TreeNet: Learning Sentence Representations with Unconstrained Tree Structure

Authors: Zhou Cheng, Chun Yuan, Jiancheng Li, Haiqin Yang

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate Tree Net achieves the state-of-the-art performance on all four typical text classification tasks.
Researcher Affiliation Academia Zhou Cheng1,2, Chun Yuan2, Jiancheng Li1,2 and Haiqin Yang3 1 Department of Computer Science and Technology, Tsinghua University 2 Graduate school at Shen Zhen, Tsinghua University 3 Department of Computing, Hang Seng Management College
Pseudocode No The paper describes the model with mathematical equations and descriptive text but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link regarding the public availability of its source code.
Open Datasets Yes MR: In the Movie Reviews dataset, each sentence is selected from one movie review with an assigned positive or negative label about reviewer s attitude.[Pang and Lee, 2005] 3https://www.cs.cornell.edu/people/pabo/movie-review-data/, the Subj (Subjectivity) dataset is also provided in this website. TREC: TREC QC consists of six typical question types...[Li and Roth, 2002]4http://cogcomp.cs.illinois.edu/Data/QA/QC/ CR: The Customer Reviews dataset...[Hu and Liu, 2004]5http://www.cs.uic.edu/ /liub/FBS/sentiment-analysis.html
Dataset Splits Yes CV means there is no standard train/dev/test split and a random 10-fold CV is used.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper mentions 'Stanford Parser' but does not specify a version number or any other software dependencies with their respective versions.
Experiment Setup Yes In the experiments without Glo Ve, namely Tree Net, the dimension of word embeddings is 100 and dimension of sentence representations is 50. In the experiments with Glo Ve,namely Tree Net glove, one word embedding is a 300-dimensional vector and a sentence representation is a 100-dimensional vector. In all these experiments, the model parameters are optimised through stochastic gradient descent over shuffled mini-batches with the Adam [Kingma and Ba, 2014] and batch size of 25. In order to get best performance in experiments, we conducted a grid search on the learning rate in the range of [1e-2, 1e-5] and L2 regularisation strength in the set of (1e-3, 1e-4, 1e-5, 0).