An End-to-End Set Transformer for User-Level Classification of Depression and Gambling Disorder

This work proposes a transformer architecture for user-level classification of gambling addiction and depression that is trainable end-to-end. As opposed to other methods that operate at the post level, we process a set of social media posts from a particular individual, to make use of the interactions between posts and eliminate label noise at the post level. We exploit the fact that, by not injecting positional encodings, multi-head attention is permutation invariant and we process randomly sampled sets of texts from a user after being encoded with a modern pretrained sentence encoder (RoBERTa / MiniLM). Moreover, our architecture is interpretable with modern feature attribution methods and allows for automatic dataset creation by identifying discriminating posts in a user's text-set. We perform ablation studies on hyper-parameters and evaluate our method for the eRisk 2022 Lab on early detection of signs of pathological gambling and early risk detection of depression. The method proposed by our team BLUE obtained the best ERDE5 score of 0.015, and the second-best ERDE50 score of 0.009 for pathological gambling detection. For the early detection of depression, we obtained the second-best ERDE50 of 0.027.

Detecting early signs of depression in the conversational domain: The role of transfer learning in low-resource scenarios

The high prevalence of depression in society has given rise to the need for new digital tools to assist in its early detection. To this end, existing research has mainly focused on detecting depression in the domain of social media, where there is a sufficient amount of data. However, with the rise of conversational agents like Siri or Alexa, the conversational domain is becoming more critical. Unfortunately, there is a lack of data in the conversational domain. We perform a study focusing on domain adaptation from social media to the conversational domain. Our approach mainly exploits the linguistic information preserved in the vector representation of text. We describe transfer learning techniques to classify users who suffer from early signs of depression with high recall. We achieve state-of-the-art results on a commonly used conversational dataset, and we highlight how the method can easily be used in conversational agents. We publicly release all source code.

Unsupervised Ranking and Aggregation of Label Descriptions for Zero-Shot Classifiers

Zero-shot text classifiers based on label descriptions embed an input text and a set of labels into the same space: measures such as cosine similarity can then be used to select the most similar label description to the input text as the predicted label. In a true zero-shot setup, designing good label descriptions is challenging because no development set is available. Inspired by the literature on Learning with Disagreements, we look at how probabilistic models of repeated rating analysis can be used for selecting the best label descriptions in an unsupervised fashion. We evaluate our method on a set of diverse datasets and tasks (sentiment, topic and stance). Furthermore, we show that multiple, noisy label descriptions can be aggregated to boost the performance.

UPV at TREC Health Misinformation Track 2021 Ranking with SBERT and Quality Estimators

Health misinformation on search engines is a significant problem that could negatively affect individuals or public health. To mitigate the problem, TREC organizes a health misinformation track. This paper presents our submissions to this track. We use a BM25 and a domain-specific semantic search engine for retrieving initial documents. Later, we examine a health news schema for quality assessment and apply it to re-rank documents. We merge the scores from the different components by using reciprocal rank fusion. Finally, we discuss the results and conclude with future works.

UPV at CheckThat! 2021: Mitigating Cultural Differences for Identifying Multilingual Check-worthy Claims

Identifying check-worthy claims is often the first step of automated fact-checking systems. Tackling this task in a multilingual setting has been understudied. Encoding inputs with multilingual text representations could be one approach to solve the multilingual check-worthiness detection. However, this approach could suffer if cultural bias exists within the communities on determining what is check-worthy.In this paper, we propose a language identification task as an auxiliary task to mitigate unintended bias.With this purpose, we experiment joint training by using the datasets from CLEF-2021 CheckThat!, that contain tweets in English, Arabic, Bulgarian, Spanish and Turkish. Our results show that joint training of language identification and check-worthy claim detection tasks can provide performance gains for some of the selected languages.

