Skip to main content

AMIC-UPV: ANALISIS AFECTIVO DE INFORMACION MULTIMEDIA CON COMUNICACION INCLUSIVA Y NATURAL

TIN2017-85854-C4-2-R

•
Funding agency name Agencia Estatal de Investigación
Funding agency acronym AEI
Program Programa Estatal de I+D+i Orientado a los Retos de la Sociedad
Call Proyectos I+D+i Retos de Investigación
Call year 2017
State Plan 2017-2020 Plan Estatal de Investigación Científica y Técnica y de Innovación
Beneficiary centre UNIVERSITAT POLITÈCNICA DE VALÈNCIA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 21
Encontrada(s) 1 página(s)

RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Spanish sentiment analysis in Twitter at the TASS workshop

  • Pla Santamaría, Ferran
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] This paper describes a support vector machine-based approach to different tasks related to sentiment analysis in Twitter for Spanish. We focus on parameter optimization of the models and the combination of several models by means of voting techniques. We evaluate the proposed approach in all the tasks that were defined in the five editions of the TASS workshop, between 2012 and 2016. TASS has become a framework for sentiment analysis tasks that are focused on the Spanish language. We describe our participation in this competition and the results achieved, and then we provide an analysis of and comparison with the best approaches of the teams who participated in all the tasks defined in the TASS workshops. To our knowledge, our results exceed those published to date in the sentiment analysis tasks of the TASS workshops., This work has been partially funded by the Spanish MINECO and FEDER founds under project ASLP-MULAN: Audio, Speech and Language Processing for Multimedia Analytics, TIN2014-54288-C4-3-R.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Self-attention for Twitter sentiment analysis in Spanish

  • González-Barba, José Ángel
  • Pla Santamaría, Ferran
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] This paper describes our proposal for Sentiment Analysis in Twitter for the Spanish language. The main characteristics of the system are the use of word embedding specifically trained from tweets in Spanish and the use of self-attention mechanisms that allow to consider sequences without using convolutional nor recurrent layers. These self-attention mechanisms are based on the encoders of the Transformer model. The results obtained on the Task 1 of the TASS 2019 workshop, for all the Spanish variants proposed, support the correctness and adequacy of our proposal., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R) and the GiSPRO project (PROMETEU/2018/176). Work of Jose-Angel Gonzalez is financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Choosing the right loss function for multi-label Emotion Classification

  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
  • González-Barba, José Ángel
  • Pla Santamaría, Ferran
[EN] Natural Language Processing problems has recently been benefited for the advances in Deep Learning. Many of these problems can be addressed as a multi-label classification problem. Usually, the metrics used to evaluate classification models are different from the loss functions used in the learning process. In this paper, we present a strategy to incorporate evaluation metrics in the learning process in order to increase the performance of the classifier according to the measure we are interested to favor. Concretely, we propose soft versions of the Accuracy, micro-F-1, and macro-F-1 measures that can be used as loss functions in the back-propagation algorithm. In order to experimentally validate our approach, we tested our system in an Emotion Classification task proposed at the International Workshop on Semantic Evaluation, SemEval-2018. Using a Convolutional Neural Network trained with the proposed loss functions we obtained significant improvements both for the English and the Spanish corpora., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R) and the GiSPRO project (PROMETEU/2018/176). Work of Jose-Angel Gonzalez is also financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

The NoisyOffice Database: A Corpus To Train Supervised Machine Learning Filters For Image Processing

  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
  • España Boquera, Salvador|||0000-0002-0335-9997
  • Pastor Pellicer, Joan
  • ZAMORA MARTÍNEZ, FRANCISCO JULIÁN
[EN] This paper presents the `NoisyOffice" database. It consists of images of printed text documents with noise mainly caused by uncleanliness from a generic office, such as coffee stains and footprints on documents or folded and wrinkled sheets with degraded printed text. This corpus is intended to train and evaluate supervised learning methods for cleaning, binarization and enhancement of noisy images of grayscale text documents. As an example, several experiments of image enhancement and binarization are presented by using deep learning techniques. Also, double-resolution images are also provided for testing super-resolution methods. The corpus is freely available at UCI Machine Learning Repository. Finally, a challenge organized by Kaggle Inc. to denoise images, using the database, is described in order to show its suitability for benchmarking of image processing systems., This research was undertaken as part of the project TIN2017-85854-C4-2-R, jointly funded by the Spanish MINECO and FEDER founds.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Research data Set de datos (Dataset)

