Traditional recommender systems such as matrix factorization methods have primarily focused on learning a shared dense embedding space to represent both items and user preferences. Subsequently, sequence models such as RNN, GRUs, and, recently, Transformers have emerged and excelled in the task of sequential recommendation. This task requires understanding the sequential structure present in users’ historical interactions to predict the next item they may like. Building upon the success of Large Language Models (LLMs) in a variety of tasks, researchers have recently explored using LLMs that are pretrained on vast corpora of text for sequential recommendation. To use LLMs for sequential recommendation, both the history of user interactions and the model’s prediction of the next item are expressed in text form. We propose CALRec, a two-stage LLM finetuning framework that finetunes a pretrained LLM in a two-tower fashion using a mixture of two contrastive losses and a language modeling loss: the LLM is first finetuned on a data mixture from multiple domains followed by another round of target domain finetuning. Our model significantly outperforms many state-of-the-art baselines (+37% in Recall@1 and +24% in NDCG@10) and our systematic ablation studies reveal that (i) both stages of finetuning are crucial, and, when combined, we achieve improved performance, and (ii) contrastive alignment is effective among the target domains explored in our experiments.

Background: Natural language processing (NLP) has the potential to promote public health. However, applying these technologies in African health systems faces challenges, including limited digital and computational resources to support the continent’s diverse languages and needs. Objective: This scoping review maps the evidence on NLP technologies for public health in Africa, addressing the following research questions: (1) What public health needs are being addressed by NLP technologies in Africa, and what unmet needs remain? (2) What factors influence the availability of public health NLP technologies across African countries and languages? (3) What stages of deployment have these technologies reached, and to what extent have they been integrated into health systems? (4) What measurable impact has these technologies had on public health outcomes, where such data are available? (5) What recommendations have been proposed to enhance the quality, cost, and accessibility of health-related NLP technologies in Africa? Methods: This scoping review includes academic studies published between January 1, 2013, and October 3, 2024. A systematic search was conducted across databases, including MEDLINE via PubMed, ACL Anthology, Scopus, IEEE Xplore, and ACM Digital Library, supplemented by gray literature searches. Data were extracted and the NLP technology functions were mapped to the World Health Organization’s list of essential public health functions and the United Nations’ sustainable development goals (SDGs). The extracted data were analyzed to identify trends, gaps, and areas for future research. This scoping review follows the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) reporting guidelines, and its protocol is publicly available. Results: Of 2186 citations screened, 54 studies were included. While existing NLP technologies support a subset of essential public health functions and SDGs, language coverage remains uneven, with limited support for widely spoken African languages, such as Kiswahili, Yoruba, Igbo, and Zulu, and no support for most of Africa’s >2000 languages. Most technologies are in prototyping phases, with only one fully deployed chatbot addressing vaccine hesitancy. Evidence of measurable impact is limited, with 15% (8/54) studies attempting health-related evaluations and 4% (2/54) demonstrating positive public health outcomes, including improved participants’ mood and increased vaccine intentions. Recommendations include expanding language coverage, targeting local health needs, enhancing trust, integrating solutions into health systems, and adopting participatory design approaches. The gray literature reveals industry- and nongovernmental organizations–led projects focused on deployable NLP applications. However, these projects tend to support only a few major languages and specific use cases, indicating a narrower scope than academic research. Conclusions: Despite growth in NLP research for public health, major gaps remain in deployment, linguistic inclusivity, and health outcome evaluation. Future research should prioritize cross-sectoral and needs-based approaches that engage local communities, align with African health systems, and incorporate rigorous evaluations to enhance public health outcomes. International Registered Report Identifier (IRRID): RR2-doi:10.1101/2024.07.02.24309815

While recent studies indicate that AI could play an important role in detecting early signs of Alzheimer's disease in speech, this use of data from individuals with cognitive decline raises numerous ethical concerns. In this paper, we identify and explain concerns related to autonomy (including consent, depersonalization and disclosure), privacy and data protection (including the handling of personal content and medical information), welfare (including distress, discrimination and reliability), transparency (including the interpretability of language features and AI-based decision-making for developers and clinicians), and fairness (including bias and the distribution of benefits). Our aim is to not only raise awareness of the ethical concerns posed by the use of AI in speech-based Alzheimer's detection, but also identify ways in which these concerns might be addressed. To this end, we conclude with a list of suggestions that could be incorporated into ethical guidelines for researchers and clinicians working in this area.

Alan Ansell, Marinela Parović, Ivan Vulić, Anna Korhonen, Edoardo Ponti. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.