Adaptive Co-attention Network for Named Entity Recognition in Tweets

Authors: Qi Zhang, Jinlan Fu, Xiaoyu Liu, Xuanjing Huang

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental To evaluate the proposed methods, we constructed a large scale labeled dataset that contained multimodal tweets. Experimental results demonstrated that the proposed method could achieve a better performance than the previous methods in most cases.
Researcher Affiliation Academia Qi Zhang, Jinlan Fu, Xiaoyu Liu, Xuanjing Huang Shanghai Key Laboratory of Intelligent Information Processing School of Computer Science, Fudan University 825 Zhangheng Road, Shanghai, P.R. China {qz, fujl16, liuxiaoyu16, xjhuang}@fudan.edu.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide explicit access to the authors' source code. It only links to third-party tools used for comparison (T-NER, Stanford NER).
Open Datasets No The authors state they "constructed a large-scale dataset" for the experiments but do not provide concrete access information (link, DOI, repository, or formal citation with authors/year) for this newly constructed dataset to make it publicly available.
Dataset Splits Yes We split the dataset into three parts: training set, development set, and testing set, which contain 4,000, 1,000, and 3,257 tweets, respectively. The named entity type counts in the training, development, and test sets are shown in Table 1.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions optimizers and parameters but does not specify software dependencies (e.g., libraries, frameworks) with version numbers.
Experiment Setup Yes The dimension for character embeddings is set to 30, and is initialized randomly from a uniform distribution of [ 0.25, 0.25]. The sentence length is set to 35, the word length is set to 30, and we apply truncating or zero-padding as necessary. In the character-level module, we use three groups of 32 filters, with window sizes of (2,3,4), while the output dimension of the bidirectional LSTM is set to 200. The optimizer is Rmsprop, and its learning rate is set to 0.19.