
2.3 Rational Decisions in Short Time
Opinion ontologies can be used as a direct source of
information to make fast rational decisions, as opons
are sharp and short.
For instance, positive reasons for booking a table
at Don Giovanni’s (by opon #28, above Fig.1)
would be a special love for Tiramisu and
indifference to kitsch. Reasons for not booking a
table could be the stingy meal and being in a hurry.
Opinion ontologies can be the input to reasoning
systems, which by comparison to a domain ontology
or by means of rule-based inference, could for
instance conclude that “white tablecloth and
napkins” imply a more expensive bill.
Finally, one could integrate off-line various opinion
ontologies into a single one, in order to make later
on inferences in a shorter time.
“opon #28: rest Don Giovanni loc downtown
food Italian, stingy meal, perfect tiramisu déco
kitsch, standard serv extremely slow.”
Opinion-Ontologies-ShortandSharp
455
3 OPINION ONTOLOGY
PROPERTIES
We here propose opinion ontologies having some
specific syntactic characteristics.
The opinion ontology is supposed to be purely
textual – without any graphical elements or colors.
The opinion ontology is always initiated by an
“opon” term and terminated by a full stop. In
between there are only words and separators
(comma or semicolon).
An opinion ontology consists of two kinds of
terms: class terms, of at most four letters (marked by
italic-bold, here in red for the online digital reader
convenience) and free-text words. Class terms are
not reserved words of a language. They rather could
be explained in a glossary and systematically used
within an application.
An opinion-ontology has as a size parameter an
upper bound to the allowed number of characters
(letters, numbers, signs) used.
Next we point out semantic and operational
characteristics of opinion-ontologies.
3.1 Sharp
Our proposed opinion-ontologies are intended to be
sharp information conveyors due to a few features:
Absence of stopwords – there is no need to
filter low information content words;
Absence of sentence structure – there is no
need to follow standard natural language
grammar, leading to parsimonious
expression;
Imposed opon structure – the linearized tree
structure facilitates reading and fast updating
of the opinion-ontology.
Once people get used to the opon structure, their
manipulation by humans will be increasingly
efficient. Of course, opinion-ontologies are easily
amenable to computerized manipulation.
3.2 Tweetable
By tweetable we mean a quite small and strict upper-
bound to the number of characters in the opinion-
ontology.
We do not mean the specific 140 character limit
of the Twitter social network.
The above referred upper-bound is a parameter to
be assigned in specific applications.
The reasons for the “tweetable” upper bound are
both practical – say the actual usage of tweeter by an
added URL – and deeper semantic arguments.
If one is forced to perform off-line compaction
analysis, before sending an opinion-ontology, one
gains information and semantic density. One thus
transmits more interesting information.
3.3 Subjective
In contrast to a typical domain ontology that is
assumed to represent the common vocabulary and
shared meanings of the domain, an opinion-ontology
is clearly subjective.
It is not necessarily subjective in an individual
sense. It could represent some group or a large
section of public consensus, but still subjective and
even being opinionated.
An opinion-ontology is an ontology, not due to
the overall domain agreement, but due to its
selective semantic character.
4 MERGING AND INFERENCE
OPERATIONS
Merging and inference are two central operations on
opinion-ontologies. Their importance stems from
two fundamental principles:
a. Size Conservation – independently of the
number of merged opinion-ontologies, the
outcome should be a standard opon with the
same syntactic and operational properties as
the original merged opons, e.g. same size
parameter;
b. Semantic Equivalence – an inference
operation on a set of opons should obtain a
semantically equivalent outcome opon.
Merging of Opinion-Ontologies
We now give a sample of merging rules for
opons. First, numerical simplification rules are
given:
a. Reinforcement – when a few
recommendations state the same opinion, use
a numerical weight to express it, say *4
means that the opinion appeared four times
in the merged opons;
b. Contradictions – in case of opposing
opinions use positive and negative weights,
e.g. *-5 *3 (five negations and 3
affirmations);
c. Excess Words – discard the less surprising
(less informative) words within the excess
words of the merging opons.