UX-comments: Evaluation of User eXperience in E-learning [Dataset]

  • Sanchis Font, Rosario
  • González Barba, José Ángel
  • Pla Santamaría, Ferran
  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
Virtual learning environments are growing in importance as fast as e-learning, which is becoming highly demanded by universities and students worldwide. We have investigated how to automatically evaluate User eXperience in this domain using sentiment analysis techniques. For this purpose, the UX-comments corpus has been built with the opinions of 583 users (107 English speakers and 476 Spanish speakers) about three learning management systems in different courses. All the collected opinions were manually labeled with polarity information (P=positive, N=negative, or NEU=neutral) by three human annotators, both at the whole opinion and sentence levels. Dataset Information P NEU N Total Spanish Observations 338 53 85 476 Spanish Sentences 404 41 142 587 English Observations 56 21 30 107 English Sentences 90 14 80 184, Special thanks to the following biomedical organizations: Fundación IVI and Medigene Press S.L.; both have provided data from their Master and Posgraduate Courses through the academic stay research of Rosario Sanchis-Font, during 2017 and 2018. Many thanks to Carlos Turró-Ribalta and Ignacio Despujol-Zabala for supporting this research with data from UPV MOOCs.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Contribución de congreso (ConferenceOutput)

ELiRF-UPV at TASS 2020: TWilBERT for Sentiment Analysis and Emotion Detection in Spanish Tweets

  • González-Barba, José Ángel
  • Arias-Moncho, José
  • Pla Santamaría, Ferran
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] This paper describes the participation of the ELiRF research group of the Universitat Politècnica de València in the TASS 2020 Workshop, framed within the XXXVI edition of the International Conference of the Spanish Society for the Processing of Natural Language (SEPLN). We present the approach used for the Monolingual Sentiment Analysis and Emotion Detection tasks of the workshop, as well as the results obtained. Our participation has focused mainly on employing an adaptation of BERT for text classification on the Twitter domain and the Spanish language. This system, that we have called TWilBERT, shown systematic improvements of the state of the art in almost all the tasks framed in the SEPLN conference of previous years, and also obtains the most competitive performance in the tasks addressed in this work., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R) and by the GiSPRO project (PROMETEU/2018/176). Work of José-Ángel González is financed by Universitat Politècnica de València under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Summarization of Spanish Talk Shows with Siamese Hierarchical Attention Networks

  • González-Barba, José Ángel
  • Segarra Soriano, Encarnación
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
  • García-Granada, Fernando|||0000-0003-2213-4213
  • Sanchís Arnal, Emilio|||0000-0002-6737-4723
[EN] In this paper, we present an approach to Spanish talk shows summarization. Our approach is based on the use of Siamese Neural Networks on the transcription of the show audios. Specifically, we propose to use Hierarchical Attention Networks to select the most relevant sentences for each speaker about a given topic in the show, in order to summarize his opinion about the topic. We train these networks in a siamese way to determine whether a summary is appropriate or not. Previous evaluation of this approach on summarization task of English newspapers achieved performances similar to other state-of-the-art systems. In the absence of enough transcribed or recognized speech data to train our system for talk show summarization in Spanish, we acquire a large corpus of document-summary pairs from Spanish newspapers and we use it to train our system. We choose this newspapers domain due to its high similarity with the topics addressed in talk shows. A preliminary evaluation of our summarization system on Spanish TV programs shows the adequacy of the proposal., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R). Work of Jose-Angel Gonzalez is financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Applying Siamese Hierarchical Attention Neural Networks for multi-document summarization

  • González-Barba, José Ángel
  • Julien Delonca
  • Segarra Soriano, Encarnación
  • Sanchís Arnal, Emilio|||0000-0002-6737-4723
  • García-Granada, Fernando|||0000-0003-2213-4213
[EN] In this paper, we present an approach to multi-document summarization based on Siamese Hierarchical Attention Neural Networks. The attention mechanism of Hierarchical Attention Networks, provides a score to each sentence in function of its relevance in the classification process. For the summarization process, only the scores of sentences are used to rank them and select the most salient sentences. In this work we explore the adaptability of this model to the problem of multi-document summarization (typically very long documents where the straightforward application of neural networks tends to fail). The experiments were carried out using the CNN/DailyMail as training corpus, and the DUC-2007 as test corpus. Despite the difference between training set (CNN/DailyMail) and test set (DUC-2007) characteristics, the results show the adequacy of this approach to multi-document summarization., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R). Work of Jose-Angel Gonzalez is also financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Extractive summarization using siamese hierarchical transformer encoders

