ABSTRACT The opioid overdose epidemic constitutes a critical public health crisis, necessitating advanced surveillance tools to enable timely intervention. Social media platforms provide a real‐time source of information on drug‐related behaviours. However, extracting structured knowledge from their informal, slang‐heavy and fragmented text presents significant technical challenges. While Named Entity Recognition (NER) enables the automated identification of drug‐related entities. Prior work has focused mainly on general biomedical domains, with limited exploration of domain‐specific scenarios. To address this gap, this study makes three key contributions. First, we present the first manually annotated, domain‐specific NER dataset for opioid overdose, comprising eight unique entity types, with a particular focus on routes of administration, sourced from Reddit posts spanning January 2021 to December 2023. Second, we provide a detailed description of the annotation process and guidelines, and systematically discuss challenges encountered during annotation, including slang, fragmented expressions and ambiguous language commonly found in social media posts. Third, we conduct an extensive series of experiments, including classical machine learning, deep learning with pretrained embeddings combined with CRF decoding, and transformer‐based models, evaluated at both token and entity levels. The proposed model achieved strong and balanced performance, with F1‐scores of 0.891 at both token and entity levels, highlighting the effectiveness of domain‐specific modelling for opioid overdose surveillance on social media.

NLP research has increasingly focused on subjective tasks such as emotion analysis. However, existing emotion benchmarks suffer fromtwo major shortcomings: (1) they largely rely on keyword-based emotion recognition, overlooking crucial cultural dimensions required fordeeper emotion understanding, and (2) many are created by translating English-annotated data into other languages, leading to potentially unreliable evaluation. To address these issues, we introduce Cultural Lenses on Emotion (CuLEmo), the first benchmark designedto evaluate culture-aware emotion prediction across six languages: Amharic, Arabic, English, German, Hindi, and Spanish. CuLEmocomprises 400 crafted questions per language, each requiring nuanced cultural reasoning and understanding. We use this benchmark to evaluate several state-of-the-art LLMs on culture-aware emotion prediction and sentiment analysis tasks. Our findings reveal that (1) emotion conceptualizations vary significantly across languages and cultures, (2) LLMs performance likewise varies by language and cultural context, and (3) prompting in English with explicit country context often outperforms in-language prompts for culture-aware emotion and sentiment understanding. The dataset and evaluation code is available.

Large Language Models (LLMs) show promising learning and reasoning abilities. Compared to other NLP tasks, multilingual and multi-label emotion evaluation tasks are under-explored in LLMs. In this paper, we present EthioEmo, a multi-label emotion classification dataset for four Ethiopian languages, namely, Amharic (amh), Afan Oromo (orm), Somali (som), and Tigrinya (tir). We perform extensive experiments with an additional English multi-label emotion dataset from SemEval 2018 Task 1. Our evaluation includes encoder-only, encoder-decoder, and decoder-only language models. We compare zero and few-shot approaches of LLMs to fine-tuning smaller language models. The results show that accurate multi-label emotion classification is still insufficient even for high-resource languages such as English, and there is a large gap between the performance of high-resource and low-resource languages. The results also show varying performance levels depending on the language and model type. EthioEmo is available publicly to further improve the understanding of emotions in language models and how people convey emotions through various languages.

The task of fake news detection is to distinguish legitimate news articles that describe real facts from those which convey deceiving and fictitious information. As the fake news phenomenon is omnipresent across all languages, it is crucial to be able to efficiently solve this problem for languages other than English. A common approach to this task is supervised classification using features of various complexity. Yet supervised machine learning requires substantial amount of annotated data. For English and a small number of other languages, annotated data availability is much higher, whereas for the vast majority of languages, it is almost scarce. We investigate whether machine translation at its present state could be successfully used as an automated technique for annotated corpora creation and augmentation for fake news detection focusing on the English-Urdu language pair. We train a fake news classifier for Urdu on (1) the manually annotated dataset originally in Urdu and (2) the machine-translated version of an existing annotated fake news dataset originally in English. We show that at the present state of machine translation quality for the English-Urdu language pair, the fully automated data augmentation through machine translation did not provide improvement for fake news detection in Urdu.

Partially observable Markov decision processes (POMDPs) are mathematical models for the planning of action sequences under conditions of uncertainty. Uncertainty in POMDPs is manifested in two ways: uncertainty in the perception of model states and uncertainty in the effects of actions on states. The diagnosis and treatment of cerebral vascular diseases (CVD) present this double condition of uncertainty, so we think that POMDP is the most suitable method to model them. In this paper, we propose a model of CVD that is based on observations obtained from neuroimaging studies such as computed tomography, magnetic resonance and ultrasound. The model is designed as a POMDP because the health status of the patient is not directly observable, and only can be deduced, with some probability, from the observations in the cerebral images. The components of the model (states, observations, actions, etc.) were defined based on specialized literature. A diagnosis of the patient’s health status is made and the most appropriate action for the recovery of health is recommended after introducing the observations when operating the model. Consultation of the probable state of health of the patient and alternative actions is also allowed.

A crucial step in plagiarism detection is text alignment. This task consists in finding similar text fragments between two given documents. We introduce an optimization methodology based on genetic algorithms to improve the performance of a plagiarism detection model by optimizing its input parameters. The implementation of the genetic algorithm is based on nonbinary representation of individuals, elitism selection, uniform crossover, and high mutation rate. The obtained parameter settings allow the plagiarism detection model to achieve better results than the state-of-the-art approaches.

We suggest the method that permits building a set of candidates to be considered semantic primitives from the standard explanatory dictionary. Our method is based on the frequencies of the words that are reachable in a semantic network constructed from the dictionary. The method implements word sense disambiguation techniques, network construction, and reachability analysis. In part of word sense disambiguation we use an improved Lesk's algorithm. In the part of analysis of reachability we show that the words to which our algorithm assigns high weight, are plausible candidates to be semantic primitives. It is also shown that better candidates to semantic primitives should be included in short vicious cycles, which is detected by our algorithm. We applied the method to a rather large Spanish explanatory dictionary.