Considering the potential of tools such as ChatGPT or Gemini to generate texts in a similar way to a human would do, having reliable detectors of AI –AI-generated content (AIGC)– is vital to combat the misuse and the surrounding negative consequences of those tools. Most research on AIGC detection has focused on the English language, often overlooking other languages that also have tools capable of generating human-like texts, such is the case of the Spanish language. This paper proposes a novel multilingual and multi-task approach for detecting machine versus human-generated text. The first task classifies whether a text is written by a machine or by a human, which is the research objective of this paper. The second task consists in detect the language of the text. To evaluate the results of our approach, this study has framed the scope of the AuTexTification shared task and also we have collected a different dataset in Spanish. The experiments carried out in Spanish and English show that our approach is very competitive concerning the state of the art, as well as it can generalize better, thus being able to detect an AI-generated text in multiple domains.

Disinformation is an important problem facing society nowadays. Given the rapid and easy access to information, news stories quickly go viral, the vast majority of which are misleading and with no prospect of verification. Specifically, the headline of a correctly designed news item must correspond to a summary of the main information of that news item and it should be neutral. However, many headlines circulating on the Internet use false or distorted information, seeking to confuse or mislead the reader. Misleading headlines indicate a dissonance between the headline and the content of the news story. From a computational perspective, this problem is being tackled as a Stance Detection problem between the headline and the body text of the news item. This paper contributes to the fight against the spread of misleading information by presenting a generic and flexible multi-level hierarchical classification. The approach is based on two stages that enable the detection of the stance between the news headline and the body text. The proposed architecture, called HeadlineStanceChecker+ uses the headline and only the essential information of the news item (not the full body text) as inputs. To extract this essential information, different summarization approaches (extractive and abstractive) are analyzed in order to determine the most relevant information for the task. The experimentation has been carried out using the Fake News Challenge (FNC-1) dataset. A 94.49% accuracy was obtained using extractive summaries, which were more helpful than abstractive ones. HeadlineStanceChecker+ improves the accuracy results of existing state-of-the-art systems. In conclusion, using automatic extractive summaries together with the two-stage generic architecture is an effective solution to the problem.

Annotated corpora are indispensable tools to train computational models in Natural Language Processing. However, in the case of more complex semantic annotation processes, it is a costly, arduous, and time-consuming task, resulting in a shortage of resources to train Machine Learning and Deep Learning algorithms. In consideration, this work proposes a methodology, based on the human-in-the-loop paradigm, for semi-automatic annotation of complex tasks. This methodology is applied in the construction of a reliability dataset of Spanish news so as to combat disinformation and fake news. We obtain a high quality resource by implementing the proposed methodology for semi-automatic annotation, increasing annotator efficacy and speed, with fewer examples. The methodology consists of three incremental phases and results in the construction of the RUN dataset. The annotation quality of the resource was evaluated through time-reduction (annotation time reduction of almost 64% with respect to the fully manual annotation), annotation quality (measuring consistency of annotation and inter-annotator agreement), and performance by training a model with RUN semi-automatic dataset (Accuracy 95% F1 95%), validating the suitability of the proposal.

Early detection of disinformation is one of the most challenging big-scale problems facing present day society. This is why the application of technologies such as Artificial Intelligence and Natural Language Processing is necessary. The vast majority of Artificial Intelligence approaches require annotated data, and generating these resources is very expensive. This proposal aims to improve the efficiency of the annotation process with a two-level semi-automatic annotation methodology. The first level extracts relevant information through summarization techniques. The second applies a Human-in-the-Loop strategy whereby the labels are pre-annotated by the machine, corrected by the human and reused by the machine to retrain the automatic annotator. After evaluating the system, the average annotation time per news item is reduced by 50%. In addition, a set of experiments on the semi-automatically annotated dataset that is generated are performed so as to demonstrate the effectiveness of the proposal. Although the dataset is annotated in terms of unreliable content, it is applied to the veracity detection task with very promising results (0.95 accuracy in reliability detection and 0.78 in veracity detection).

Discovering the main features of virality patterns in Twitter is the focus of this research. Five trending topics related to the COVID-19 pandemic were selected for the study, with Spanish as the target language. To carry out the discovery of virality patterns, we applied opinion mining techniques that enable us to structure the information based on the polarity of the messages and the emotions they contain. After transforming the information from an unstructured textual representation to a structured one, data mining techniques were applied, specifically association rules mining. Message patterns with the highest virality (high shares and high likes), and at the same time the most relevant characteristics of the patterns with less impact were extracted. After an exhaustive analysis of the most relevant non-redundant rules, it can be concluded that messages with a high-negative polarity and a very high emotional charge, especially emotions that have intensified with the COVID-19 pandemic, such as fear, sadness, anger and surprise are more likely to go viral in social media. By contrast, messages with little news coverage in the media, few authors, and the absence of surprise are relevant features when it comes to seeing messages with very low dissemination in social media.

