[推荐系统] 笔记 01 - 初识

课程：【推荐系统 python】推荐系统从入门到实战，18课时，based on Python。

油管基础：Recommender System

 

The Search & Recommendations Group is working to enhance its search retrieval and relevance capabilities. We are expanding our use of ML-based approaches as we continue to scale up across languages and markets, design content types, and creator marketplace contributions.

The Core Technical Pieces To Support These Capabilities Include

Indexing - Visual content representation and content understanding
Retrieval - Query understanding, language understanding
Ranking - Topical, contextual, personalised and business objective feature modelling and ranking systems
User experience - Universal search systems, diversity-aware ranking
Query assistance - Autocomplete, popular and related searches
Metrics and experimentation - Development of sensitive offline and online metrics and more efficient and predictive experimentation systems.

Responsibilities

Working in one or more of the search layer areas listed above
Applying knowledge of information retrieval technologies, e.g., OpenSearch, ElasticSearch, Solr, Learning to Rank algorithms and toolkits
Building scalable ML solutions that meet our SLA guidelines, beyond just ML model training
Model deployments and feature engineering as part of large-scale systems using a service-oriented architecture
Analytical skills with hypothesis-driven problem solving and turning data into actionable insights
Practical and ethical considerations of ML data sets for training and evaluation

Background

Requirement to have worked in search, ranking, ads, etc.
Good knowledge in one or more of the following areas: machine learning, learning to rank, information retrieval, search-specific experimentation and metrics.
(Ideal) Experience at working in hyper-growth companies that incorporate search or recommendations as part of a product experience (high growth teams, rapidly evolving requirements, and building E2E ranking systems)
(Ideal) Specific image/video search experience and/or image/video understanding and feature representation via state-of-the-art models.
(Bonus) Interest & experience in responsible AI considerations with ML-based systems.

 

 

Ref: https://www.youtube.com/playlist?list=PLmOn9nNkQxJE3UX1L1bkI23mSJr5afIeL

10:03 / 38:03 亚马逊牛逼，Netflix牛逼。

14:17 / 38:03 不同业务场景不同推荐方案。

23:12 / 38:03 三个思考角度

 

 29:02 / 38:03 收集分析数据

用户：个人信息、喜好标签、（上下文信息，例如浏览器 cookie）

物品：内容信息、分类标签、关键词

用户的行为：对物品的偏好，评分，查看记录，购买记录等。

 

Ref: https://www.youtube.com/watch?v=osPyGEgZR_I&list=PLmOn9nNkQxJE3UX1L1bkI23mSJr5afIeL&index=2

2:50 / 31:03

 

7:48 / 31:03 个性化推荐系统，根据数据源分类 为主流。

16:46 / 31:03

协同过滤，同时跟“用户”和“物品”都有关系，就是一张评价表。

 

 

/* implement */