Research map / Vision-language and multimodal / Human and embodied modeling
BreakingNews: Article Annotation by Image and Text Processing
Arnau Ramisa A, Fei Yan, Francesc Moreno-Noguer, Krystian Mikolajczyk
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CNN on full Word2Vec matrix instead of SVM on best shallow feature | SVM BoW TF-IDF · Source detection | balanced accuracy (%): 72.6 -> 80.7 | Improved |
| use full BoW instead of truncated 5000-d BoW (ablation) | SVM · Source detection | accuracy (%): 70.8 -> 72.9 | Improved |
| use mean of Word2Vec instead of max (ablation) | SVM · Source detection | accuracy (%): 57.3 -> 68.6 | Improved |
| combine VGG19 and Places features instead of VGG19 alone (ablation) | CCA W2V mean · Article illustration | median rank: 499 -> 415 | Improved |
| use CNN with full Word2Vec matrix instead of CCA with W2V mean | CCA W2V mean + VGG19+Places · Article illustration | median rank: 415 -> 405 | No clear effect |
| replace article image with related Internet images (ablation) | CCA W2V mean + VGG19 · Article illustration | median rank: 499 -> 161 | Improved |
| add original image caption to visual representation (ablation) | CCA W2V mean + VGG19 · Article illustration | median rank: 499 -> 7 | Improved |
| anonymize proper nouns in the added caption (ablation) | CCA with captions · Article illustration | median rank: 7 -> 42 | Hurt |
| use CNN instead of SVR with W2V mean | SVR W2V mean · Popularity prediction | median l1 error: 3.63 -> 1.09 | Improved |
| use CNN instead of SVR on mean l1 error | SVR W2V mean · Popularity prediction | mean l1 error: 69.75 -> 70.72 | No clear effect |
| use GCD loss instead of Euclidean loss (ablation) | CNN · Geolocation prediction | median GCD (thousand km): 1.42 -> 0.90 | Improved |
| train geolocation and popularity jointly (multitask) (ablation) | CNN single-task · Geolocation + popularity | median GCD / median l1: 0.90 -> 1.17; 1.09 -> 1.68 | Hurt |
Closest papers by meaning
- NewsStories: Illustrating Articles with Visual Summaries (ECCV 2022 · similarity 0.87)
- Transform and Tell: Entity-Aware News Image Captioning (CVPR 2020 · similarity 0.86)
- Good News, Everyone! Context Driven Entity-Aware Captioning for News Images (CVPR 2019 · similarity 0.85)
- Retrieval of Sentence Sequences for an Image Stream via Coherence Recurrent Convolutional Networks (TPAMI 2017 · similarity 0.85)
- Knowledge Completes the Vision: A Multimodal Entity-aware Retrieval-Augmented Generation Framework for News Image Captioning (AAAI 2026 · similarity 0.84)
- Newswire: A Large-Scale Structured Database of a Century of Historical News (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
