Learning to Give Feedback: Modeling Attributes Affecting Argument Persuasiveness in Student Essays
Authors: Zixuan Ke, Winston Carlile, Nishant Gurrapadi, Vincent Ng
IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Using this corpus, we train the first set of neural models that predict the persuasiveness score of an argument in a student essay as well as the scores of its various attributes. Persuasiveness scoring results of the three models on the development set obtained via five-fold cross validation are shown in Table 5. |
| Researcher Affiliation | Academia | Zixuan Ke, Winston Carlile, Nishant Gurrapadi, Vincent Ng Human Language Technology Research Institute, University of Texas at Dallas Richardson, TX 75083-0688 {zixuan,winston}@hlt.utdallas.edu, Nishant.Gurrapadi@utdallas.edu, vince@hlt.utdallas.edu |
| Pseudocode | No | The paper describes neural network architectures and model components, but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured, code-like steps. |
| Open Source Code | No | The paper mentions that a corpus is publicly available: 'we have recently annotated and made publicly available a corpus of persuasive student essays [Carlile et al., 2018]', but it does not provide an explicit statement or link for the source code of the models (Baseline, Pipeline, Joint) developed in this paper. |
| Open Datasets | Yes | The corpus we use is composed of 102 essays randomly chosen from the Argument Annotated Essays corpus [Stab and Gurevych, 2014a]. These essays were taken from essayforum2, a site offering feedback to students wishing to improve their ability to write persuasive essays for tests. |
| Dataset Splits | Yes | We first randomly partition our 102 essays into five folds, each of which contains 20 21 essays, and then conduct five-fold cross-validation experiments. In each fold experiment, we employ three folds for training, one fold for development, and one fold for testing. |
| Hardware Specification | No | The paper describes the models and experimental setup but does not specify any hardware details such as GPU models, CPU types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions using 'Facebook Fast Text pre-trained word embeddings' and 'bi LSTMs' but does not provide specific version numbers for any software dependencies or libraries used in the implementation. |
| Experiment Setup | Yes | The resulting vector first goes through a dense layer, which reduces the vector s dimension to 150 (with Leaky Re LU as the activation function), then goes through another dense layer for scoring (again with Leaky Re LU as the activation function). To represent the words, we use the 300-dimensional Facebook Fast Text pre-trained word embeddings [Bojanowski et al., 2017]. The network is trained to minimize mean absolute error. Early stopping is used to choose the best epoch. Specifically, training stops when the loss on development data stops improving after 20 epochs. |