盘点4种常用的推荐算法

导读:推荐系统大量使用了机器学习技术,本文简单介绍一下推荐系统常用的策略与算法。

作者:刘强

[En]

Author: Liu Qiang

来源:大数据DT(ID:hzdassuju)

[En]

Source: big data DT (ID:hzdashuju)

盘点4种常用的推荐算法

01 基于内容的推荐

推荐系统通过技术手段将主题与人关联,主题包含许多属性,用户会通过与主题的交互生成行为日志。通过这些行为日志,可以挖掘出衡量用户对主题的偏好的标签(将主题的属性赋予喜欢它的用户,这样用户就有了这个标签)通过这些偏好标签向用户推荐是一种基于内容的推荐算法。

[En]

The recommendation system associates the subject matter with people through technical means, and the subject matter contains many attributes, and the user will generate a behavior log through the interaction with the subject matter. Through these behavior logs, you can mine tags that measure users’ preference for the subject matter (give the attributes of the subject matter to the user who likes it, so that the user has this tag) * recommending users through these preference tags is a content-based recommendation algorithm. *

以视频推荐为例,视频中有标题、国家、年龄、演职人员、标签等信息。用户之前看过某些类型的视频,这意味着用户对这些视频感兴趣,比如用户对恐怖和科幻电影的偏好。通过这种方式,用户的电影偏好被贴上了恐怖和科幻的标签。我们可以根据这些兴趣特征向用户推荐恐怖和科幻电影。

[En]

Take video recommendation, for example, videos have title, country, age, acting staff, tags and other information. Users have seen certain types of videos before, which means that users are interested in these videos, such as users’ preference for horror and science fiction movies. In this way, users’ movie preferences are labeled as horror and science fiction. We can recommend horror and science fiction movies to users according to these interest characteristics.

02 协同过滤

用户在产品上的互动行为已经给用户留下了印记,我们可以用物以类聚,人以群分的简单思路,为用户提供个性化的推荐。

[En]

Users’ interactive behavior on the product has left a mark for users, and we can use the simple idea of “birds of a feather flock together, people are divided into groups” to provide personalized recommendations for users.

具体来说,[人分群组]是找到兴趣相同(行为相似)的用户,将这些兴趣相同的用户浏览的对象推荐给用户,这是一种基于用户的协同过滤算法。

[En]

Specifically, * “people are divided into groups” * is to find users with the same interests (who have had similar behavior) and recommend the objects browsed by these users with the same interests to users, which is a user-based collaborative filtering algorithm.

“物以类聚”就是如果有很多用户都对某两个标的物有相似的偏好,说明这两个标的物是”相似”的,我们可以通过推荐与用户喜欢过的标的物相似的标的物这种方式为用户提供个性化推荐,这就是基于物品的协同过滤推荐算法。

图1-2简单说明了这两类协同过滤算法。

盘点4种常用的推荐算法

图1-2:两种协同过滤推荐算法

[En]

Figure 1-2: two types of collaborative filtering recommendation algorithms

03 基于模型的推荐

一般来说,可以根据用户行为记录、用户相关信息(年龄、性别、地域、消费习惯等)构建算法模型。以及与主题相关的信息来预测用户对物品的偏好。常用算法包括Logistic回归、矩阵分解、分解器等

[En]

Generally speaking, you can build an algorithm model based on user behavior records, user-related information (age, sex, region and consumption habits, etc.) and subject matter-related information to predict users’ preference for items. * common algorithms include logistic regression, matrix decomposition, decomposer, etc. **

随着深度学习技术的发展,许多与深度学习相关的算法已经落地到推荐系统上,并取得了良好的效果。

[En]

With the development of deep learning technology, many algorithms related to deep learning have landed on the recommendation system and achieved good results.

04 基于社交关系的推荐

我们在日常生活中经常为别人或者要求别人给我们推荐书籍、 餐厅、电影等,这种推荐方式往往效果较好,大家也更容易接受。

微信“看一看”模块中的“看一看”,是通过向你展示微信好友看过的文章来实现推荐的。张小龙在2019年是微信八周年纪念日的微信公开课上说,在《看一看》模块中,《看》比《选》要好很多,而《选》是算法实现的推荐。

[En]

The “looking” in Wechat’s “take a look” module realizes the recommendation by showing you articles that your Wechat friends have read. Zhang Xiaolong said in the Wechat open class, which marks the eighth anniversary of Wechat in 2019, that “watching” is much better than the “selection” in the “take a look” module, and “selection” is the recommendation realized by algorithm.

在这些推荐算法中,基于内容的推荐和协同过滤推荐是最古老和最常用的推荐算法,它们相对简单有效,在行业中得到了广泛的应用。

[En]

Among these recommendation algorithms, content-based recommendation and collaborative filtering recommendation are the oldest and most commonly used recommendation algorithms, which are relatively simple and effective, and have been widely used in industry.

作者简介:刘强,硕士,2009年毕业于中国科学技术大学数学系。大数据拥有12年的推荐系统实践经验,精通企业级推荐系统建设。从零到一建成了拥有千万DAU视频APP的推荐体系,推荐体系占APP总流量的30%。在过去的3年里,我们为多家中小互联网公司(流媒体、在线教育、跨境电商等)提供了技术咨询。帮助他们建立从零到一的推荐系统。

[En]

About the author: Liu Qiang, master’s degree, graduated from the Department of Mathematics, University of Science and Technology of China in 2009. Big data has 12 years of practical experience with recommendation system, proficient in the construction of enterprise-level recommendation system. A recommendation system with tens of millions of DAU video APP has been built from zero to one, and the recommendation system accounts for 30% of the total APP traffic. In the past 3 years, we have provided technical advice to a number of small and medium-sized Internet companies (streaming media, online education, cross-border e-commerce, etc.) to help them build recommendation systems from zero to one.

本文摘自《构建企业推荐系统》,由出版商授权。

[En]

This article is excerpted from “Building an Enterprise recommendation system” and is authorized by the publisher.

盘点4种常用的推荐算法

《构建企业推荐体系》的延伸阅读

[En]

Extended Reading of “Building an Enterprise recommendation system”

点击上面的图片了解并购买

[En]

Click on the picture above to learn and purchase

转载请联系微信:DoctorData

[En]

For reprint, please contact Wechat: DoctorData

推荐语:推荐算法工程师必读。推荐系统专家集10年实践经验之作,从场景、算法、工程、运营、实践等多维度深度梳理构建企业级推荐系统的方法。

盘点4种常用的推荐算法

划重点????

干货直达????

更多精彩 ????

在公众号对话框中输入以下关键字

[En]

Enter the following * keywords * in the official account dialog box

查看更多优质内容!

[En]

See more quality content!

PPT | 读书 | 书单 | 硬核 | 干货| 讲明白| 神操作

大数据 | 云计算 | 数据库 | Python | 爬虫 | 可视化

AI | 人工智能 | 机器学习 | 深度学习 | NLP

5G| 中台| 用户画像| 1024 | 数学| 算法| 数字孪生

据统计,99%的明星都关注这个公众号。

[En]

According to statistics, 99% of the celebrities follow this official account.

????

Original: https://blog.csdn.net/zw0Pi8G5C1x/article/details/119620321
Author: 大数据v
Title: 盘点4种常用的推荐算法

原创文章受到原创版权保护。转载请注明出处:https://www.johngo689.com/6476/

转载文章受原作者版权保护。转载请注明原作者出处!

(0)

大家都在看

最近整理资源【免费获取】:   👉 程序员最新必读书单  | 👏 互联网各方向面试题下载 | ✌️计算机核心资源汇总