This paper presents a novel architecture for dealing with Automatic Fake News detection. The architecture factors in the discourse structure of news in traditional digital media and is based on two premises. First, fake news tends to mix true and false information with the purpose of confusing readers. Second, this research is focused on fake news delivered in traditional digital media , so our approach considers the influence of the journalistic structure of news, and the way journalists tend to introduce the essential content in a news story using 5W1H answer. Considering both premises, this proposal deals with the news components separately because some may be true or false, instead of considering the veracity value of the news article as a unit. A two-layer architecture is proposed, Structure and Veracity layers. To demonstrate the validity of the proposal, a new dataset was created and annotated with a new fine-grained annotation scheme (FNDeepML) that considers the different elements of the news document and their veracity. Due to the severity of the COVID-19 pandemic crisis, health is the chosen domain, and Spanish is the language used to validate the architecture, given the lack of research in this language. However, the proposal can be applied to any other language or domain. The performance of the Veracity layer of our proposal, which factors in the traditional news article structure and the 5W1H annotation, is capable of delivering a result of F 1 =0.807. This represents a strong improvement when compared to the baseline, which uses the whole document with a single veracity value, obtaining F 1 =0.605. These findings validate the suitability and effectiveness of our approach.

The headline of a news article is designed to succinctly summarize its content, providing the reader with a clear understanding of the news item. Unfortunately, in the post-truth era, headlines are more focused on attracting the reader’s attention for ideological or commercial reasons, thus leading to mis- or disinformation through false or distorted headlines. One way of combating this, although a challenging task, is by determining the relation between the headline and the body text to establish the stance. Hence, to contribute to the detection of mis- and disinformation, this paper proposes an approach (HeadlineStanceChecker) that determines the stance of a headline with respect to the body text to which it is associated. The novelty rests on the use of a two-stage classification architecture that uses summarization techniques to shape the input for both classifiers instead of directly passing the full news body text, thereby reducing the amount of information to be processed while keeping important information. Specifically, summarization is done through Positional Language Models leveraging on semantic resources to identify salient information in the body text that is then compared to its corresponding headline. The results obtained show that our approach achieves 94.31% accuracy for the overall classification and the best FNC-1 relative score compared with the state of the art. It is especially remarkable that the system, which uses only the relevant information provided by the automatic summaries instead of the whole text, is able to classify the different stance categories with very competitive results, especially in the discuss stance between the headline and the news body text. It can be concluded that using automatic extractive summaries as input of our approach together with the two-stage architecture is an appropriate solution to the problem.

Post-truth is a term that describes a distorting phenomenon that aims to manipulate public opinion and behavior. One of its key engines is the spread of Fake News. Nowadays most news is rapidly disseminated in written language via digital media and social networks. Therefore, to detect fake news it is becoming increasingly necessary to apply Artificial Intelligence (AI) and, more specifically Natural Language Processing (NLP). This paper presents a review of the application of AI to the complex task of automatically detecting fake news. The review begins with a definition and classification of fake news. Considering the complexity of the fake news detection task, a divide-and-conquer methodology was applied to identify a series of subtasks to tackle the problem from a computational perspective. As a result, the following subtasks were identified: deception detection; stance detection; controversy and polarization; automated fact checking; clickbait detection; and, credibility scores. From each subtask, a PRISMA compliant systematic review of the main studies was undertaken, searching Google Scholar. The various approaches and technologies are surveyed, as well as the resources and competitions that have been involved in resolving the different subtasks. The review concludes with a roadmap for addressing the future challenges that have emerged from the analysis of the state of the art, providing a rich source of potential work for the research community going forward.

To provide with easy and optimal access to digital information, narrative summaries must have a coherent and natural structure. Depending on how a summary is produced, a distinction can be made between extractive and abstractive summaries. Using an abstractive summarization approach, the relevant information (e.g., who? what?, when?, where?,...) could be fused together, leading to the generation of one or more new sentences. However, in order to do this it is necessary to obtain and process the temporal information in a text. A very effective way is the generation of timelines starting from multiple documents so that the generation of summaries is supported by the generated timeline, without losing the relevant temporal information of the texts. In this proposal, a enriched timeline is generated automatically, and the process of generating abstractive summaries is presented using this timeline as a basis [Barros et al., 2019]. Finally, potential applications of the automatic timeline generation would be presented, as for example its application to Fake News detection.

Nowadays, the automatic processing of digitalized documents is crucial to cope with the increasing amount of information available. This issue is addressed from the natural language processing (NLP) research field. One of the tasks required for many NLP applications is temporal information processing. It involves the automatic extraction and interpretation of temporal expressions, events, and their relations. Specifically, the identification and the categorization of temporal relations are the most complex subtasks yet to solve, judging from the results reported in the latest international evaluation exercise. Temporal relation identification has been addressed by very few approaches, and the current categorization approaches are still not a definitive solution. This paper presents a system that approaches temporal relation identification and categorization. The former is approached with a knowledge-driven strategy and the later with data-driven strategy based on different machine-learning techniques. Our proposal has been empirically evaluated over the currently available English data sets annotated with temporal information (TimeBank and AQUAINT) in a 10-fold cross-validated experiment. The results obtained support that the presented approach achieves a high performance. It improves the baseline F1 by 46% and outperforms the state of the art. © 2012 Wiley Periodicals, Inc.

