The availability of large annotated corpora from social media and the development of powerful classification approaches have contributed in an unprecedented way to tackle the challenge of monitoring users’ opinions and sentiments in online social platforms across time. Such linguistic data are strongly affected by events and topic discourse, and this aspect is crucial when detecting phenomena such as hate speech, especially from a diachronic perspective. We address this challenge by focusing on a real case study: the “Contro l’odio” platform for monitoring hate speech against immigrants in the Italian Twittersphere. We explored the temporal robustness of a BERT model for Italian (AlBERTo), the current benchmark on non-diachronic detection settings. We tested different training strategies to evaluate how the classification performance is affected by adding more data temporally distant from the test set and hence potentially different in terms of topic and language use. Our analysis points out the limits that a supervised classification model encounters on data that are heavily influenced by events. Our results show how AlBERTo is highly sensitive to the temporal distance of the fine-tuning set. However, with an adequate time window, the performance increases, while requiring less annotated data than a traditional classifier.

The ability to understand people through spoken language is a skill that many human beings take for granted. On the contrary, the same task is not as easy for machines, as consequences of a large number of variables which vary the speaking sound wave while people are talking to each other. A sub-task of speeches understanding is about the detection of the emotions elicited by the speaker while talking, and this is the main focus of our contribution. In particular, we are presenting a classification model of emotions elicited by speeches based on deep neural networks (CNNs). For the purpose, we focused on the audio recordings available in the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset. The model has been trained to classify eight different emotions (neutral, calm, happy, sad, angry, fearful, disgust, surprise) which correspond to the ones proposed by Ekman plus the neutral and calm ones. We considered as evaluation metric the F1 score, obtaining a weighted average of 0.91 on the test set and the best performances on the "Angry" class with a score of 0.95. Our worst results have been observed for the sad class with a score of 0.87 that is nevertheless better than the state-of-the-art. In order to support future development and the replicability of results, the source code of the proposed model is available on the following GitHub repository: https://github.com/marcogdepinto/Emotion-Classification-Ravdess

In this article, we present HealthAssistantBot, an intelligent virtual assistant able to talk with patients in order to understand their symptomatology, suggest doctors, and monitor treatments and health parameters. In a simple way, by exploiting a natural language-based interaction, the system allows the user to create her health profile, to describe her symptoms, to search for doctors or to simply remember a treatment to follow. Specifically, our methodology exploits machine learning techniques to process users symptoms and to automatically infer her diseases. Next, the information obtained is used by our recommendation algorithm to identify the nearest doctor who can best treat the user's condition, considering the community data. In the experimental session we evaluated our HealthAssistantBot with both an offline and online evaluation. In the first case, we assessed the performance of our internal components, while in the second one we carried out a study involving 102 subjects who interacted with the conversational agent in a daily use scenario. Results are encouraging and showed the effectiveness of the strategy in supporting the patients in taking care of their health.

Personalized content recommendations have been pivotal to the content experience in digital media from video streaming to social networks. However, several domain specific challenges have held back adoption of recommender systems in news publishing. To address these challenges, we introduce the Ekstra Bladet News Recommendation Dataset (EB-NeRD). The dataset encompasses data from over a million unique users and more than 37 million impression logs from Ekstra Bladet. It also includes a collection of over 125, 000 Danish news articles, complete with titles, abstracts, bodies, and metadata, such as categories. EB-NeRD served as the benchmark dataset for the RecSys ’24 Challenge, where it was demonstrated how the dataset can be used to address both technical and normative challenges in designing effective and responsible recommender systems for news publishing. The dataset is available at: https://recsys.eb.dk.