Extended Discriminative Random Walk: A Hypergraph Approach to Multi-View Multi-Relational Transductive Learning

Authors: Sai Nageswar Satchidanand, Harini Ananthapadmanaban, Balaraman Ravindran

IJCAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experiments aim to show the effectiveness of the different aspects of the Extended DRW method construction of the attribute view hypergraph, inference with very few labeled samples; with multiple views and multiple relations; and when the data exhibits high class-imbalance. Towards this, we have run experiments on synthetic and real datasets.
Researcher Affiliation Academia Sai Nageswar Satchidanand, Harini Ananthapadmanaban, Balaraman Ravindran Indian Institute of Technology Madras, Chennai, India sainageswar@gmail.com, harini.nsa@gmail.com, ravi@cse.iitm.ac.in
Pseudocode Yes Algorithm 1 Extended Discriminative Random Walk
Open Source Code Yes Along with the code of EDRW, the details and code for synthetic graph generation is available at https://github. com/Harini A/EDRW.
Open Datasets Yes We have also used real-world data with varying number of views and relations, to compare our method against other multi-view and collective classification approaches. The details about the various datasets used can be found in Table 1. ... Web KB [Sen et al., 2008] ... Cora and Citeseer [Sen et al., 2008] datasets ... Twitter Olympics and Twitter Football [Greene, 2013]
Dataset Splits Yes For a given percentage of unknowns, experiments were run over multiple partitions of the dataset into training and test data, and the average of all the runs have been reported. ... For data with more than one graph or hypergraph, the weights for each of the graphs and hypergraphs were found using 5-fold cross-validation.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions methods like 'Hellinger distance decision trees' and 'transductive SVM' but does not provide specific software names with version numbers for implementation.
Experiment Setup Yes In all the experiments, we found that the performance was similar for walk lengths of 2, 3 and 4, reminiscent of results reported in [Callut et al., 2008]. The results reported here are for L = 2. For data with more than one graph or hypergraph, the weights for each of the graphs and hypergraphs were found using 5-fold cross-validation.