=Paper=
{{Paper
|id=None
|storemode=property
|title=EUROSENTIMENT: Linked Data Sentiment Analysis
|pdfUrl=https://ceur-ws.org/Vol-1272/paper_116.pdf
|volume=Vol-1272
|dblpUrl=https://dblp.org/rec/conf/semweb/Sanchez-RadaVIB14
}}
==EUROSENTIMENT: Linked Data Sentiment Analysis==
https://ceur-ws.org/Vol-1272/paper_116.pdf
EUROSENTIMENT: Linked Data Sentiment
Analysis
J. Fernando Sánchez-Rada1 , Gabriela Vulcu2 , Carlos A. Iglesias1 , and Paul
Buitelaar2
1
Dept. Ing. Sist. Telemáticos, Universidad Politécnica de Madrid,
{jfernando,cif}@gsi.dit.upm.es,
http://www.gsi.dit.upm.es
2
Insight, Centre for Data Analytics at National University of Ireland, Galway
{gabriela.vulcu,paul.buitelaar}@insight-centre.org,
http://insight-centre.org/
Abstract. Sentiment and Emotion Analysis strongly depend on quality
language resources, especially sentiment dictionaries. These resources are
usually scattered, heterogeneous and limited to specific domains of appli-
cation by simple algorithms. The EUROSENTIMENT project addresses
these issues by 1) developing a common language resource representation
model for sentiment analysis, and APIs for sentiment analysis services
based on established Linked Data formats (lemon, Marl, NIF and ONYX)
2) by creating a Language Resource Pool (a.k.a. LRP) that makes avail-
able to the community existing scattered language resources and services
for sentiment analysis in an interoperable way. In this paper we describe
the available language resources and services in the LRP and some sam-
ple applications that can be developed on top of the EUROSENTIMENT
LRP.
Keywords: Language Resources, Sentiment Analysis, Emotion Analy-
sis, Linked Data, Ontologies
1 Introduction
This paper reports our ongoing work in the European R&D project EUROSEN-
TIMENT, where we have created a multilingual Language Resource Pool (LRP)
for Sentiment Analysis based on a Linked Data approach for modelling linguistic
resources.
Sentiment Analysis requires language resources such as dictionaries that pro-
vide a sentiment or emotion value to each word. Just as words have different
meanings in different domains, the associated sentiment or emotion also varies.
Hence, every domain has its own dictionary. The information about what each
domain represents or how the entries for each domain are related is usually un-
documented or implied by the name of each dictionary. Moreover, it is common
that dictionaries from different providers use different representation formats.
Thus, it is very difficult to use different dictionaries at the same time.
2
In order to overcome these limitations, we have defined a Linked Data Model
for Sentiment and Emotion Analysis, which is based on the combination of sev-
eral vocabularies: the NLP Interchange Format (NIF) [1], to represent informa-
tion about texts, referencing text in the web with unique URIs; the Lexicon
Model for Ontologies (lemon) [2], to provide lexical information, and differen-
tiate between different domains and senses of a word; Marl [5], to link lexical
entries or senses with a sentiment; and Onyx [3], that adds emotive information.
The use of a semantic format not only eliminates the interoperability issue,
but it also makes information from other Linked Data sources available for the
sentiment analysis process. The EUROSENTIMENT LRP generates language
resources from legacy corpora, linking them with other Linked Data sources, and
shares this enriched version with other users.
In addition to language resources, the pool also offers access to sentiment
analysis services with a unified interface and data format. This interface builds on
the NIF Public API, adding several extra parameters that are used in Sentiment
Analysis. Results are formatted using JSON-LD and the same vocabularies as
for language resources. The NIF-compatible API allows for the aggregation of
results from different sources.
The project documentation3 contains further information about the EU-
ROSENTIMENT format, APIs and tools.
2 Language Resources
The EUROSENTIMENT LRP contains a set of language resources (lexicons and
corpora). The available EUROSENTIMENT language resources can be found
here.4 The user can see the domain and the language of each language resource.
