FeiFei's Blog

DeepFM draws inspiration from the structure of the Wide & Deep model but replaces the Wide component with the Factorization Machine (FM) model. This eliminates the need for manual feature engineering. What makes DeepFM particularly clever is that it shares weights between the second-order part of the FM model and the embedding layer of the neural network. This weight sharing reduces the number of parameters significantly and speeds up the training process. DeepFM从Wide & Deep模型的结构中汲取灵感，但将Wide组件替换为因子分解机（FM）模型。这消除了手动特征工程的需求。DeepFM特别巧妙之处在于它在FM模型的二阶部分和神经网络的嵌入层之间共享权重。这种权重共享显著减少了参数数量并加快了训练过程。

a model that could both learn high-order features and capture non-linear relationships effectively, which led to the development of the NFM model. NFM模型的开发是为了创建一种既能够有效学习高阶特征又能够捕获非线性关系的模型。

its most notable feature is the use of an attention network to learn the importance of different combinations of features.assign a weight to each combination feature (cross feature). 它最显著的特点是使用了一个注意力网络来学习不同特征组合的重要性，为每个特征组合（交叉特征）分配权重。

Din-model is a term used in the field of computer vision and machine learning. It refers to a type of neural network architecture that is specifically designed for image segmentation tasks. Din-models are known for their efficiency and accuracy in segmenting images into regions of similar pixels. They have been used in a variety of applications, including medical imaging, autonomous driving, and facial recognition. "Din模型"是计算机视觉和机器学习领域中使用的一个术语。它指的是一种专门设计用于图像分割任务的神经网络架构。Din模型以其在将图像分割成相似像素区域方面的高效性和准确性而闻名。它已经被应用于各种领域，包括医学成像、自动驾驶和人脸识别等应用。

整体介绍在聚类算法领域，有几种基本方法，每种方法可能会导致一个或多个具体的算法。这些基本方法包括： In the field of clustering algorithms, there are several fundamental approaches, each of which may lead to one or more specific algorithms. These fundamental approaches are: 基于划分的算法; 基于层次的算法; 基于密度的算法; 基于网格的算法; 基于约束的算法; Partition-Based Algorithms; Hierarchical-Based Algorithms; Density-Based Algorithms; Grid-Based Algorithms; Constraint-Based Algorithms; 这些基本方法中的每一种都产生了许多衍生算法，这些衍生算法在学术界和工业界都找到了应用。这些衍生算法通常提供了对特定类型的数据或聚类挑战的改进或专门能力。 Each ...

Improved Learning of Category Feature Embeddings; Incorporation of Second-Order and High-Order Feature Interactions 改进的类别特征嵌入学习；包括二阶和高阶特征交互的融合。

The FM model (Factorization Machine) is an improvement over Poly2 to better handle data sparsity issues. Additionally, the FM model employs matrix factorization techniques, allowing it to train models with near-linear time complexity efficiency FM模型（因子分解机）是对Poly2的改进，以更好地处理数据稀疏性问题。此外，FM模型采用了矩阵因子分解技术，使其能够以近线性时间复杂度的高效性进行模型训练。

These are various categorized knowledge points about attention as summarized by the website's author, aiming to provide assistance and insights to the readers. 以下是网站作者总结的有关注意力的各种分类知识点，旨在为读者提供帮助和见解。

DIEN extracts the user's interest sequence based on their historical behavior. It aims to understand how a user's interests change or evolve when considering a specific item. DIEN（深度兴趣演化网络）根据用户的历史行为提取用户的兴趣序列，旨在了解用户在考虑特定项目时兴趣如何变化或演化。

CTR modeling involves predicting the probability that a user will click on a particular item or ad when presented with a set of options. CTR models are used to personalize content and ads for users, improve user engagement, and optimize advertising campaigns. CTR（点击率）建模涉及预测用户在提供一组选项时是否会点击特定项目或广告的概率。CTR模型用于为用户个性化内容和广告，提高用户参与度，并优化广告活动。

The "slimming down" of Alibaba's Taobao advertising CTR model 阿里巴巴淘宝广告CTR模型的“瘦身”

Exploring Advances in Click-Through Rate Estimation in the Era of Deep Learning 探索在深度学习时代点击率估计的进展

Deep Learning-Based Advertising CTR Estimation Algorithm

Discussing the Development Cycle of Deep Learning-Powered Advertising and Recommendation Technologies

提出了一种单一模型能够适用于多种不同业务场景的方法。 Proposed a method for a single model to be applicable to multiple different business scenarios.