A5071401829- Natural Language Processing Techniques
- Topic Modeling
- Authorship Attribution and Profiling
- Sentiment Analysis and Opinion Mining
- Text Readability and Simplification
- Advanced Text Analysis Techniques
- Public Relations and Crisis Communication
- Disaster Management and Resilience
- Hate Speech and Cyberbullying Detection
- Explainable Artificial Intelligence (XAI)
- Spam and Phishing Detection
- Speech and dialogue systems
- Knowledge Management and Technology
- Innovation Diffusion and Forecasting
- Historical and Religious Studies of Rome
- Seismology and Earthquake Studies
- Swearing, Euphemism, Multilingualism
- Text and Document Classification Technologies
- Advanced Malware Detection Techniques
- Intellectual Property and Patents
- Complex Network Analysis Techniques
- Stock Market Forecasting Methods
- Second Language Acquisition and Learning
- Misinformation and Its Impacts
- Interpreting and Communication in Healthcare
Institute for Computational Linguistics “A. Zampolli”
2014-2021
National Academies of Sciences, Engineering, and Medicine
2018
National Research Council
2013-2018
University of Groningen
2017
Leipzig University
2015
University of Tübingen
2014
Much progress has been made in the field of sentiment analysis past years. Researchers relied on textual data for this task, while only recently they have started investigating approaches to predict sentiments from multimedia content. With increasing amount shared social media, there is also a rapidly growing interest that work "in wild", i.e. are able deal with uncontrolled conditions. In work, we faced challenge training visual classifier starting large set user-generated and unlabeled...
This work focuses on the analysis of Italian social media messages for disaster management and aims at detection carrying critical information damage assessment task. A main novelty this study consists in focus out-domain cross-event detection, investigation most relevant tweet-derived features these tasks. We devised different experiments by resorting to a wide set linguistic qualifying lexical grammatical structure text as well ad-hoc specifically implemented investigated effective that...
The paper investigates the problem of sentence readability assessment, which is modelled as a classification task, with specific view to text simplification.In particular, it addresses two open issues connected it, i.e. corpora be used for training, and identification most effective features determine readability.An existing assessment tool developed Italian was specialized at level training corpus learning algorithm.A maximum entropy-based feature selection ranking algorithm (grafting)...
In this paper we present PaCCSS-IT, a Parallel Corpus of Complex-Simple Sentences for ITalian.To build the resource develop new method automatically acquiring corpus complex-simple paired sentences able to intercept structural transformations and particularly suitable text simplification.The requires wide amount texts that can be easily extracted from web making it also less-resourced languages.We test on Italian language available biggest automatic simplification.
In recent years, the explainable artificial intelligence (XAI) paradigm is gaining wide research interest. The natural language processing (NLP) community also approaching shift of paradigm: building a suite models that provide an explanation decision on some main task, without affecting performances. It not easy job for sure, especially when very poorly interpretable are involved, like almost ubiquitous (at least in NLP literature last years) transformers. Here, we propose two different...
In this paper, we describe the approach of ItaliaNLP Lab team to native language identification and discuss results submitted as participants essay track NLI Shared Task 2017. We introduce for first time a 2-stacked sentence-document architecture that is able exploit both local sentence information wide set general-purpose features qualifying lexical grammatical structure whole document. When evaluated on official test set, our stacked obtained best result among all with an F1 score 0.8818.
The task of witness detection in social media is crucial for many practical applications, including rumor debunking, emergency management, and public opinion mining. Yet to date, it has been approached an approximated way. We propose a method addressing strict realistic fashion. By employing hybrid crowdsensing over Twitter, we contact real-life witnesses use their reactions build strong ground-truth, thus avoiding manual, subjective annotation the dataset. Using this dataset, develop system...