Next, rules are related to semantic characteristics:
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
456
d. Different terminologies – choose the most
frequent term among the different ones;
e. Ambiguities – disambiguate terms using
opons before they are merged;
5 CASE STUDIES
5.1 Restaurant Recommendation
Here we report the following experiment. We looked
at a restaurant recommendation web-site. The
recommendations were of free text with a typical 50
words length and about 5 keywords categorization.
We took a small sample of these
recommendations and condensed them into the
opinion-ontology format, as follows:
We made a series of worthwhile observations
from this sample. Some of them are:
a- Semantic content – there is an obvious
semantic character to these opons, which may
be used for making inferences; they are not
just dry facts on eating meals;
b- Connotations – classical wood furniture,
Retro touches and take-away do have
implications about food and quality;
c- Branding names – chosen for branding, e.g.
Post-Modern in the Museum of Art, but
induce expectations on food and enable
inferences about quality and price;
d- Incompleteness – classes may lack property
instances, say the last opon serv; but these
may be completed later on.
5.2 Hotel Recommendation
Hotel recommendations – e.g. those in web-sites
offering travel advice – have more complex
characteristics than restaurant recommendations.
Essentially one could have a quite similar structure
as:
In this example the hotel name is “The Hotel”, it
is conveniently located near a metro station, the
neighbourhood is not very recommendable, the room
has air conditioning and it is clean. The amenities
include sauna, and an unreliable wi-fi. The service is
friendly, but overall pricey.
This example shows that the language is quite
fuzzy, leaving a lot of margin for interpretation. For
instance:, it is not precise about the distance to the
metro station; so-so location is probably negative,
but may be acceptable for a given budget.
Summarizing, the loosely structured
information may still be quite challenging.
6 DISCUSSION
This paper proposed opinion-ontologies, short,
sharp, tweetable opinions loosely structured as a
small ontology tree.
The motivation for opinion-ontologies is both:
efficiency of information transmission and deeper
concerns with high density of important information.
The case studies reveal some interesting
observations. The restaurant booking case study,
shows clearly that opinion mining and understanding
is inherently not syntactically based upon presence
of positive words like “nice” or negative words as
“nasty” – agreeing with (Cambria, 2010).
Opinions are subtly expressed through
sophisticated expressions such as “Retro touches”,
eminently context dependent.
6.1 Are Opinion-Ontologies Real
Subjective Ontologies?
We are asking here two different but related
questions:
1. Are opons real ontologies?
2. Are opons subjective ontologies?
In order to answer the first question we cite the
ontology definition by Gruber (Gruber, 1993): an
ontology is a specification of a representational
vocabulary for a shared domain of discourse. The
important points seem to be the “specification of a
“opon #31: rest The Steak House loc
neighborhood food meat, grill déco classical
wood furniture, Retro touches serv meticulous.”
“opon #34: rest The Coffee Network loc
crossroads food take away, coffee house,
breakfast déco standard serv efficient.”
“opon #39: rest Le Bistro loc downtown food
French, gourmet, chef déco dim room serv
culinarian trip.”
“opon #42: rest Post-Modern loc Museum of
Art food chef, meat, pasta, vegetarian déco post-
modern serv.”
“opon #52: hotl The Hotel loc near metro stop,
so-so location room air condition, clean amen
sauna, wi-fi unreliable serv friendly, pricey.”
Opinion-Ontologies-ShortandSharp
457
representational vocabulary” and the “shared”
aspect.
An opon satisfies both important points mentioned
above. It is a specification, it has a representational
vocabulary – although a limited partial one for a
given domain – and it is “shared” among people
expressing and receiving the recommendation.
The second question may be more controversial.
One may claim that an opon is just a set of instances
of clearly non-subjective domain ontology. But we
wish to provide two arguments against this
viewpoint. First, the fact that the domain ontology is
not subjective does not necessarily imply that the
opon also is non-subjective, because the essence of
subjectivity is its dependence on interpretation.
Second, the logic of opons is most probably non-
monotonic. For instance, ‘classical furniture implies
quality food’ is sometimes true, sometimes not.
