Learning Greedy Policies for the Easy-First Framework
Authors: Jun Xie, Chao Ma, Janardhan Rao Doppa, Prashanth Mannem, Xiaoli Fern, Thomas G. Dietterich, Prasad Tadepalli
AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate our approach in two NLP domains: within-document entity coreference resolution and cross-document joint entity and event coreference resolution. Our results demonstrate that the proposed approach achieves statistically significant performance improvement over the baseline training approaches for the Easy-first framework and is less prone to overfitting. |
| Researcher Affiliation | Academia | School of Electrical Engineering and Computer Science, Oregon State University {xie,machao,mannemp,xfern,tgd,tadepall}@eecs.oregonstate.edu School of Electrical Engineering and Computer Science, Washington State University jana@eecs.wsu.edu |
| Pseudocode | Yes | Algorithm 1 Easy-first inference algorithm with learning option. ... Algorithm 2 The MM algorithm to solve Equation 4 |
| Open Source Code | No | The paper does not provide concrete access to source code (specific repository link, explicit code release statement, or code in supplementary materials) for the methodology described in this paper. |
| Open Datasets | Yes | ACE04: (NIST 2004) We employ the same train-ing/testing partition as ACE2004-CULOTTA-TEST (Culotta et al. 2007; Bengtson and Roth 2008). ... Onto Notes-5.0: (Pradhan et al. 2012) We employ the of-ficial split for training, validation, and testing. ... We employ the benchmark EECB corpus (Lee et al. 2012) for our experiments. |
| Dataset Splits | Yes | Onto Notes-5.0: (Pradhan et al. 2012) We employ the of-ficial split for training, validation, and testing. There are 2802 documents for training; 343 documents for validation; and 345 documents for testing. ... ACE04: (NIST 2004) ... 268 documents are used for training, 68 documents for validating, and 107 documents for testing. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions external tools like 'Stanford multi-pass sieve system' and 'Co NLL scorer (7.0)' but does not provide specific version numbers for its own implementation's software dependencies (e.g., programming languages or libraries with versions). |
| Experiment Setup | Yes | For BGBB, we tune the learning rate (η {10 1, ..., 10 5}) and the maximum number of repeated perceptron updates (k {1, 5, 10, 20, 50}) for each mistake step. For RBGVB and RBGBB, we tune the regularization parameter (λ {10 4, 10 3, ..., 103}). For MM-based method including BGVB, RBGVB, RBGBB, we tune the maximum number of MM iterations (T {1, 5, 10, 20, 50}) and the maximum number of gradient descent steps (t {1, 5, 10, 20, 50}). |