  • González-Barba, José Ángel
  • Segarra Soriano, Encarnación
  • García-Granada, Fernando|||0000-0003-2213-4213
  • Sanchís Arnal, Emilio|||0000-0002-6737-4723
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] In this paper, we present an extractive approach to document summarization, the Siamese Hierarchical Transformer Encoders system, that is based on the use of siamese neural networks and the transformer encoders which are extended in a hierarchical way. The system, trained for binary classification, is able to assign attention scores to each sentence in the document. These scores are used to select the most relevant sentences to build the summary. The main novelty of our proposal is the use of self-attention mechanisms at sentence level for document summarization, instead of using only attentions at word level. The experimentation carried out using the CNN/DailyMail summarization corpus shows promising results in-line with the state-of-the-art., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R). Work of Jose Angel Gonzalez is also financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Fallback Variable History NNLMs: Efficient NNLMs by precomputation and stochastic training

  • Zamora Martínez, Francisco Julián
  • Palacios Corella
  • España Boquera, Salvador|||0000-0002-0335-9997
  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
[EN] This paper presents a new method to reduce the computational cost when using Neural Networks as Language Models, during recognition, in some particular scenarios. It is based on a Neural Network that considers input contexts of different length in order to ease the use of a fallback mechanism together with the precomputation of softmax normalization constants for these inputs. The proposed approach is empirically validated, showing their capability to emulate lower order N-grams with a single Neural Network. A machine translation task shows that the proposed model constitutes a good solution to the normalization cost of the output softmax layer of Neural Networks, for some practical cases, without a significant impact in performance while improving the system speed., This work was partially supported by the Spanish MINECO and FEDER founds under project TIN2017-85854-C4-2-R (to MJCB). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

NASca and NASes: Two Monolingual Pre-Trained Models for Abstractive Summarization in Catalan and Spanish

  • Ahuir-Esteve, Vicent|||0000-0001-5636-651X
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
  • González-Barba, José Ángel
  • Segarra Soriano, Encarnación
[EN] Most of the models proposed in the literature for abstractive summarization are generally suitable for the English language but not for other languages. Multilingual models were introduced to address that language constraint, but despite their applicability being broader than that of the monolingual models, their performance is typically lower, especially for minority languages like Catalan. In this paper, we present a monolingual model for abstractive summarization of textual content in the Catalan language. The model is a Transformer encoder-decoder which is pretrained and fine-tuned specifically for the Catalan language using a corpus of newspaper articles. In the pretraining phase, we introduced several self-supervised tasks to specialize the model on the summarization task and to increase the abstractivity of the generated summaries. To study the performance of our proposal in languages with higher resources than Catalan, we replicate the model and the experimentation for the Spanish language. The usual evaluation metrics, not only the most used ROUGE measure but also other more semantic ones such as BertScore, do not allow to correctly evaluate the abstractivity of the generated summaries. In this work, we also present a new metric, called content reordering, to evaluate one of the most common characteristics of abstractive summaries, the rearrangement of the original content. We carried out an exhaustive experimentation to compare the performance of the monolingual models proposed in this work with two of the most widely used multilingual models in text summarization, mBART and mT5. The experimentation results support the quality of our monolingual models, especially considering that the multilingual models were pretrained with many more resources than those used in our models. Likewise, it is shown that the pretraining tasks helped to increase the degree of abstractivity of the generated summaries. To our knowledge, this is the first work that explores a monolingual approach for abstractive summarization both in Catalan and Spanish., This work was partially supported by the Spanish Ministerio de Ciencia, Innovacion y Universidades and FEDER founds under the project AMIC (TIN2017-85854-C4-2-R), and by the Agencia Valenciana de la Innovacio (AVI) of the Generalitat Valenciana under the GUAITA (INNVA1/2020/61) project.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

