"Book Notes" recommendation System (Recommender systems An Introduction) Chapter I Introduction

A long time did not come up to write a blog, the first two months to catch the school recruit season, are busy school recruit.

I have bought this book long ago, but I have seen a bright "recommendation system practice", to see the directory structure of the book and Xiangliang that the same, has been left to see. Recently in a recommendation system, encountered some practical problems, want to open Xiangliang book to find inspiration, but for the things that they have seen, do not want to look at the second time again. It happened that the book was just like that of Xiangliang, and it was read directly from the book. By the way take notes, posted here for more people to refer to.

1. Recommendations for collaborative filtering

Basic idea: The user has the same preference in the past, e.g. they browse or buy the same books, and they have similar preferences in the future.

Key issues:

(1) How to find similar users

(2) How to measure similarity

(3) cold start problem

(4) Is there any other information available

2. Content-based recommendations

Recommended two purposes: (1) Inspire the user to choose an item or do something-for example, to entice a user to buy a book; (2) to solve the problem of information overload-for example, news recommendation, the user itself is to see, but the news too much or find trouble, it is best to the user want to see directly to him, save him.

The second objective is close to the field of information filtering and information retrieval. Content-based recommendation is essentially a method based on the field of information retrieval.

Core issues based on content recommendation: How to get a description of an item, whether manually generated or automatically extracted. Further, the user's behavior to the item, can also be converted into the user's "description" of the item, after accumulation, the user itself is equivalent to an item. Borrow the idea of information retrieval, the user as a query, then with this query similar items, after sorting, is recommended to the user's items.

Key issues:

(1) How to obtain and constantly update the user's records

(2) Calculation of similarity between goods and users

(3) How to automatically extract the item description

Advantages:

(1) There is no need for large-scale users (like collaborative filtering) to get the relationship between items

(2) Once the property of the item is obtained, the item can be recommended to the user immediately.

3. Knowledge-based recommendations

In some areas, such as the consumer electronics sector (e.g), the vast majority of data is a single purchase record. If you apply both of these methods, the data is too sparse to even get the recommended results. What if we have to recommend it to the user in this case?

The recommendation method based on knowledge (rule, inference) requires people to add some domain knowledge (such as rules) to the system.

Finish.

"Book Notes" recommendation System (Recommender systems An Introduction) Chapter I Introduction