深度学习驱动的神经推荐系统：从协同过滤到内容与上下文增强
需积分: 14 112 浏览量
更新于2024-07-10
收藏 3.18MB PDF 举报
“TKDE最新「神经推荐系统」综述论文，深度学习在计算机视觉和语言理解领域的成功推动了推荐系统向基于神经网络模型的转变，近年来神经推荐模型取得显著进展，超越传统推荐模型。”
这篇论文是IEEE Transactions on Knowledge and Data Engineering上的一篇关于神经推荐系统的综述。随着深度学习在计算机视觉和自然语言处理领域的突破性成果，推荐系统的研究也开始转向利用神经网络构建新型推荐模型。这些神经推荐模型由于神经网络的强大表征能力，能够更好地概括和超越传统的推荐算法。
作者们对神经推荐模型进行了系统性的回顾，其目标是总结这一领域的现状，以便推动未来的发展。不同于以往根据深度学习技术分类的方法，他们选择从推荐模型构建的角度来总结，这样的视角对于从事推荐系统研究和实践的人员更具指导意义。
文章的核心内容可能包括以下几个方面：
1. **协同过滤的神经网络实现**：传统的协同过滤方法主要依赖于用户行为历史和物品相似度，而神经网络可以学习更复杂的用户和物品表示，提高推荐的准确性和个性化。
2. **内容增强的推荐**：结合深度学习，神经推荐系统能从大量数据中提取物品内容特征，将这些信息融入到推荐过程中，提高推荐的关联性和多样性。
3. **上下文感知的推荐**：通过考虑用户的实时环境、时间、位置等上下文信息，神经推荐模型可以动态调整推荐结果，提供更加适时和情境相关的建议。
4. **模型架构与技术**：可能涵盖多层感知机（MLP）、卷积神经网络（CNN）、循环神经网络（RNN）、Transformer等模型在推荐系统中的应用，以及如何利用这些技术进行序列建模、注意力机制、自编码器等。
5. **优化与训练策略**：讨论神经推荐模型的损失函数设计、优化算法、正则化策略以及如何有效地处理大规模稀疏数据。
6. **评估与挑战**：介绍现有的评价指标，如精度、召回率、覆盖率、多样性等，并分析当前神经推荐系统面临的挑战，如冷启动问题、过拟合、解释性等。
7. **未来趋势与展望**：作者可能还对未来的研究方向进行了预测，比如深度强化学习在推荐中的应用、联邦学习以保护用户隐私、以及如何将模型的可解释性提升到新的水平。
这篇综述论文为读者提供了全面理解神经推荐系统发展现状和未来趋势的窗口，对于深入研究和应用推荐系统具有重要的参考价值。
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING 4
researchers argued that these neural graph based CF models
differ from the classical GNNs as CF models do not contain
any user or item features, and directly borrowing complex
steps such as embedding transformation, and non-linear
activations in GNNs may not be a good choice. Simplified
neural graph CF models, including LR-GCCF [32], and
LightGCN [33] have been proposed, which eliminate
unnecessary deep learning operations. These simplified
neural graph based models show superior performance in
practice without the need of carefully chosen activation
functions.
2.2 Interaction Modeling
Let p
u
and q
i
denote the learned embeddings of users
and items from representation models, this component
aims at interaction function modeling that estimates the
user’s preference towards the target item based on their
representations. In the following, we describe how to
model users’ predicted preference, denoted as ˆr
ui
based
on the learned embeddings. For ease of explanation, as
shown in Table 2, we summarize three main categories
for interaction modeling: classical inner product based
approaches, distance based modeling and neural network
based approaches.
Most previous recommendation models relied on
the inner product between user embedding and item
embedding to estimate the user-item pair score as: ˆr
ui
=
p
>
u
q
i
=
P
d
f=1
p
uf
q
if
. Despite its great success and
simplicity, prior efforts suggest that simply conducting
inner product would have two major limitations. First, the
triangle inequality is violated [38]. That is, inner product
only encourages the representations of users and historical
items to be similar, but lacks guarantees for the similarity
propagation between user-user and item-item relationships.
Second, it models the linear interaction, and may fail
to capture the complex relationships between users and
items [41].
2.2.1 Distance based Metrics
In order to solve the first issue, a line of research [38], [39],
[40] borrows ideas from the translation principles and uses
distance metric as the interaction function. The inherent
triangle inequality assumption plays an important role in
helping capture the underlying relationships among users
and items. For instance, if user u tends to purchase items i
and j, the representations of i and j should be close in the
latent space.
