Co-Occurrence Estimation from Aggregated Data with Auxiliary Information
Authors: Tomoharu Iwata, Naoki Marumo4247-4254
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of our proposed method using user-item rating datasets. We evaluated the proposed method using sushi and Movielens data sets. |
| Researcher Affiliation | Industry | Tomoharu Iwata, Naoki Marumo NTT Communication Science Laboratories, Kyoto, Japan {tomoharu.iwata.gy, naoki.marumo.ec}@hco.ntt.co.jp |
| Pseudocode | Yes | Algorithm 1: Estimation procedure with the proposed method. Input: marginal counts y, total count U, auxiliary information S, small records R, hyperparameter λ, batchsize T Output: estimation of co-occurrence counts X, neural network parameters Ψ, parameter α 1 Initialize parameters X, Ψ, α; 2 Calculate empirical marginal occurrence probability ˆθi = yi U for all items using y; 3 Calculate co-occurrence counts for small records X using R; 5 Sample a set of T item pairs Q randomly; 6 Calculate the objective function Eq.(10) for the sampled item pairs and its gradients Update parameters X, Ψ, α with a gradient-based optimization method; 7 until end condition is satisfied; |
| Open Source Code | No | The paper does not provide any specific links to open-source code for the methodology described, nor does it state that code is available in supplementary materials or upon request. |
| Open Datasets | Yes | The original sushi data were obtained from http://www.kamishima.net/sushi/. The original Movielens data were obtained from https://grouplens.org/datasets/movielens/. |
| Dataset Splits | No | The paper only explicitly states 'We randomly sampled the purchase records of ten users for the validation data.' without providing specific splits for training or testing sets, or a full breakdown of all data partitions. |
| Hardware Specification | Yes | The average computational time with the proposed method for training using the sushi data with I = 10, Movielens with I = 100, and Movielens with I = 200 were 0.12, 5.95 and 24.19 minutes, respectively, on computers with 2.60GHz CPUs, where we fixed U = 20. |
| Software Dependencies | No | The paper mentions 'PyTorch (Paszke et al. 2017)' as the implementation framework but does not specify the version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | For neural network ρ0( ), ρ1( ), φo( ), φ1( ), f01( ) in the EB+A, CGM+A and proposed methods, we used three-layer feed-forward neural networks with 32 hidden units. [...] We optimized using ADAM (Kingma and Ba 2015) with learning rate 10^-2, weight decay 10^-2 and dropout rate 0.1. For each batch, we randomly sampled 512 item pairs in training. The validation data were used for early stopping, where the maximum number of training epochs was 1,000. We used fixed hyperparameter λ = 0.5 in our experiments. |