HAN: Hierarchical Association Network for Computing Semantic Relatedness
Authors: Xiaolong Gong, Hao Xu, Linpeng Huang
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments are conducted on real datasets and the results show that our proposed method improves correlation precision compared with the state-of-the-art approaches. We conduct extensive experiments on a real dataset to evaluate the effectiveness of our proposed solution using HAN. The results show that our solution achieves improvement in correlation precision, compared with other state-of-the-art approaches. |
| Researcher Affiliation | Academia | Xiaolong Gong, Hao Xu, Linpeng Huang Department of Computer Science, Shanghai Jiao Tong University, Shanghai, China {gxl121438, insanelun, lphuang}@sjtu.edu.cn |
| Pseudocode | No | The paper does not contain explicitly labeled pseudocode or algorithm blocks. It describes methods and equations but not in pseudocode format. |
| Open Source Code | No | The paper does not provide a specific link or statement about releasing the source code for the methodology described. |
| Open Datasets | Yes | In this paper, we constructed two hierarchical association networks HANwiki and HANfree, HANwiki is based on the Wikipedia dump on October 2, 2015. You can download here: https://dumps.wikimedia.org/. The vertices of HANfree are based on the Florida norms (see Table 2). The original Florida free association norms data contains 5019 cue words and a total of 72,176 cue-target pairs. This training set is generated by a well-studied psychological process called free association. Table 2 shows a fragment of the free association norms collected by University of South Florida (Nelson, Mc Evoy, and Schreiber 2004). |
| Dataset Splits | No | The paper mentions 'training set Florida Norms' and 'Conceptual test set: Concept Rel-250' but does not specify explicit training/validation/test splits (e.g., percentages or counts) for the main datasets (MC, RG, WS353) used for evaluation. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments. It only mentions general environment like 'wiki set'. |
| Software Dependencies | No | The paper mentions using 'word2vec (Mikolov et al. 2013)' and 'GloVe (Pennington, Socher, and Christopher 2014)' but does not list specific version numbers for these or other software dependencies required to reproduce the experiments. |
| Experiment Setup | Yes | Parameters tuning We now evaluate the performance of HAN performance by varying the following parameters: The parameter K1 and K2 controls the context size in word co-occurrence Fw and concept relatedness Fc respectively. (Section 3) The parameter β trades off the importance of two related items. (Section 3) The parameter λ adjusts the contribution between Fw and CREL(wi, wj). We found the optimal correlation is obtained when relatedness of concept pairs are taken with the fixed-size window K2 = 10 and β = 0.85. The optimal result is obtained when λ = 0.2. We selected 100 as an vector dimensionality. |