1,721,015 research outputs found
An SVM Ensamble Approach to Detect Irony and Stereotype Spreaders on Twitter
Full text linkThe problem we address in this work is classifying whether a Twitter user has spread Irony and Stereotype or not. We used a text vectorization layer to generate Bag-Of-Words sequences. Then such sequences are passed to three different text classifiers (Decision Tree, Convolutional Neural Network, Naive Bayes). Our final classifier is an SVM. To test and validate our approach we used the dataset provided for the author profiling task organized by PAN@CLEF 2022. Our team (missino) submitted the predictions on the provided test set to participate at the shared task. Over several cross fold validation our approach was able to reach a maximum binary accuracy on the best validation split equal to 0.9474. On the test set provided for the shared task our model is able to reach an accuracy of 0.9389
Improving Irony and Stereotype Spreaders Detection using Data Augmentation and Convolutional Neural Network
Full text linkIn this paper we describe a deep learning model based on a Data Augmentation (DA) layer followed by a Convolutional Neural Network (CNN). The proposed model was developed by our team for the Profiling Irony and Stereotype Spreaders (ISSs) task proposed by the PAN 2022 organizers. As a first step, to classify an author as ISS or not (nISS), we developed a DA layer that expands each sample in the dataset provided. Using this augmented dataset we trained the CNN. Then, to submit our predictions, we apply our DA layer on the samples within the unlabeled test set too. Finally we fed our trained CNN with the augmented test set to generate our final predictions. To develop and test our model we used a 5-fold cross validation on the labelled training set. The proposed model reaches a maximum accuracy of 0.92 and an average accuracy of 0.89 over the five folds. Meanwhile, on the provided test set the proposed model reaches an accuracy of 0.9278
T100: A modern classic ensemble to profile irony and stereotype spreaders
Full text linkIn this work we propose a novel ensemble model based on deep learning and non-deep learning classifiers. The proposed model was developed by our team for participating at the Profiling Irony and Stereotype Spreaders (ISSs) task hosted at PAN@CLEF2022. Our ensemble (named T100), include a Logistic Regressor (LR) that classifies an author as ISS or not (nISS) considering the predictions provided by a first stage of classifiers. All these classifiers are able to reach state-of-the-art results on several text classification tasks. These classifiers (namely, the voters) are a Convolutional Neural Network (CNN), a Support Vector Machine (SVM), a Decision Tree (DT) and a Naive Bayes (NB) classifier. The voters are trained on the provided dataset and then generate predictions on the training set. Finally, the LR is trained on the predictions made by the voters. For the simulation phase the LR considers the predictions of the voters on the unlabelled test set to provide its final prediction on each sample. To develop and test our model we used a 5-fold cross validation on the labelled training set. Over the five validation splits, the proposed model achieves a maximum accuracy of 0.9342 and an average accuracy of 0.9158. As announced by the task organizers, the trained model presented here is able to reach an accuracy of 0.9444 on the unlabelled test set provided for the task
Spatio-temporal log-Gaussian Cox processes on eartquake events
No full textIn this paper we aim at studying some extensions of complex space-time models,
useful for the description of earthquake data. In particular we want to focus on the
Log-Gaussian Cox Process (LGCP) model estimation approach, with some results
on global informal diagnostics. Indeed, in our opinion the use of Cox processes that
are natural models for point process phenomena that are environmentally driven
could be a new approach for the description of seismic events. These models can
be useful in estimating the intensity surface of a spatio-temporal point process, in
constructing spatially continuous maps of earthquake risk from spatially discrete
data, and in real-time seismic activity surveillance. Moreover, covariate information
varying in space-time can be considered into the LGCP model, providing complex
models useful for a proper description of seismic events. LGCP is a Cox process with
a stochastic intensity function, depending on a Gaussian process. This construction
has some advantages, related to the multivariate Normal distribution features, since
the moment properties of the intensity function are inherited by the Cox process. In
particular, both estimation and diagnostics, can deal with some higherorder properties,
expressed for instance by the intensity and the pair correlation function of the
LGCP
Narratives and Counter-Narratives about Radicalization: Experiences of Moderation of an Online Communication Campaign
No full textThis paper presents the results of a cyber-ethnographic study. The research analyzes the dynamics that make,
calm and increase the radicalization narratives. This study is part of the Oltre project (ISF - DG Migration
and Home Affairs, EU) which directly involved 42 Italian and second generation youths in the dissemination
and moderation of an online communication campaign in order to prevent radicalized behaviors. This paper
illustrates how the young “moderators” interacted each other, highlighting how counter-narratives can
represent useful tools for deconstructing "complex" issues such as radicalization. Furthermore, the paper
shows (using social network analysis) how on the social media the communicative dynamics are influenced
by the characteristics of virtual networks that convey media messages. Finally, this study elucidates the
content analysis results, in order to compare narratives and counter-narratives, identifying different
meanings, specific vocabularies and relevant thematic clusters
Pratiche partecipative e educational commons per contrastare le disuguaglianze: il caso di studio del progetto SMOOTH in contesti multiculturali
No full textQuesto capitolo presenta i risultati preliminari del primo ciclo di attuazione di un caso di studio incluso nel progetto SMOOTH di Horizon 2020. L'obiettivo principale del progetto è quello di introdurre e studiare il paradigma emergente dei beni comuni educativi come sistema alternativo di valori e azioni per la promozione del dialogo interculturale e intergenerazionale e la creazione di spazi di cittadinanza democratica che supportino lo sviluppo delle comunità locali.This chapter presents the preliminary findings of the first round of implementation of a case study included in the Horizon 2020 project SMOOTH. The project's main objective is to introduce and study the emergent paradigm of the educational commons as an alternative system of values and actions for promoting intercultural and intergenerational dialogue and establishing spaces of democratic citizenship that support the development of local communities. 
