Latent Emotion Memory for Multi-Label Emotion Classification

Authors: Hao Fei, Yue Zhang, Yafeng Ren, Donghong Ji7692-7699

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on two benchmark datasets show that the proposed model outperforms strong baselines, achieving the state-of-the-art performance.
Researcher Affiliation Academia 1Key Laboratory of Aerospace Information Security and Trusted Computing, Ministry of Education, School of Cyber Science and Engineering, Wuhan University, Wuhan, China 2School of Engineering, Westlake University, Hangzhou, China 3Guangdong Collaborative Innovation Center for Language Research & Services, Guangdong University of Foreign Studies, Guangzhou, China
Pseudocode No The paper describes the model architecture and training process in text and diagrams, but does not provide pseudocode or a clearly labeled algorithm block.
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We conduct experiments on two benchmark datasets, including the English dataset Sem Eval2018 (Mohammad et al. 2018) and the Chinese dataset Ren-CECps (Quan and Ren 2010).
Dataset Splits Yes Statistics of datasets is shown in Table 1. In our experiments, the latent emotion module takes emotional Bo W as input. We first filter out stopword tokens1, keeping the words existed in a sentiment dictionary as the lexicons. For English lexicon, we employ GI (Stone, Dunphy, and Smith 1966), LIWC (Pennebaker, Francis, and Booth 2001), MPQA (Wilson, Wiebe, and Hoffmann 2005), Opinion Lexicon (Hu and Liu 2004) and Senti Word Net (Baccianella, Esuli, and Sebastiani 2010). For Chinese lexicon, we use How Net. Note that we also keep the emotion-rich symbols (e.g. emoji, emoticons). For the memory module, we keep the original token sequence as the input. Table 1: Statistics of datasets. Avg.len. is the average length of sentences. Emo. denotes the numbers of emotion categories. #N co.e.l. denotes the N numbers of co-existing emotion labels in one sentence. Dataset Sent. Words Avg.len. Emo. 3 co.e.l.(%) 2 co.e.l.(%) 1 co.e.l.(%) Train Dev Test Ren-CECps 35,096 228,455 24.56 8 1,824(5.2) 11,416(32.5) 18,812(53.6) 24,567 3,510 7,019 Sem Eval2018 10,983 32,557 16.04 11 3,419(31.1) 4,442(40.4) 1,563(14.2) 6,838 886 3,259
Hardware Specification No The paper does not provide specific hardware details (e.g., specific GPU/CPU models, memory amounts) used for running its experiments, only mentioning general terms like "trained on 6 billion words from Wikipedia and web text" and "train 300-dimensional word embeddings on Chinese Wikipedia...using word2vec".
Software Dependencies No The paper mentions "gensim package" with a URL, but does not provide specific version numbers for it or other software dependencies used in the experiments.
Experiment Setup Yes We set the same max length for Xe Bo W and X. In the learning process, we set 300 epochs for pre-training latent emotion module, and 1000 total training epochs with early-stop strategy (Caruana, Lawrence, and Giles 2001). To mitigate overfitting, we apply dropout with a rate of 0.01. We use Adam (Kingma and Ba 2014) for the optimization with the initial rate of 0.001.