TWilBert: Pre-trained deep bidirectional transformers for Spanish Twitter

  • González-Barba, José Ángel
  • Pla Santamaría, Ferran
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] In recent years, the Natural Language Processing community have been moving from uncontextualized word embeddings towards contextualized word embeddings. Among these contextualized architectures, BERT stands out due to its capacity to compute bidirectional contextualized word representations. However, its competitive performance in English downstream tasks is not obtained by its multilingual version when it is applied to other languages and domains. This is especially true in the case of the Spanish language used in Twitter. In this work, we propose TWiLBERT, a specialization of BERT architecture both for the Spanish language and the Twitter domain. Furthermore, we propose a Reply Order Prediction signal to learn inter-sentence coherence in Twitter conversations, which improves the performance of TWilBERT in text classification tasks that require reasoning on sequences of tweets. We perform an extensive evaluation of TWilBERT models on 14 different text classification tasks, such as irony detection, sentiment analysis, or emotion detection. The results obtained by TWilBERT outperform the state-of-the-art systems and Multilingual BERT. In addition, we carry out a thorough analysis of the TWilBERT models to study the reasons of their competitive behavior. We release the pre-trained TWilBERT models used in this paper, along with a framework for training, evaluating, and fine-tuning TWilBERT models., This work has been partially supported by the Spanish Ministerio de Ciencia, Innovacion y Universidades and FEDER founds under project AMIC (TIN2017-85854-C4-2-R), and the Generalitat Valenciana under GiSPRO (PROMETEU/2018/176) and GUAITA (INNVA1/2020/61) projects. Work of Jose Angel Gonzalez is financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Efficient Embedded Decoding of Neural Network Language Models in a Machine Translation System

  • Zamora Martínez, Francisco Julián
  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
[EN] Neural Network Language Models (NNLMs) are a successful approach to Natural Language Processing tasks, such as Machine Translation. We introduce in this work a Statistical Machine Translation (SMT) system which fully integrates NNLMs in the decoding stage, breaking the traditional approach based on n-best list rescoring. The neural net models (both language models (LMs) and translation models) are fully coupled in the decoding stage, allowing to more strongly influence the translation quality. Computational issues were solved by using a novel idea based on memorization and smoothing of the softmax constants to avoid their computation, which introduces a trade-off between LM quality and computational cost. These ideas were studied in a machine translation task with different combinations of neural networks used both as translation models and as target LMs, comparing phrase-based and N-gram-based systems, showing that the integrated approach seems more promising for N-gram-based systems, even with nonfull-quality NNLMs., This work was partially supported by the Spanish MINECO and FEDER found under project TIN2017-85854-C4-2-R.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Cross-Domain Polarity Models to Evaluate User eXperience in E-learning

  • Sanchis-Font, Rosario
  • González-Barba, José Ángel
  • Pla Santamaría, Ferran
  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] Virtual learning environments are growing in importance as fast as e-learning is becoming highly demanded by universities and students all over the world. This paper investigates how to automatically evaluate User eXperience in this domain using sentiment analysis techniques. For this purpose, a corpus with the opinions given by a total of 583 users (107 English speakers and 476 Spanish speakers) about three learning management systems in different courses has been built. All the collected opinions were manually labeled with polarity information (positive, negative or neutral) by three human annotators, both at the whole opinion and sentence levels. We have applied our state-of-the-art sentiment analysis models, trained with a corpus of a different semantic domain (a Twitter corpus), to study the use of cross-domain models for this task. Cross-domain models based on deep neural networks (convolutional neural networks, transformer encoders and attentional BLSTM models) have been tested. In order to contrast our results, three commercial systems for the same task (MeaningCloud, Microsoft Text Analytics and Google Cloud) were also tested. The obtained results are very promising and they give an insight to keep going the research of applying sentiment analysis tools on User eXperience evaluation. This is a pioneering idea to provide a better and accurate understanding on human needs in the interaction with virtual learning environments and a step towards the development of automatic tools that capture the feed-back of user perception for designing virtual learning environments centered in user's emotions, beliefs, preferences, perceptions, responses, behaviors and accomplishments that occur before, during and after the interaction., Partially supported by the Spanish MINECO and FEDER founds under Project TIN2017-85854-C4-2-R. Work of J.A. Gonzalez is financed under Grant PAID-01-17
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

