Jointly Modeling Topics and Intents with Global Order Structure

Authors: Bei Chen, Jun Zhu, Nan Yang, Tian Tian, Ming Zhou, Bo Zhang

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform experiments in both unsupervised and supervised settings, results show the superiority of our model over several state-of-the-art baselines.
Researcher Affiliation Collaboration Dept. of Comp. Sci. & Tech., State Key Lab of Intell. Tech. & Sys., Center for Bio-Inspired Computing Research, TNList, Tsinghua University, Beijing, 100084, China Microsoft Research Asia, Beijing, 100080, China
Pseudocode Yes As shown in Fig.3, we present an approximate three-step algorithm to obtain the canonical permutation π0. Step 1: We compute π d for each labeled document sd... Step 2: We introduce variables gij(i, j [K])... Step 3: We obtain the π0 by calculating the topological sequence of G.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We use two real datasets: 1) Chemical (Guo et al. 2010): It contains 965 abstracts of scientific papers...; and 2) Elements (Chen et al. 2009): It consists of 118 articles from the English Wikipedia...
Dataset Splits No For supervised classification, the paper states 'we randomly choose 20% documents; annotate their sentences with intent labels; and use them for training. Our goal is to learn the intent labels for the sentences in the remaining 80% documents.' However, it does not explicitly mention a separate validation set or split.
Hardware Specification No The paper does not provide any specific hardware details (e.g., CPU/GPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions 'SVMLight tools (Joachims 1998)' but does not provide specific version numbers for it or any other software dependencies.
Experiment Setup Yes For hyperparameters, we set θ0 = 0.1, λ0 = 0.1, α0 = 0.1, β0 = 0.1 and γ0 = 1... ν0 is set to be 0.1 times the number of documents in the corpus. For EGMM-LDA, we set the regularization parameter c to be 0.1. We set ρ0 = 2 for all the experiments except for Elements with K = 10, in which we set ρ0 = 1.