At the moment the LRP contains resources for electronics and hotel domains in
six languages (Catalan, English, Spanish, French, Italian and Portuguese) and
we are currently working on adding more language resources from other domains
like telco, movies, food and music. Table 1 shows the number of reviews in each
available corpus and the number of lexical entries in each available lexicon.
A detailed description of the methodology for creating the domain-specific
sentiment lexicons and corpora to be added in the EUROSENTIMENT LRP
was presented at LREC 2014 [4].
The EUROSENTIMENT demonstrator5 shows how users can benefit from
the LRP, including an interactive SPARQL query editor to access the resources
and a faceted browser.
3 Sentiment Services
In addition to a model for language resources, EUROSENTIMENT also provides
an API for sentiment and emotion analysis services. Several already existing ser-
3
http://eurosentiment.readthedocs.org
4
http://portal.eurosentiment.eu/home_resources
5
http://eurosentiment.eu/demo
3
Lexicons
Corpora
Language Domains #Entities
Language Domains #Entities
German General 107417
English Hotel,Electronics 22373
English Hotel,Electronics 8660
Spanish Hotel,Electronics 18191
Spanish Hotel,Electronics 1041
Catalan Hotel,Electronics 4707
Catalan Hotel,Electronics 1358
Portuguese Hotel,Electronics 6244
Portuguese Hotel,Electronics 1387
French Electronics 22841
French Hotel,Electronics 651
Table 1. Summary of the resources in the LRP
vices in different languages have been adapted to expose this API. Any user can
benefit from these services, which are conveniently listed in the EUROSENTI-
MENT portal. At the moment, the following services are provided in several
languages: language detection, domain detection, sentiment and emotion detec-
tion, and text analysis.
Fig. 1. The LRP provides a list of available services
4 Applications Using the LRP
To demonstrate the capabilities of the EUROSENTIMENT LRP, we open-
sourced the code of several applications that make use of the services and re-
sources of the EUROSENTIMENT LRP. The applications are written in dif-
ferent programming languages and are thoroughly documented. Using these ap-
plications as a template, it is straightforward to immediately start consuming
the services and resources. The code can be found on the EUROSENTIMENT
Github repositories.6
6
http://github.com/eurosentiment
4
Fig. 2. Simple service that uses the resources in EUROSENTIMENT to analyse opin-
ions in different languages and domains
Acknowledgements
This work has been funded by the European project EUROSENTIMENT under
grant no. 296277
References
1. Hellmann, S., Lehmann, J., Auer, S., Nitzschke, M.: Nif combinator: Combining nlp
tool output. In: Knowledge Engineering and Knowledge Management, pp. 446–449.
Springer (2012)
2. McCrae, J., Spohr, D., Cimiano, P.: Linking lexical resources and ontologies on the
semantic web with lemon. In: Antoniou, G., Grobelnik, M., Simperl, E., Parsia, B.,
Plexousakis, D., De Leenheer, P., Pan, J. (eds.) The Semantic Web: Research and
Applications, Lecture Notes in Computer Science, vol. 6643, pp. 245–259. Springer
Berlin Heidelberg (2011)
3. Sánchez-Rada, J.F., Iglesias, C.A.: Onyx: Describing emotions on the web of data.
In: ESSEM@ AI* IA. pp. 71–82. Citeseer (2013)
4. Vulcu, G., Buitelaar, P., Negi, S., Pereira, B., Arcan, M., Coughlan, B., Sánchez-
Rada, J.F., Iglesias, C.A.: Generating Linked-Data based Domain-Specific Senti-
ment Lexicons from Legacy Language and Semantic Resources. In: th International
Workshop on EMOTION, SOCIAL SIGNALS, SENTIMENT & LINKED OPEN
DATA, co-located with LREC 2014,. LREC2014, Reykjavik, Iceland (May 2014)
5. Westerski, A., Iglesias, C.A., Tapia, F.: Linked Opinions: Describing Sentiments on
the Structured Web of Data. In: Proceedings of the 4th International Workshop
Social Data on the Web (2011)