Umeå University's logo

Algorithms frequently manage online advertising markets, aligning advertisements with article topics. Our work investigates how users perceive the relevance of ads to articles when ads are placed using different keyword extraction algorithms, including Large Language Models (LLMs), and how transparency about the placement procedure influences these perceptions and behavioral intentions. We conducted an online user experiment (N = 498) where ads are matched with news articles using the keyword extraction methods TF-IDF, KeyBERT, and DeepSeek. Results indicate that lightweight methods can match advanced LLMs in delivering high user-perceived ad-article relevance, which in turn fosters click and purchase intentions. However, providing explanations for the ad-article placements by displaying extracted keywords reduces ad interest and thereby weakens behavioral intentions, while simultaneously increasing perceived relevance and moderating algorithm effects. These findings highlight the complex impact of transparency-increasing explanations and suggest that algorithmic precision metrics must be complemented by user perception and intention measures.

Keyword extraction is a foundational task in natural language processing, underpinning countless real-world applications. One of these is contextual advertising, where keywords help predict the topical congruence between ads and their surrounding media contexts to enhance advertising effectiveness. Recent advances in artificial intelligence have improved keyword extraction capabilities but also introduced concerns about computational cost. Moreover, although the end-user experience is of vital importance, human evaluation of keyword extraction performances remains under-explored. This study provides a comparative evaluation of prevalent keyword extraction algorithms with different levels of complexity represented by TF-IDF, KeyBERT, and Llama 2. To evaluate their effectiveness, a mixed-methods approach is employed, combining quantitative benchmarking with qualitative assessments from 855 participants through four survey-based experiments. The findings demonstrate that KeyBERT achieves an effective balance between user preferences and computational efficiency, compared to the other algorithms. We observe a clear overall preference for gold-standard keywords, but there is a misalignment between algorithmic benchmark performance and user ratings. This reveals a long-overlooked gap between traditional precision-focused metrics and user-perceived algorithm efficiency. The study underscores the importance of human-in-the-loop evaluation methodologies and proposes analytical tools to facilitate their implementation.

Stricter regulations on personal data are causing a shift towards contextual advertising, where keywords are used to predict the topical congruence between ads and their surrounding media contexts - an alignment shown to enhance advertising effectiveness. Recent advances in AI, particularly large language models, have improved keyword extraction capabilities but also introduced concerns about computational cost. This study conducts a comparative, survey-based evaluation experiment of three prominent keyword extraction approaches, emphasising user-perceived accuracy and efficiency. Based on responses from 552 participants, the embedding-based approach emerges as the preferred method. The findings underscore the importance of human-in-the-loop evaluation in real-world settings.

We perform a mixed-method frame semantics-based analysis on a dataset of more than 49,000 sentences collected from 5846 news articles that mention AI. The dataset covers the twelve-month period centred around the launch of OpenAI’s chatbot ChatGPT and is collected from the most visited open-access English-language news publishers. Our findings indicate that during the six months succeeding the launch, media attention rose tenfold—from already historically high levels. During this period, discourse has become increasingly centred around experts and political leaders, and AI has become more closely associated with dangers and risks. A deeper review of the data also suggests a qualitative shift in the types of threat AI is thought to represent, as well as the anthropomorphic qualities ascribed to it.

We introduce opinion units, a contribution to the field Aspect-Based Sentiment Analysis (ABSA) that extends aspect- sentiment pairs by including substantiating excerpts, derived through hybrid abstractive-extractive summarisation. The goal is to provide fine-grained information without sacrificing succinctness and abstraction. Evaluations on review datasets demonstrate that large language models (LLMs) can accurately extract opinion units through few-shot learning. The main types of errors are providing incomplete contexts for opinions and and mischaracterising objective statements as opinions. The method reduces the need for labelled data and allows the LLM to dynamically define aspect types. As a practical evaluation, we present a case study on similarity search across academic datasets and public review data. The results indicate that searches leveraging opinion units are more successful than those relying on traditional data-segmentation strategies, showing robustness across datasets and embeddings.

