Syntax-Based Deep Matching of Short Texts

Authors: Mingxuan Wang, Zhengdong Lu, Hang Li, Qun Liu

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

Reproducibility Variable Result LLM Response
Research Type Experimental We test our algorithm on the problem of matching a tweet and a response in social media, a hard matching problem proposed in [Wang et al., 2013], and show that DEEPMATCHtree can outperform a number of competitor models including one without using dependency trees and one based on word-embedding, all with large margins. and 5 Experiments We report our empirical study of DEEPMATCHtree and compare it to competitors, with a brief analysis and case studies.
Researcher Affiliation Collaboration Mingxuan Wang1 Zhengdong Lu2 Hang Li2 Qun Liu3,1 1Institute of Computing Technology, Chinese Academy of Sciences 2Noah s Ark Lab, Huawei Technologies 3Centre for Next Generation Localisation, Dublin City University
Pseudocode Yes Algorithm 1: Discriminative Mining of Parse Trees for Parallel Texts
Open Source Code No The paper provides a link for dataset access ('Data: data.noahlab.com.hk/conversation/') but does not state that the code for the described methodology is open-source or publicly released.
Open Datasets Yes The experiments are on two Weibo datasets in two settings. ... The second dataset, denoted as Data Labeled, consists of 422 tweets and around 30 labeled responses for each tweet 1, as introduced in [Wang et al., 2013] for retrieval-based conversation. 1Data: data.noahlab.com.hk/conversation/
Dataset Splits Yes The first dataset, denoted as Data Orignal, consists of 4.8 million (tweet, response) pairs. ... We use 485,282 original (tweet, response) pairs not used in the training for testing. and For each model, we use 5-fold cross validation to choose the hyper-parameter of the ranking model Rank SVM and report the best result.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions various algorithms and models (e.g., 'multi-layer perceptron (MLP)', 'Rank SVM', 'stochastic gradient descent (SGD)') but does not specify any software names with version numbers for replication.
Experiment Setup Yes As it shows, we have 1,000 units in the first hidden layer (sigmoid active function), 400 in the second hidden layer, 30 in the third hidden layer, and one in the output layer. and the performance peaks around Node Density=10 and In the experiments, we use m = 1.