RpBERT: A Text-image Relation Propagation-based BERT Model for Multimodal NER

Authors: Lin Sun, Jiquan Wang, Kai Zhang, Yindu Su, Fangsheng Weng13860-13868

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In the experiments, we deeply analyze the changes in visual attention before and after the use of text-image relation propagation. Our model achieves state-of-the-art performance on the MNER datasets. Experiments Datasets In the experiments, we use three datasets to evaluate the performance.
Researcher Affiliation Academia 1 Department of Computer Science, Zhejiang University City College, Hangzhou, China 2 College of Computer Science and Technology, Zhejiang University, Hangzhou, China 3 Department of Computer Science and Technology, Tsinghua University, Beijing, China
Pseudocode Yes Algorithm 1 Multitask training procedure of Rp BERT for MNER. Input: The TRC dataset and MNER dataset. Output: θRp BERT , θRes Net, θF Cs, θbi LST M, and θCRF .
Open Source Code No The paper does not provide an explicit statement or a link indicating that the source code for their methodology is publicly available.
Open Datasets Yes TRC dataset of Bloomberg LP (Vempala and Preot iuc-Pietro 2019) MNER dataset of Fudan University (Zhang et al. 2018) MNER dataset of Snap Research (Lu et al. 2018)
Dataset Splits Yes The authors labeled 8,257 tweet texts using the BIO2 tagging scheme and used a 4,000/1,000/3,257 train/dev/test split. The authors labeled 6,882 tweet texts using the BIO tagging scheme and used a 4,817/1,032/1,033 train/dev/test split. We follow the same split of 8:2 for train/test sets as in (Vempala and Preot iuc-Pietro 2019).
Hardware Specification No The paper mentions models like ResNet-152 and BERT-Base/BERT-Large but does not specify any hardware details such as GPU or CPU models, memory, or cloud computing instances used for experiments.
Software Dependencies No The paper mentions tools and models like "fast Text Crawl", "Res Net-152", "BERT-base-uncased model", "Adam", "bi LSTM-CRF", and "Viterbi algorithm", but it does not provide specific version numbers for any software dependencies.
Experiment Setup Yes Table 2: Hyperparameters of the Rp BERT and bi LSTM-CRF models. Hyperparameter Value LSTM hidden state size 256 +Rp BERT 1024 LSTM layer 2 mini-batch size 8 char embedding dimension 25 optimizer Adam learning rate 1e-4 learning rate for finetuning Rp BERT and Res Net 1e-6 dropout rate 0.5