Telepath: Understanding Users from a Human Vision Perspective in Large-Scale Recommender Systems
Authors: Yu Wang, Jixing Xu, Aohan Wu, Mantian Li, Yang He, Jinghe Hu, Weipeng Yan
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In practice, the Telepath model has been launched to JD s recommender system and advertising system and outperformed the former state-of-the-art method. For one of the major item recommendation blocks on the JD app, clickthrough rate (CTR), gross merchandise value (GMV) and orders have been increased 1.59%, 8.16% and 8.71% respectively by Telepath. For several major ad publishers of JD demand-side platform, CTR, GMV and return on investment have been increased 6.58%, 61.72% and 65.57% respectively by the first launch of Telepath, and further increased 2.95%, 41.75% and 41.37% respectively by the second launch. |
| Researcher Affiliation | Industry | Yu Wang, Jixing Xu, Aohan Wu, Mantian Li, Yang He, Jinghe Hu, Weipeng P. Yan Business Growth Division. JD.com {wangyu5, xujixing, wuaohan1, limantian, landy, hujinghe, paul.yan}@jd.com |
| Pseudocode | No | The paper provides architectural diagrams and descriptions of its components (CNN, RNN, DNN), including a table detailing the CNN subnetwork architecture. However, it does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | No | The paper does not contain any statements about releasing source code for the methodology described, nor does it provide any links to a code repository. |
| Open Datasets | No | The paper states it uses 'huge amounts of click-through log data daily, all weakly labeled' from 'JD's recommender system' and 'JD's online system'. This is internal data and no public access information, citations, or names of standard public datasets are provided. |
| Dataset Splits | No | The paper mentions 'all experiments are performed 5 times on training data and validation data' and that a 'dropout rate of 0.5 is applied' to avoid overfitting, implying a validation process. However, it does not provide specific details on the dataset splits (e.g., percentages or exact counts) for training, validation, or testing, nor does it mention cross-validation folds or specific files for custom splits. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU or CPU models, memory, or cloud instance types used for running its experiments. |
| Software Dependencies | No | The paper mentions using CNN, RNN, and DNN architectures, and optimization algorithms like FTRL and Ada Grad, but it does not specify any software names with version numbers (e.g., Python, TensorFlow, PyTorch versions) needed to replicate the experiment. |
| Experiment Setup | Yes | In the CNN subnetwork, a dropout rate of 0.5 is applied in all layers to avoid over-fitting. The momentum for the updating gradient is set to 0.9 and the weight decay is set to 0.0005 for regularization. We employ a global learning rate of 0.01 for all layers that decays by a factor of 0.1 after every 100K steps with a mini-batch size of 64. |