Is text preprocessing still worth the time? A comparative survey on the influence of popular preprocessing methods on Transformers and traditional classifiers
Full text linkWith the advent of the modern pre-trained Transformers, the text preprocessing has started to be neglected and not specifically addressed in recent NLP literature. However, both from a linguistic and from a computer science point of view, we believe that even when using modern Transformers, text preprocessing can significantly impact on the performance of a classification model. We want to investigate and compare, through this study, how preprocessing impacts on the Text Classification (TC) performance of modern and traditional classification models. We report and discuss the preprocessing techniques found in the literature and their most recent variants or applications to address TC tasks in different domains. In order to assess how much the preprocessing affects classification performance, we apply the three top referenced preprocessing techniques (alone or in combination) to four publicly available datasets from different domains. Then, nine machine learning models – including modern Transformers – get the preprocessed text as input. The results presented show that an educated choice on the text preprocessing strategy to employ should be based on the task as well as on the model considered. Outcomes in this survey show that choosing the best preprocessing technique – in place of the worst – can significantly improve accuracy on the classification (up to 25%, as in the case of an XLNet on the IMDB dataset). In some cases, by means of a suitable preprocessing strategy, even a simple Naïve Bayes classifier proved to outperform (i.e., by 2% in accuracy) the best performing Transformer. We found that Transformers and traditional models exhibit a higher impact of the preprocessing on the TC performance. Our main findings are: (1) also on modern pre-trained language models, preprocessing can affect performance, depending on the datasets and on the preprocessing technique or combination of techniques used, (2) in some cases, using a proper preprocessing strategy, simple models can outperform Transformers on TC tasks, (3) similar classes of models exhibit similar level of sensitivity to text preprocessing
Detection of Hate Speech Spreaders using convolutional neural networks
Full text linkIn this paper we describe a deep learning model based on a Convolutional Neural Network (CNN). The model was developed for the Profiling Hate Speech Spreaders (HSSs) task proposed by PAN 2021 organizers and hosted at the 2021 CLEF Conference. Our approach to the task of classifying an author as HSS or not (nHSS) takes advantage of a CNN based on a single convolutional layer. In this binary classification task, on the tests performed using a 5-fold cross validation, the proposed model reaches a maximum accuracy of 0.80 on the multilingual (i.e., English and Spanish) training set, and a minimum loss value of 0.51 on the same set. As announced by the task organizers, the trained model presented is able to reach an overall accuracy of 0.79 on the full test set. This overall accuracy is obtained averaging the accuracy achieved by the model on both languages. In particular, with regard to the Spanish test set, the organizers announced that our model achieves an accuracy of 0.85, while on the English test set the same model achieved - as announced by the organizers too - an accuracy of 0.73. Thanks to the model presented in this paper, our team won the 2021 PAN competition on profiling HSSs
WhoSNext: Recommending Twitter Users to Follow Using a Spreading Activation Network Based Approach
No full textThe huge number of modern social network users has made the web a fertile ground for the growth and development of a plethora of recommender systems. To date, recommending a new user profile X to a given user U that could be interested in creating a relationship with X has been tackled using techniques based on content analysis, existing friendship relationships and other pieces of information coming from different social networks or websites. In this paper we propose a recommending architecture-called WhoSNext (WSN)-tested on Twitter and which aim is promoting the creation of new relationships among users. As recent researches show, this is an interesting recommendation problem: for a given user U, find which other user might be proposed to U as a new friend. Instead of conducting a study based on a semantic approach (e.g. analyzing tweet content), the proposed algorithm exploits a graph created from a set of Twitter users called seeds. In this work-and, to the best of our knowledge, for the first time-this issue is addressed using only user ID for building a particular Spreading Activation Network. This network was firstly trained and then tested on a set consisting of over 400,000 real users. Experimental results show that this approach outperforms the results obtained from many well-known state-of-the-art systems, which are much more expensive in terms of either data preprocessing or computational resources
Text Enrichment with Japanese Language to Profile Cryptocurrency Influencers
Full text linkFrom a few-shot learning perspective, we propose a strategy to enrich the latent semantic of the text provided in the dataset provided for the Profiling Cryptocurrency Influencers with Few-shot Learning, the task hosted at PAN@CLEF2023. Our approach is based on data augmentation using the backtranslation forth and back to and from Japanese language. We translate samples in the original training dataset to a target language (i.e. Japanese). Then we translate it back to English. The original sample and the backtranslated one are then merged. Then we fine-tuned two state-of-the-art Transformer models on this augmented version of the training dataset. We evaluate the performance of the two fine-tuned models using the Macro and Micro F1 accordingly to the official metric used for the task. After the fine-tuning phase, ELECTRA and XLNet obtained a Macro F1 of 0.7694 and 0.7872 respectively on the original training set. Our best submission obtained a Macro F1 equal to 0.3851 on the official test set provided
- …