ALBAYZIN Query-by-example Spoken Term Detection 2016 evaluation

  • Tejedor, Javier
  • Toledano, Doroteo T.
  • Lopez-Otero, Paula
  • Docio-Fernandez, Laura
  • Proença, Jorge
  • Perdigão, Fernando
  • Pompili, Anna
  • Abad, Alberto
  • García-Granada, Fernando|||0000-0003-2213-4213
  • Sanchís Arnal, Emilio|||0000-0002-6737-4723
[EN] Query-by-example Spoken Term Detection (QbE STD) aims to retrieve data from a speech repository given an acoustic (spoken) query containing the term of interest as the input. This paper presents the systems submitted to the ALBAYZIN QbE STD 2016 Evaluation held as a part of the ALBAYZIN 2016 Evaluation Campaign at the IberSPEECH 2016 conference. Special attention was given to the evaluation design so that a thorough post-analysis of the main results could be carried out. Two different Spanish speech databases, which cover different acoustic and language domains, were used in the evaluation: the MAVIR database, which consists of a set of talks from workshops, and the EPIC database, which consists of a set of European Parliament sessions in Spanish. We present the evaluation design, both databases, the evaluation metric, the systems submitted to the evaluation, the results, and a thorough analysis and discussion. Four different research groups participated in the evaluation, and a total of eight template matching-based systems were submitted. We compare the systems submitted to the evaluation and make an in-depth analysis based on some properties of the spoken queries, such as query length, single-word/multi-word queries, and in-language/out-of-language queries., This work was partially supported by Fundacao para a Ciencia e Tecnologia (FCT) under the projects UID/EEA/50008/2013 (pluriannual funding in the scope of the LETSREAD project) and UID/CEC/50021/2013, and Grant SFRH/BD/97187/2013. Jorge Proenca is supported by the SFRH/BD/97204/2013 FCT Grant. This work was also supported by the Galician Government ('Centro singular de investigacion de Galicia' accreditation 2016-2019 ED431G/01 and the research contract GRC2014/024 (Modalidade: Grupos de Referencia Competitiva 2014)), the European Regional Development Fund (ERDF), the projects "DSSL: Redes Profundas y Modelos de Subespacios para Deteccion y Seguimiento de Locutor, Idioma y Enfermedades Degenerativas a partir de la Voz" (TEC2015-68172-C2-1-P) and the TIN2015-64282-R funded by Ministerio de Economia y Competitividad in Spain, the Spanish Government through the project "TraceThem" (TEC2015-65345-P), and AtlantTIC ED431G/04.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Siamese hierarchical attention networks for extractive summarization

  • González-Barba, José Ángel
  • Segarra Soriano, Encarnación
  • García-Granada, Fernando|||0000-0003-2213-4213
  • Sanchís Arnal, Emilio|||0000-0002-6737-4723
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] In this paper, we present an extractive approach to document summarization based on Siamese Neural Networks. Specifically, we propose the use of Hierarchical Attention Networks to select the most relevant sentences of a text to make its summary. We train Siamese Neural Networks using document-summary pairs to determine whether the summary is appropriated for the document or not. By means of a sentence-level attention mechanism the most relevant sentences in the document can be identified. Hence, once the network is trained, it can be used to generate extractive summaries. The experimentation carried out using the CNN/DailyMail summarization corpus shows the adequacy of the proposal. In summary, we propose a novel end-to-end neural network to address extractive summarization as a binary classification problem which obtains promising results in-line with the state-of-the-art on the CNN/DailyMail corpus., This work has been partially supported by the Spanish MINECO and FEDER founds under project AMIC (TIN2017-85854-C4-2-R). Work of Jose-Angel Gonzalez is also financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Towards a Universal Semantic Dictionary

  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
  • Iklódi, E.
  • Recski, G.
  • Borbély, G.
[EN] A novel method for finding linear mappings among word embeddings for several languages, taking as pivot a shared, multilingual embedding space, is proposed in this paper. Previous approaches learned translation matrices between two specific languages, while this method learns translation matrices between a given language and a shared, multilingual space. The system was first trained on bilingual, and later on multilingual corpora as well. In the first case, two different training data were applied: Dinu¿s English¿Italian benchmark data, and English¿Italian translation pairs extracted from the PanLex database. In the second case, only the PanLex database was used. The system performs on English¿Italian languages with the best setting significantly better than the baseline system given by Mikolov, and it provides a comparable performance with more sophisticated systems. Exploiting the richness of the PanLex database, the proposed method makes it possible to learn linear mappings among an arbitrary number of languages., This research was funded by Spanish MINECO and FEDER grant number TIN2017-85854-C4-2-R.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Transformer based contextualization of pre-trained word embeddings for irony detection in Twitter

