Long Short-Term Memory Over Recursive Structures
Authors: Xiaodan Zhu, Parinaz Sobihani, Hongyu Guo
ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We leverage the models for semantic composition to understand the meaning of text, a fundamental problem in natural language understanding, and show that it outperforms a state-of-the-art recursive model by replacing its composition layers with the S-LSTM memory blocks. We also show that utilizing the given structures is helpful in achieving a performance better than that without considering the structures. We experiment with the models on the Stanford Sentiment Treebank dataset (Socher et al., 2013) to determine the sentiment for different granularities of phrases in a tree. We experimentally show that S-LSTM outperforms a state-of-the-art recursive model by simply replacing the original tensor-enhanced composition with the S-LSTM memory block that we propose here. We show that utilizing the given structures is helpful in achieving a better performance than that without considering the structures. Table 1 shows the accuracies of different models on the test set of the Stanford Sentiment Treebank. |
| Researcher Affiliation | Collaboration | Xiaodan Zhu XIAODAN.ZHU@NRC-CNRC.GC.CA National Research Council Canada, 1200 Montreal Road M-50, Ottawa, ON K1A 0R6 CANADA Parinaz Sobhani PSOBH090@UOTTAWA.CA School of Electrical Engineering and Computer Science, University of Ottawa, 800 King Edward Avenue, Ottawa, ON K1N 6N5 CANADA Hongyu Guo HONGYU.GUO@NRC-CNRC.GC.CA National Research Council Canada, 1200 Montreal Road M-50, Ottawa, ON K1A 0R6 CANADA |
| Pseudocode | No | The paper provides mathematical equations (1)-(7) for forward computation and (9)-(14) for backpropagation. While these describe the algorithm, they are not presented as structured pseudocode or an algorithm block. |
| Open Source Code | No | The paper does not provide any statement or link indicating the availability of its source code. |
| Open Datasets | Yes | We specifically attempt to determine the sentiment of different granularities of phrases in a tree, within the Stanford Sentiment Treebank benchmark data (Socher et al., 2013). |
| Dataset Splits | Yes | For the root sentiment, the training, development, and test sentences are 8544, 1101, and 2210, respectively. The phrase sentiment task includes 318582, 41447, and 82600 phrases for the three sets. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the experiments (e.g., GPU model, CPU type, memory). |
| Software Dependencies | No | The paper mentions using "Ada Grad" for optimization, but it does not specify any software libraries or their version numbers (e.g., TensorFlow, PyTorch, scikit-learn, with their versions). |
| Experiment Setup | Yes | The S-LSTM results reported here were obtained by setting the size of the hidden units to be 100, batch size to be 10, and learning rate to be 0.1. |