6.2 Future Work
The next stage of this work is to implement, and test
the whole approach and run extensively a system
with the capabilities proposed here:
Compacting free text – into short and sharp
opinion-ontologies;
Merging opons – i.e. given two or more
opinion-ontologies, to merge their
information into a new unique one without
expanding the opon sizes;
Making inferences from opons – by using
rules such as a restaurant with “white
tablecloth and napkins” is more expensive
than another one in which tables are without
tablecloth.
6.3 Main Contribution
The main contribution of this work is the concept
and detailed characterization of opinion-ontologies,
for efficient transmission and manipulation of
recommendations.
REFERENCES
Akbar, Z, Garcia, J.M, Toma, I. and Fensel, D., 2014. “On
Using Semantically-Aware Rules for Efficient Online
Communication”, in Proc. 8th International Web Rule
Symposium (RuleML 2014), August 18-20, Prague,
Czech Republic.
Biagioli, C., 1997. “Towards a legal rules functional
micro-ontology”, in Proc. LEGONT’97 Workshop.
Cambria, E., Speer, R., Havasi,C. and Hussain, A., 2010.
“SenticNet: A Publicly Available Semantic Resource
for Opinion Mining”, in AAAI Fall Symposium, pp.
14-18.
Chang, E., Hussain, F.K. and Dillon, T., 2005.
“Reputation ontology for reputation systems”, in
SWWS International Workshop on Web Semantics,
pp. 957-966, Springer-Verlag. DOI:
10.1007/11575863_117.
Exman, I., 2012. “A Non-Concept is Not a ¬Concept”, in
Proc. KEOD Conference, pp. 401-404. DOI:
10.5220/0004149704010404.
Exman, I., 2013. “Misbehavior Discovery through Unified
Software-Knowledge Models”, in CCIS, Vol. 348, pp.
350-361. Springer-Verlag. DOI: 10.1007/978-3-642-
37186-8_23.
Gruber, T.R., 1993. “A translation approach to portable
ontology specifications”. Knowledge Acquisition 5
(2), pp. 199-220.
Kiciman, E., 2012. “OMG, I Have to Tweet That! A study
of factors that influence tweet rates”, in Proc. 6
th
Int.
AAAI Conf. on Weblogs and Social Media, pp. 170-
177.
Li, F. and Du, T.C., 2011. “Who is talking? An ontology-
based opinion leader identification framework for
word-of-mouth marketing in online social blogs”,
Decision Support Systems, Vol. 51, (1) pp. 190-197.
DOI: 10.1016/j.dss.2010.12.007.
Lu, Y., Duan, H., Wang, H. and Zhai, C.X., 2010.
“Exploiting structured ontology to organize scattered
online opinions”, in Proc. COLING’10, 23
rd
Int. Conf.
on Computational Linguistics, pp. 734-742.
Schema.org web site: http://schema.org/ Last visited
August 2014.
STI (Semantic Technology Institute) at the University of
Innsbruck, web site: http://oc.sti2.at/results/papers.
Last visited August 2014.
Toma, I., Fensel, D., Oberhauser, A., Fuchs, C., Stanciu,
C. and Larizgoitia, I., June 2013. “SESA: A Scalable
Multi-Channel Communication and Booking Solution
for e-Commerce in the Tourism Domain”, in Proc.
10th IEEE International Conference on E-Business
Engineering (ICEBE 2013), Coventry, UK.
Toma, I., Fensel, D., Gagiu, A.-E., Stavrakantonakis, I.,
Fensel, A., Leiter, B., Thalhammer, A., Larizgoitia, I.,
and Garcia, J.-M., November 2013. “Enabling
Scalable Multi-Channel Communication through
Semantic Technologies”, in Proc. IEEE/WIC/ACM
International Conference on Web Intelligence,
November 17-20, 2013, Atlanta, USA, IEEE
Computer Society Press.
Toma, I., Stanciu, C., Fensel, A., Stavrakantonakis, I. and
Fensel, D., 2014. “Improving the online visibility of
touristic service providers by using semantic
annotations”, in Proc. 11th European Semantic Web
Conference (ESWC'2014), 25-29 May 2014, Crete,
Greece, Springer-Verlag, LNCS.
KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
458