As the Internet grows, it becomes essential to find efficient tools to deal with all the available information. Question answering (QA) and text summarization (TS) research fields focus on presenting the information requested by users in a more concise way. In this paper, the appropriateness and benefits of using summaries in semantic QA are analyzed. For this purpose, a combined approach where a TS component is integrated into a Web-based semantic QA system is developed. The main goal of this paper is to determine to what extent TS can help semantic QA approaches, when using summaries instead of search engine snippets as the corpus for answering questions. In particular, three issues are analyzed: (i) the appropriateness of query-focused (QF) summarization rather than generic summarization for the QA task, (ii) the suitable length comparing short and long summaries, and (iii) the benefits of using TS instead of snippets for finding the answers, tested within two semantic QA approaches (named entities and semantic roles). The results obtained show that QF summarization is better than generic (58% improvement), short summaries are better than long (6.3% improvement), and the use of TS within semantic QA improves the performance for both named-entity-based (10%) and, especially, semantic-role-based QA (47.5%). © 2011 Wiley Periodicals, Inc.

This paper presents a multilayered architecture that enhances the capabilities of current QA systems and allows different types of complex questions or queries to be processed. The answers to these questions need to be gathered from factual information scattered throughout different documents. Specifically, we designed a specialized layer to process the different types of temporal questions. Complex temporal questions are first decomposed into simple questions, according to the temporal relations expressed in the original question. In the same way, the answers to the resulting simple questions are recomposed, fulfilling the temporal restrictions of the original complex question. A novel aspect of this approach resides in the decomposition which uses a minimal quantity of resources, with the final aim of obtaining a portable platform that is easily extensible to other languages. In this paper we also present a methodology for evaluation of the decomposition of the questions as well as the ability of the implemented temporal layer to perform at a multilingual level. The temporal layer was first performed for English, then evaluated and compared with: a) a general purpose QA system (F-measure 65.47% for QA plus English temporal layer vs. 38.01% for the general QA system), and b) a well-known QA system. Much better results were obtained for temporal questions with the multilayered system. This system was therefore extended to Spanish and very good results were again obtained in the evaluation (F-measure 40.36% for QA plus Spanish temporal layer vs. 22.94% for the general QA system).

This paper presents the automatic extension of TERSEO to other languages, a knowledge-based system for the recognition and normalization of temporal expressions, originally developed for Spanish. TERSEO was extended to English and Italian through the automatic translation of the temporal expressions, and it was presented in previous works (see Saquete et al.), but a new methodology has been designed with the purpose of obtaining better results in this issue. This new methodology is based on the use of parallel corpora for extending the TERSEO temporal model to other languages. In this case, two different methods have been tested: (1) automatic translation of TERSEO patterns to other languages and (2) automatic corpora annotation in the target side of parallel corpora. The main idea is focused on annotating the Spanish side of a parallel corpora, projecting the analysis to the second language, and then obtaining new TERSEO patterns (1) and new annotated corpus (2). The set of new patterns will be used to improve the current TERSEO language independent modules. Whereas the new annotated corpus will be used to train a ML system. This system will annotate new temporal expressions in the new language.

The knowledge-based system TERSEO was originally developed for the recognition and normalization of temporal expressions in Spanish and then extended to other languages: to English first, through the automatic translation of the temporal expressions, and then to Italian, applying a porting process where the automatic translation of the rules was combined with the extraction of expressions from an annotated corpus. In this paper we present a new automatic porting procedure, where resolution rules are automatically assigned to the temporal expressions that have been acquired in a new language, thus eliminating the need for automatic translation and consequently minimizing the errors produced. This is achieved by exploiting the rules of the temporal model, which are language independent, and the information extracted from the annotated corpus. Evaluation results of the updated version of TERSEO for English show a considerable improvement in recognition performance (+ 14% F-measure) with respect to the original system

This paper presents a multi-layered Question Answering (Q.A.) architecture suitable for enhancing current Q.A. capabilities with the possibility of processing complex questions. That is, questions whose answer needs to be gathered from pieces of factual information scattered in different documents. Specifically, we have designed a layer oriented to process the different types of temporal questions. Complex temporal questions are first decomposed into simpler ones, according to the temporal relationships expressed in the original question.In the same way, the answers of each simple question are re-composed, fulfilling the temporal restrictions of the original complex question.Using this architecture, a Temporal Q.A. system has been developed.In this paper, we focus on explaining the first part of the process: the decomposition of the complex questions. Furthermore, it has been evaluated with the TERQAS question corpus of 112 temporal questions. For the task of question splitting our system has performed, in terms of precision and recall, 85% and 71%, respectively.