docs: forbidden link jump

This commit is contained in:
camera-2018
2024-02-09 03:05:08 +08:00
parent d952b7ac3e
commit 3bac8037ab
9 changed files with 15 additions and 16 deletions

View File

@@ -45,7 +45,7 @@ DSSM(Deep Structured Semantic Model)是由微软研究院于CIKM在2013年提出
## SENet双塔模型
SENet由Momenta在2017年提出当时是一种应用于图像处理的新型网络结构。后来张俊林大佬将SENet引入了精排模型[FiBiNET](https%3A//arxiv.org/abs/1905.09433)中其作用是为了将大量长尾的低频特征抛弃弱化不靠谱低频特征embedding的负面影响强化高频特征的重要作用。那SENet结构到底是怎么样的呢为什么可以起到特征筛选的作用
SENet由Momenta在2017年提出当时是一种应用于图像处理的新型网络结构。后来张俊林大佬将SENet引入了精排模型[FiBiNET](https://arxiv.org/abs/1905.09433)中其作用是为了将大量长尾的低频特征抛弃弱化不靠谱低频特征embedding的负面影响强化高频特征的重要作用。那SENet结构到底是怎么样的呢为什么可以起到特征筛选的作用
<div align=center>
<img src="https://camo.githubusercontent.com/ccf54fc4fcac46667d451f22368e31cf86855bc8bfbff40b7675d524bc899ecf/68747470733a2f2f696d672d626c6f672e6373646e696d672e636e2f32303231303730333136313830373133392e706e673f782d6f73732d70726f636573733d696d6167652f77617465726d61726b2c747970655f5a6d46755a33706f5a57356e6147567064476b2c736861646f775f31302c746578745f6148523063484d364c7939696247396e4c6d4e7a5a473475626d56304c336431656d6876626d6478615746755a773d3d2c73697a655f312c636f6c6f725f4646464646462c745f3730237069635f63656e746572" style="zoom:80%;"/>

View File

@@ -250,7 +250,7 @@ class ItemToItemBatchSampler(IterableDataset):
### 邻居节点采样
再得到训练样本之后接下来主要是在训练图上为heads节点采用其邻居节点。在DGL中主要是通过sampler_module.NeighborSampler来实现具体地通过**sample_blocks**方法回溯生成各层卷积需要的block即所有的邻居集合。其中需要注意的几个地方基于随机游走的重要邻居采样DGL已经实现具体参考**[dgl.sampling.PinSAGESampler](https://link.zhihu.com/?target=https%3A//docs.dgl.ai/generated/dgl.sampling.PinSAGESampler.html%3Fhighlight%3Dpinsagesampler)**其次避免信息泄漏代码中先将head → tails,head → neg_tails从frontier中先删除再生成block。
再得到训练样本之后接下来主要是在训练图上为heads节点采用其邻居节点。在DGL中主要是通过sampler_module.NeighborSampler来实现具体地通过**sample_blocks**方法回溯生成各层卷积需要的block即所有的邻居集合。其中需要注意的几个地方基于随机游走的重要邻居采样DGL已经实现具体参考**[dgl.sampling.PinSAGESampler](https://docs.dgl.ai/generated/dgl.sampling.PinSAGESampler.html)**其次避免信息泄漏代码中先将head → tails,head → neg_tails从frontier中先删除再生成block。
```python
class NeighborSampler(object): # 图卷积的邻居采样

View File

@@ -273,5 +273,5 @@ OK 这就是AutoInt比较核心的部分了当然上面自注意部分
**参考资料**
* [AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks](https://link.zhihu.com/?target=https%3A//arxiv.org/abs/1810.11921)
* [AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks](https://arxiv.org/abs/1810.11921)
* [AutoInt基于Multi-Head Self-Attention构造高阶特征](https://zhuanlan.zhihu.com/p/60185134)

View File

@@ -138,7 +138,7 @@ def ESSM(dnn_feature_columns, task_type='binary', task_names=['ctr', 'ctcvr'],
测试数据集:
adult[https://archive.ics.uci.edu/ml/datasets/census+income](https://link.zhihu.com/?target=https%3A//archive.ics.uci.edu/ml/datasets/census%2Bincome)
adult[https://archive.ics.uci.edu/ml/datasets/census+income](https://archive.ics.uci.edu/dataset/20/census+income)
将里面两个特征转为label完成两个任务的预测
@@ -157,6 +157,6 @@ https://www.zhihu.com/question/475787809
https://zhuanlan.zhihu.com/p/37562283
美团:[https://cloud.tencent.com/developer/article/1868117](https://link.zhihu.com/?target=https%3A//cloud.tencent.com/developer/article/1868117)
美团:[https://cloud.tencent.com/developer/article/1868117](https://cloud.tencent.com/developer/article/1868117)
Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate (SIGIR'2018)

View File

@@ -382,9 +382,8 @@ Progressive Layered Extraction (PLE): A Novel Multi-Task Learning (MTL) Model fo
https://zhuanlan.zhihu.com/p/291406172
爱奇艺:[https://www.6aiq.com/article/1624916831286](https://link.zhihu.com/?target=https%3A//www.6aiq.com/article/1624916831286)
爱奇艺:[https://www.6aiq.com/article/1624916831286](https://www.6aiq.com/article/1624916831286)
美团:[https://mp.weixin.qq.com/s/WBwvfqOTDKCwGgoaGoSs6Q](https://link.zhihu.com/?target=https%3A//mp.weixin.qq.com/s/WBwvfqOTDKCwGgoaGoSs6Q)
多任务loss优化[https://blog.csdn.net/wuzhongqi](https://link.zhihu.com/?target=https%3A//blog.csdn.net/wuzhongqiang/article/details/124258128)
美团:[https://mp.weixin.qq.com/s/WBwvfqOTDKCwGgoaGoSs6Q](https://mp.weixin.qq.com/s/WBwvfqOTDKCwGgoaGoSs6Q)
多任务loss优化[https://blog.csdn.net/wuzhongqi](https://blog.csdn.net/wuzhongqi)