Surveying the Research on Fake News in Social Media: a Tale of Networks and Language

The history of journalism and news diffusion is tightly coupled with the effort to dispel hoaxes, misinformation, propaganda, unverified rumours, poor reporting, and messages containing hate and divisions. With the explosive growth of online social media and billions of individuals engaged with consuming, creating, and sharing news, this ancient problem has surfaced with a renewed intensity threatening our democracies, public health, and news outlets credibility. This has triggered many researchers to develop new methods for studying, understanding, detecting, and preventing fake-news diffusion; as a consequence, thousands of scientific papers have been published in a relatively short period, making researchers of different disciplines to struggle in search of open problems and most relevant trends. The aim of this survey is threefold: first, we want to provide the researchers interested in this multidisciplinary and challenging area with a network-based analysis of the existing literature to assist them with a visual exploration of papers that can be of interest; second, we present a selection of the main results achieved so far adopting the network as an unifying framework to represent and make sense of data, to model diffusion processes, and to evaluate different debunking strategies. Finally, we present an outline of the most relevant research trends focusing on the moving target of fake-news, bots, and trolls identification by means of data mining and text technologies; despite scholars working on computational linguistics and networks traditionally belong to different scientific communities, we expect that forthcoming computational approaches to prevent fake news from polluting the social media must be developed using hybrid and up-to-date methodologies.

FakeFlow: Fake News Detection by Modeling the Flow of Affective Information

Fake news articles often stir the readers' attention by means of emotional appeals that arouse their feelings. Unlike in short news texts, authors of longer articles can exploit such affective factors to manipulate readers by adding exaggerations or fabricating events, in order to affect the readers' emotions. To capture this, we propose in this paper to model the flow of affective information in fake news articles using a neural architecture. The proposed model, FakeFlow, learns this flow by combining topic and affective information extracted from text. We evaluate the model's performance with several experiments on four real-world datasets. The results show that FakeFlow achieves superior results when compared against state-of-the-art methods, thus confirming the importance of capturing the flow of the affective information in news articles.

Analysis and tuning of hierarchical topic models based on Renyi entropy approach

Hierarchical topic modeling is a potentially powerful instrument for determining the topical structure of text collections that allows constructing a topical hierarchy representing levels of topical abstraction. However, tuning of parameters of hierarchical models, including the number of topics on each hierarchical level, remains a challenging task and an open issue. In this paper, we propose a Renyi entropy-based approach for a partial solution to the above problem. First, we propose a Renyi entropy-based metric of quality for hierarchical models. Second, we propose a practical concept of hierarchical topic model tuning tested on datasets with human mark-up. In the numerical experiments, we consider three different hierarchical models, namely, hierarchical latent Dirichlet allocation (hLDA) model, hierarchical Pachinko allocation model (hPAM), and hierarchical additive regularization of topic models (hARTM). We demonstrate that hLDA model possesses a significant level of instability and, moreover, the derived numbers of topics are far away from the true numbers for labeled datasets. For hPAM model, the Renyi entropy approach allows us to determine only one level of the data structure. For hARTM model, the proposed approach allows us to estimate the number of topics for two hierarchical levels.

Multilingual Irony Detection with Dependency Syntax and Neural Models

This paper presents an in-depth investigation of the effectiveness of dependency-based syntactic features on the irony detection task in a multilingual perspective (English, Spanish, French and Italian). It focuses on the contribution from syntactic knowledge, exploiting linguistic resources where syntax is annotated according to the Universal Dependencies scheme. Three distinct experimental settings are provided. In the first, a variety of syntactic dependency-based features combined with classical machine learning classifiers are explored. In the second scenario, two well-known types of word embeddings are trained on parsed data and tested against gold standard datasets. In the third setting, dependency-based syntactic features are combined into the Multilingual BERT architecture. The results suggest that fine-grained dependency-based syntactic information is informative for the detection of irony.

Classifier Combination Approach for Question Classification for Bengali Question Answering System

Question classification (QC) is a prime constituent of automated question answering system. The work presented here demonstrates that the combination of multiple models achieve better classification performance than those obtained with existing individual models for the question classification task in Bengali. We have exploited state-of-the-art multiple model combination techniques, i.e., ensemble, stacking and voting, to increase QC accuracy. Lexical, syntactic and semantic features of Bengali questions are used for four well-known classifiers, namely Na\"{\i}ve Bayes, kernel Na\"{\i}ve Bayes, Rule Induction, and Decision Tree, which serve as our base learners. Single-layer question-class taxonomy with 8 coarse-grained classes is extended to two-layer taxonomy by adding 69 fine-grained classes. We carried out the experiments both on single-layer and two-layer taxonomies. Experimental results confirmed that classifier combination approaches outperform single classifier classification approaches by 4.02% for coarse-grained question classes. Overall, the stacking approach produces the best results for fine-grained classification and achieves 87.79% of accuracy. The approach presented here could be used in other Indo-Aryan or Indic languages to develop a question answering system.