Disambiguating Spatial Prepositions Using Deep Convolutional Networks
Authors: Kaveh Hassani, Won-Sook Lee
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The comprehensive experimental results suggest that the combination of the hybrid feature and a convolutional neural network outperforms state-of-the-art methods and reaches the accuracy of 94.21% and F1-score of 0.9398. |
| Researcher Affiliation | Academia | Kaveh Hassani, Won-Sook Lee School of Electrical Engineering and Computer Science, University of Ottawa, Canada kaveh.hassani@uottawa.ca and wslee@uottawa.ca |
| Pseudocode | No | The paper describes the architecture of the convolutional neural network in Figure 1 and outlines the training process, but it does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any statement about releasing source code or provide a link to a code repository for the methodology described. |
| Open Datasets | Yes | The Pattern Dictionary of English Prepositions (PDEP) (Litkowski 2014) is a publicly available lexical resource collected as a part of The Preposition Project (TPP) for the study of preposition behavior. |
| Dataset Splits | Yes | We split the dataset into a train set with 39,000 samples and a test set with 4,129 samples (2,000 spatial and 2,129 non-spatial samples). ... For this purpose, 10% of the training set is sampled as the dev set. |
| Hardware Specification | No | The paper mentions using the TensorFlow library for deep learning models and lists various machine learning classifiers, but it does not specify any hardware details such as CPU, GPU models, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions several software tools and libraries (scikit-learn, TensorFlow, Adam optimizer, genism, Stanford Core NLP, NLTK) but does not provide specific version numbers for any of them, which is necessary for reproducible software dependency description. |
| Experiment Setup | Yes | Both deep learning models are trained using Adam stochastic optimizer (Kingma and Ba 2015) with the learning rate of 1E-4 over the mini-batches of size 250. ...They also use dropout regularization (Srivastava et al. 2014) with probability of p=0.5 and batch normalization (Ioffe and Szegedy 2015) on the input layer. Both models also utilize the rectified linear units (Re LU) in their hidden layers and a softmax function in their output layer. The DNN model is defined as a four layer network (3 hidden layers + softmax layer). ... We fixed the number of epochs to 50,000, and used 32, 64, and 128 filters with sizes of 2 5, 3 5, and 5 5 with stride of 1. ...The sizes of the hidden layers are considered as hyper-parameters and are optimized using the random search. ... these sizes are set to 500, 800, and 1,000. ... The number of neighbors in the K-NN classifier is set to 10. The number of estimators and the learning rate in the Ada-boost classifier are set to 50 and 0.9, respectively. For the random forest classifier, the number of estimators is set to 15 and the Gini index is used to decide the split. ... The number of estimators of the bagging classifier is also set to 15. The penalty parameter of the SVM models is set to 0.98. |