AutoSense Model for Word Sense Induction
Authors: Reinald Kim Amplayo, Seung-won Hwang, Min Song6212-6219
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experiments, we estimate the parameters of the model using collapsed Gibbs sampling and get the sense distribution of each instance as the WSI solution. We evaluate our model using the Sem Eval 2010 and 2013 WSI datasets (Manandhar et al. 2010; Jurgens and Klapaftis 2013). Results show that Auto Sense performs superior than previous state-of-the-art models. |
| Researcher Affiliation | Academia | Reinald Kim Amplayo, Seung-won Hwang, Min Song Yonsei University {rktamplayo, seungwonh, min.song}@yonsei.ac.kr |
| Pseudocode | No | The paper describes the generative process and inference using mathematical formulas but does not provide pseudocode or algorithm blocks. |
| Open Source Code | Yes | We share our data and code here: https://github.com/rktamplayo/Auto Sense. |
| Open Datasets | Yes | We use two publicly available datasets: Sem Eval 2010 Task 14 (Manandhar et al. 2010) and Sem Eval 2013 Task 13 (Jurgens and Klapaftis 2013). The Sem Eval 2010 dataset2 consists of 50 verbs and 50 nouns, each with different number of instances for a total of 8915 instances. Sem Eval 2013 dataset3 consists of 20 verbs, 20 nouns, and 10 adjectives, with a total of 4664 instances. |
| Dataset Splits | Yes | Auto Senses=X is a tuned and best version of the model, where the number of senses is tuned over a separate development set provided by the shared tasks and X is the tuned number of sense, different for each dataset |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU models, or memory specifications used for running the experiments. |
| Software Dependencies | No | For preprocessing, we do tokenization, lemmatization, and removing of symbols to build the word lists using Stanford Core NLP (Manning et al. 2014). |
| Experiment Setup | Yes | We set the hyperparameters to α = 0.1, β = 0.01, γ = 0.3, following the conventional setup (Griffiths and Steyvers 2004; Chemudugunta, Smyth, and Steyvers 2006). We arbitrarily set the number of senses to S = 15, and the number of topics T = 2S = 30, following (Wang et al. 2015). We set the number of iterations to 2000 and run the Gibbs sampler. |