Towards this end, CML [38] minimizes the distance d
ui
between each user-item interaction < u, i > in Euclidean
space as: d
ui
= kp
u
− q
i
k
2
2
. Instead of minimizing the
distance between each observed user-item pair, TransRec
exploits the translation principle to model the sequential
behaviors of users [39]. In particular, the representation
of user u is treated as the translation vector between the
representations of the items i and the item j to visit next,
namely, q
j
+ p
u
≈ q
i
.
Distinct from CML that uses simple metric learning that
assumes each user’s embedding is equally close to every
item embedding she likes, LRML introduces the relation
vectors r to capture the relationships between user and item
pairs [40] . More formally, the score function is defined as:
s
ui
= kp
u
+ e − q
i
k
2
F
, (6)
where the relation vector e ∈ R
d
is constructed using
a neural attention mechanism over a memory matrix M.
M ∈ R
m×d
is the trainable memory module, hence E is the
attentive sum of m memory slots. As a result, the relation
vectors not only ensure the triangle inequality, but also
achieve better representation ability.
2.2.2 Neural network based Metrics
Distinct from the foregoing that employs linear the
metrics, recent works adopt a diverse array of neural
architectures, spanning from MLP, Convolutional Neural
Network (CNN), and AE as the main building block to mine
complex and nonlinear patterns of user-item interactions.
Researchers made attempts to replace similarity
modeling between users and items with MLPs, as MLPs
are general function approximators to model the any
complex continuous function. NCF is proposed to model
the interaction function between each user-item pair
with MLPs as: ˆr
ui
= f
MLP
(p
u
||q
i
). Besides, NCF also
incorporates a generic MF component into the interaction
modeling, thereby making use of both linearity of MF and
non-linearity of MLP to enhance recommendation quality.
Researchers also proposed to leverage CNN based
architecture for interaction modeling. These kinds of
models first generate interaction maps via outer product
of user and item embeddings, explicitly capturing the
pairwise correlations between embedding dimensions [42],
[43]. These CNN based CF focuses on high order
correlations among representation dimensions. However,
such improvements on performance come at the cost of
increasing model complexity and time cost.
Besides, a line of research exploits AEs to fulfill the
blanks of the user-item interaction matrix directly in the
decoder part [20], [21], [22], [23], [44], [45], [46]. As
the encoder and decoder can be implemented via deep
neural networks, such stacks of nonlinear transformations
give the recommenders more capacity to model the user
representation from complex combinations of all historically
interacted items.
3 CONTENT-ENRICHED RECOMMENDATION
Besides the general user-item interaction information,
recommendation problems are often accompanied with
auxiliary data. The auxiliary data could be classified into
two categories: content based information and context-
aware data. Specifically, the first category of content
information is associated with users and items, including
general user and item features, textual content (a.k.a, item
tags, item textual descriptions and users’ reviews for items),
multimedia descriptions (a.k.a, images, videos, and audio
information), user social networks, and knowledge graphs.
In contrast, contextual information shows the environment
when users make item decisions, which usually denotes
descriptions that beyond users and items [2]. Contextual
information includes time, location, and specific data
that are collected from sensors (such as speed, and
剩余18页未读，继续阅读
2023-05-01 上传
2023-06-02 上传
syp_net
- 粉丝: 158
- 资源: 1187
我的内容管理
展开
最新资源
- 前端面试必问：真实项目经验大揭秘
- 永磁同步电机二阶自抗扰神经网络控制技术与实践
- 基于HAL库的LoRa通讯与SHT30温湿度测量项目
- avaWeb-mast推荐系统开发实战指南
- 慧鱼SolidWorks零件模型库：设计与创新的强大工具
- MATLAB实现稀疏傅里叶变换(SFFT)代码及测试
- ChatGPT联网模式亮相，体验智能压缩技术.zip
- 掌握进程保护的HOOK API技术
- 基于.Net的日用品网站开发：设计、实现与分析
- MyBatis-Spring 1.3.2版本下载指南
- 开源全能媒体播放器：小戴媒体播放器2 5.1-3
- 华为eNSP参考文档：DHCP与VRP操作指南
- SpringMyBatis实现疫苗接种预约系统
- VHDL实现倒车雷达系统源码免费提供
- 掌握软件测评师考试要点：历年真题解析
- 轻松下载微信视频号内容的新工具介绍
