Using Graphs of Classifiers to Impose Declarative Constraints on Semi-supervised Learning

Authors: Lidong Bing, William W. Cohen, Bhuwan Dhingra

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experiment with two classes of tasks, link-based text classification and relation extraction. We show modest improvements on well-studied link-based classification benchmarks, and state-of-the-art results on relation-extraction tasks for two realistic domains.
Researcher Affiliation Collaboration Lidong Bing AI Lab Tencent Inc. lyndonbing@tencent.com William W. Cohen Machine Learning Department Carnegie Mellon University wcohen@cs.cmu.edu Bhuwan Dhingra Language Technologies Institute Carnegie Mellon University bdhingra@cs.cmu.edu
Pseudocode Yes Figure 1: Declarative specifications of the models for supervised learning, on the left, and for a mutual-exclusivity constraint, on the right. (Contains pseudocode-like rules such as predict(X,Y) pick Label(Y) classify(X,Y).)
Open Source Code No The paper links to external tools (Spearmint, baseline models) that have open-source code, but it does not state that the authors' own D-Learner implementation or its source code is publicly released or available.
Open Datasets Yes We use three datasets from [Sen et al., 2008]: Cite Seer, Cora and Pub Med, with their statistical information given in Table 1.
Dataset Splits No The paper mentions training and testing sets, but does not explicitly describe a separate validation split or its size/proportion. It discusses tuning parameters using Bayesian optimization, which often implies validation, but no explicit split details for a validation set are provided in the text.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, memory).
Software Dependencies No The paper mentions the use of 'Pro PPR [Wang et al., 2013]' and 'Spearmint3', but does not provide specific version numbers for Pro PPR or other key software components used in their implementation.
Experiment Setup Yes After the examples are prepared, we employ Pro PPR to learn multi-class classifiers with α = 0.1. The maximum epoch number is 40, and we find the training usually converges in less than 10 epochs. ... As listed in Table 5, we have 10 parameters: the number of training relation examples (#R)...