Bridging Recommendation and Marketing via Recurrent Intensity Modeling
Authors: Yifei Ma, Ge Liu, Anoop Deoras
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We run experiments where we use marketing as an alternative to coldstart item exploration, by setting a minimal-exposure constraint for every item in the audience base.Our experiments are available at https://github.com/ awslabs/recurrent-intensity-model-experiments |
| Researcher Affiliation | Industry | Yifei Ma, Ge Liu & Anoop Deoras AWS AI Labs {yifeim,gliua,adeoras}@amazon.com |
| Pseudocode | Yes | Algorithm 1 Dual planner for online (and offline) matching |
| Open Source Code | Yes | Our experiments are available at https://github.com/ awslabs/recurrent-intensity-model-experiments |
| Open Datasets | Yes | Movielens (ML) (Harper & Konstan, 2015)..., Netflix (NF)2... 2https://www.kaggle.com/netflix-inc/netflix-prize-data, Yoochoose (YC)3... 3https://www.kaggle.com/phhasian0710/yoochoose |
| Dataset Splits | Yes | We hold out time windows only on the test users (Group-B in Table S1)... All training users (Group A) and the observed histories of the testing users (Group B left part) are considered training data.RNN-HP and GCMC(*) require further splitting of the training set. On NF, we create a set-back window between [T , T) from all users and on ML/YC, we keep the same time [T, T + T) but change the user base to Group A for validation. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used to run the experiments, such as GPU or CPU models. |
| Software Dependencies | No | The paper mentions 'tick software package', 'Implicit package', and 'Light FM package' but does not specify their version numbers. |
| Experiment Setup | Yes | We fix α = 1% in equation 9 and vary 0 β 1% as item min-exposure constraints.Algorithm 1 Dual planner for online (and offline) matching Require: λxy = λ(x, y), ideally scaled to λ 1; user-capacity α = K/Y; item-constraint β; user-state distribution P(X) from a past period of time; step-size γ... init ˆvy Unif( 1, 0), y Y for k in [0, 1, . . . , 100] do set ϵ = 0.8k; |