The central decision-making processes involved in online advertising are often supported by Reinforcement Learning (RL), which serves to optimise long-term accumulative rewards through interactions with evolving environments. While RL’s potential in various real-world applications has been reviewed in extant survey works, the specific ways RL algorithms address online advertising challenges remain unchartered. Therefore, this paper reviews RL applications in this practice area, identifying core challenges and key issues including trust concerns. We categorize reviewed work based on problem domains and propose potential directions for future research. Our goal is to bridge the cross-disciplinary gap in this field, offering perspectives and guidance for researchers and practitioners.

We introduce a reliable approach for enhancing multilingual language models that preserves cultural knowledge while improving reasoning capabilities, focusing on low-resource languages. Using Qwen as a base model, we demonstrate that trust-aware model merging can verifiably improve performance without compromising cultural understanding. Our proposed approach achieves quantifiable improvements in both reasoning tasks and cultural benchmarks while maintaining computational efficiency. Results on Vietnamese and Arabic language tasks show consistent performance gains while preserving cultural knowledge, offering a reliable path for developing trustworthy multilingual AI systems. Our models are available at github.com/WARA-ML/waraml-mini-brains.

The Visual Object Tracking Segmentation VOTS2024 challenge is the twelfth annual tracker benchmarking activity of the VOT initiative. This challenge consolidates the new tracking setup proposed in VOTS2023, which merges short-term and long-term as well as single-target and multiple-target tracking with segmentation masks as the only target location specification. Two sub-challenges are considered. The VOTS2024 standard challenge, focusing on classical objects and the VOTSt2024, which considers objects undergoing a topological transformation. Both challenges use the same performance evaluation methodology. Results of 28 submissions are presented and analyzed. A leaderboard, with participating trackers details, the source code, the datasets, and the evaluation kit are publicly available on the website (https://www.votchallenge.net/vots2024/).

The VOTS2025 is the third edition of the Visual Object Tracking Segmentation benchmark. Organised the VOT initiative, VOTS builds on 10 years of experience in organising VOT challenges. Building on the tracking setup introduced in VOTS2023, the challenge continues to integrate short-term and long-term tracking, as well as single-target and multi-target scenarios, using segmentation masks as the sole form of target annotation. This year's benchmark features three sub-challenges. VOTS2025 and VOTSt2025, evaluate tracking of conventional objects and objects undergoing topological changes, respectively. A new addition, VOTS-RT2025, aims to foster the development of efficient tracking models by introducing constraints that highlight realtime performance. All sub-challenges adopt a consistent evaluation protocol, with VOTS-RT2025 introducing specific modifications to reflect latency-aware performance. We report and analyze results from 32 submissions. Full tracker descriptions, source code, datasetsand the evaluation toolkit are available on the project website (https://www.votchallenge.net/vots2025/=.

We present VOICE, a novel approach to science communication that connects large language models' conversational capabilities with interactive exploratory visualization. VOICE introduces several innovative technical contributions that drive our conversational visualization framework. Based on the collected design requirements, we introduce a two-layer agent architecture that can perform task assignment, instruction extraction, and coherent content generation. We employ fine-tuning and prompt engineering techniques to tailor agents' performance to their specific roles and accurately respond to user queries. Our interactive text-to-visualization method generates a flythrough sequence matching the content explanation. In addition, natural language interaction provides capabilities to navigate and manipulate 3D models in real-time. The VOICE framework can receive arbitrary voice commands from the user and respond verbally, tightly coupled with a corresponding visual representation, with low latency and high accuracy. We demonstrate the effectiveness of our approach by implementing a proof-of-concept prototype and applying it to the molecular visualization domain: analyzing three 3D molecular models with multiscale and multi-instance attributes. Finally, we conduct a comprehensive evaluation of the system, including quantitative and qualitative analyses on our collected dataset, along with a detailed public user study and expert interviews. The results confirm that our framework and prototype effectively meet the design requirements and cater to the needs of diverse target users.