CROSSLINGMIND

Automated analysis of opinions in a multilingual context

 Partecipanti

Mappa

 Word cloud

Esplora la "nuvola delle parole (Word Cloud) per avere un'idea di massima del progetto.

 Obiettivo del progetto (Objective)

'In the recent years, massive amounts of personal opinions have been posted on the Internet. This allows the companies and institutions to get feedback about their services, provided that they are able to analyse this mass of information.Given that most people prefer to write and read information in their own language, it is clearly relevant to be able to analyse in one language opinions expressed in another language. This is an interdisciplinary topic, called cross-lingual opinion analysis, which involves opinion analysis and machine translation. Although both topics involved have reached a high maturity level, cross-lingual opinion analysis has been little explored. Most of the work done in this field aimed at leveraging resources in English to perform monolingual opinion analysis in another language. The objective of this project is to develop a cross-lingual opinion analysis system. To achieve this goal, the fellow will apply his knowledge in machine translation to adapt the monolingual opinion analysis system existing at the host institution. The fellow will develop a novel approach to map across languages concepts such as the opinion target entity, the corresponding attribute and its polarity. This approach consists of enriching the machine translation system with relevant contextual information. Furthermore several possible architectures of the system will be evaluated and finally a demonstrator will be developed. The underlying technology will eventually be transferred to the industrial sector, since this is one of the missions of the host institution. Thus the fellow will gain interdisciplinary and inter-sectoral competences. In addition, European companies will be able to improve their competitiveness. The citizens will be the final beneficiary since companies and public institutions will be able to better take their opinions into account.'

Introduzione (Teaser)

The incredible volume of personal opinions posted on the Internet is helping companies and institutions garner feedback on their services. However, this mass of information is not always in just one language, posing a problem regarding its analysis and processing.

Descrizione progetto (Article)

The 'Automated analysis of opinions in a multilingual context' (http://glicom.upf.edu/lambert/crosslingmind/ (CROSSLINGMIND)) project was established to develop a cross-lingual opinion mining (CLOM) system, applying knowledge in machine translation to adapt an existing monolingual opinion analysis system.

The existing monolingual opinion mining pipeline was adapted to the type of content generated by social media users and to languages of high interest for potential industrial partners. Related work resulted in the publication of 'Selection of correction candidates for the normalization of Spanish user-generated content' in the journal Natural Language Engineering. An opinion mining pipeline was also built for the Portuguese language.

At the same time, CROSSLINGMIND conducted an in-depth study of the state of the art in CLOM, with particular emphasis on cross-lingual algorithms and applications used in natural language processing. The project also generated a lexicon of sentiment words (the LIWC lexicon) in Catalan via triangulation from similar lexicons in other languages.

CROSSLINGMIND developed an aspect-based CLOM system that performs an analysis at the level of the aspects of the entities about which opinions are expressed. Trained on data from the Seventh Framework Programme (FP7) OPENER project, the system was annotated at the opinionated unit level, in several languages, in the hotel review domain.

The various activities and deliverables have been presented at conferences and developed as tutorials, courses and special journal issues.

The project-developed system proved that aspect-based CLOM is achievable and can return competitive classification results. It has been designed for use in real-life settings and has the potential to be part of a technology transfer framework. This development is slated to help improve the competitiveness of European companies, organisations and institutions that rely on general public opinion to effect changes and growth. It will also break down language barriers hindering communication and fragmenting exploitable data and the digital market. Eventually, the general public should also be able to use it.