  • González-Barba, José Ángel
  • Pla Santamaría, Ferran
  • Hurtado Oliver, Lluis Felip|||0000-0002-1877-0455
[EN] Human communication using natural language, specially in social media, is influenced by the use of figurative language like irony. Recently, several workshops are intended to explore the task of irony detection in Twitter by using computational approaches. This paper describes a model for irony detection based on the contextualization of pre-trained Twitter word embeddings by means of the Transformer architecture. This approach is based on the same powerful architecture as BERT but, differently to it, our approach allows us to use in-domain embeddings. We performed an extensive evaluation on two corpora, one for the English language and another for the Spanish language. Our system was the first ranked system in the Spanish corpus and, to our knowledge, it has achieved the second-best result on the English corpus. These results support the correctness and adequacy of our proposal. We also studied and interpreted how the multi-head self-attention mechanisms are specialized on detecting irony by means of considering the polarity and relevance of individual words and even the relationships among words. This analysis is a first step towards understanding how the multi-head self-attention mechanisms of the Transformer architecture address the irony detection problem., This work has been partially supported by the Spanish Ministerio de Ciencia, Innovacion y Universidades and FEDER founds under project AMIC (TIN2017-85854-C4-2-R) and the GiSPRO project (PROMETEU/2018/176). Work of Jose-Angel Gonzalez is financed by Universitat Politecnica de Valencia under grant PAID-01-17.
RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication Artículo científico (JournalArticle)

Handwriting recognition by using deep learning to extract meaningful features

  • Pastor Pellicer, Joan
  • Zamora-Martinez, Francisco Julián
  • Castro-Bleda, Maria Jose|||0000-0003-1001-8258
  • España Boquera, Salvador|||0000-0002-0335-9997
[EN] Recent improvements in deep learning techniques show that deep models can extract more meaningful data directly from raw signals than conventional parametrization techniques, making it possible to avoid specific feature extraction in the area of pattern recognition, especially for Computer Vision or Speech tasks. In this work, we directly use raw text line images by feeding them to Convolutional Neural Networks and deep Multilayer Perceptrons for feature extraction in a Handwriting Recognition system. The proposed recognition system, based on Hidden Markov Models that are hybridized with Neural Networks, has been tested with the IAM Database, achieving a considerable improvement., Work partially supported by the Spanish MINECO and FEDER founds under project TIN2017-85854-C4-2-R.
RUA. Repositorio Institucional de la Universidad de Alicante
Publication Otros (Other)

Applying Siamese Hierarchical Attention Neural Networks for multi-document summarization, Aplicando Redes Neuronales Siamesas con Atención Jerárquica para resúmenes multi-documento

  • González, José Angel
  • Delonca, Julien
  • Sanchis Arnal, Emilio
  • García Granada, Fernando
  • Segarra Soriano, Encarnación
In this paper, we present an approach to multi-document summarization based on Siamese Hierarchical Attention Neural Networks. The attention mechanism of Hierarchical Attention Networks, provides a score to each sentence in function of its relevance in the classification process. For the summarization process, only the scores of sentences are used to rank them and select the most salient sentences. In this work we explore the adaptability of this model to the problem of multi-document summarization (typically very long documents where the straightforward application of neural networks tends to fail). The experiments were carried out using the CNN/DailyMail as training corpus, and the DUC-2007 as test corpus. Despite the difference between training set (CNN/DailyMail) and test set (DUC-2007) characteristics, the results show the adequacy of this approach to multi-document summarization.
Zaguán. Repositorio Digital de la Universidad de Zaragoza
Publication Artículo científico (JournalArticle)

AMIC: Affective multimedia analytics with inclusive and natural communication

  • Ortega, A.
  • Lleida, E.
  • San-Segundo, R.
  • Ferreiros, J.
  • Hurtado, L.
  • Sanchis, E.
  • Torres, M.I.
  • Justo, R.
Traditionally, textual content has been the main source of information extraction and indexing, and other technologies that are capable of extracting information from the audio and video of multimedia documents have joined later. Other major axis of analysis is the emotional and affective aspect intrinsic in human communication. This information of emotions, stances, preferences, figurative language, irony, sarcasm, etc. is fundamental and irreplaceable for a complete understanding of the content in conversations, speeches, debates, discussions, etc. The objective of this project is focused on advancing, developing and improving speech and language technologies as well as image and video technologies in the analysis of multimedia content adding to this analysis the extraction of affective-emotional information. As additional steps forward, we will advance in the methodologies and ways for presenting the information to the user, working on technologies for language simplification, automatic reports and summary generation, emotional speech synthesis and natural and inclusive interaction.