An Association Network for Computing Semantic Relatedness

Authors: Keyang Zhang, Kenny Zhu, Seung-won Hwang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our evaluation results validate that simple algorithms on this network give competitive results in computing semantic relatedness between words and between short texts.This section primarily evaluates two association networks, one constructed only using the original free association norms (denoted as ANfree), and the other constructed through the approach proposed in Section (denoted as ANwiki).
Researcher Affiliation Academia Keyang Zhang 1 and Kenny Q. Zhu 2 Shanghai Jiao Tong University, Shanghai, China 1keyangzh@gmail.com, 2kzhu@cs.sjtu.edu.cn Seung-won Hwang POSTECH, Pohang, Republic of Korea swhwang@postech.ac.kr
Pseudocode Yes Algorithm 1 Generate super node
Open Source Code No 3A demo of our system is available at http://adapt.seiee.sjtu. edu.cn/ keyang/assoc/.
Open Datasets Yes The original Florida free association norms data contains 5,019 cue words (which form the set of normed words) and a total of 72,176 cue-response pairs. ... known as Florida Norms from now on. Our test set for evaluting term relatedness is the well-known Word Similarity-353 (Finkelstein et al. 2002) (a.k.a. WS-353 with 353 word pairs) For testing short text similarity, we use the well-known public set Li30 (Li et al. 2006), comprising 30 pairs of short texts. A newly constructed dataset STSS-131 (O shea, Bandar, and Crockett 2013) is used to tune the parameter K decribed in Algorithm 2.
Dataset Splits No The paper mentions various datasets (Florida Norms, WS-353, Li30, STSS-131) but does not provide specific details on training, validation, and test splits (e.g., percentages, sample counts, or explicit splitting methodologies) for any of them.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies or version numbers for libraries, frameworks, or programming languages used in the implementation.
Experiment Setup Yes Recall that, Algorithm 2 is parameterized by K determining the extent of expansion. Our reported results use K = 10, empirically tuned based on STSS-131 dataset. As a result, we follow (Wettler 1993) to set α to be 0.66, which, according to them, perform the best in estimating word association.