Post 1: Derived Social Recommendation

Review

In the first lecture, we had an overview about social media and social networking. From the lecture, I learnt about different characteristics of social networking, such as the user age. And I found that mobile apps are bringing more and more influence to people in daily life, especially in the aspect of social networking.

In the second lecture, we discussed about some kinds of social tasks and social media, and then go to social media marketing. I was surprised to found that there are already a large number of different social tasks in our life, and various companies and people are benefited by ideas provided by different individuals.

My Topic: Derived Social Recommendation

I am especially interested in the topic of derived social recommendations. Along with the rapid development of different social networking applications, techniques of derived social recommendations are applied more and more widely. Derived social recommendation is a system used to provide different recommendations to different individuals according to their personalized information.

Types and Applications

There are different types of derived social recommendations: collaborative filtering (including user-based filtering and item-based filtering) and content-based filtering. Let me show you some applications with these different types.

For user-based filtering, it groups similar users and find out what they like, and then recommends items that a lot of users like to another similar user. For content-based filtering, it groups similar items and recommends high correlated items to individuals. Such collaborative filtering is used on people who indicated their “targets”, for example, friends relations like Facebook, product purchase records like Taobao and Amazon, and music sharing platform like Douban FM, Last.fm. The advantage of such filtering is that it provides personalized choice to users, and you can find what you like easily by this system. The more items you indicated that you like, you will get more customized recommendations.

A part of recommendation page of Taobao, obviously using user-based filtering.

My friend recommendation interface of the app “LINE”, although I don’t know which collaborative filtering it is using, I still see some friends’ name (and I don’t have their numbers!) This is because I have a lot of common friends with them.

For content-based filtering, it relies more on an item called “tag cloud”. When you read a piece of blog or news article, for example, an article introducing Hong Kong food, you will see recommended articles about Hong Kong food too. This is because they have similar tags, and system tends to mark them as high correlated items. Of course, they are more applied on news sites, blog providers, since they are usually described in words. However, other contents can also apply this system as long as they can be described in words and categories! Its advantage is obviously that it doesn’t rely heavily on the “user-target” relationship, and you don’t need to indicate a lot of items you like in order to see a new one. But it may need more artificial work, such as tagging the items.

A news article and its related articles grabbed from Yahoo! (Original Article)

Drawbacks It Causes

Well, benefits it brings are well-known: it helps people discover new items they like easily. Therefore, I will talk more about its drawbacks.

1. Biased information

Do you know that your internet browser is filtering your information when you are unconscious? If you don't believe, just open Google and search a keyword, then ask a friend in another country to do so, you will find the results are different. The author Eli Pariser expresses his worry in his book The Filter Bubble: What the Internet is Hiding From You [1]. When the internet presents a biased world to you, your view will also be biased. For example, if you are politically radical, and you may find all radical post on Facebook, and conservative views will be hidden. This is somehow scary since you will be more and more radical, and may finally fail to judge fairly. In another way, does this process make people separated into different circles and communicate different ideas less and less?

2. Inappropriate tags

As I mentioned above, in content-based filtering, more artificial work may be needed such as tagging items. For simplicity purpose, many websites let users do the work of tagging by themselves, like blog providers will let the blog holders tag their own articles. Another example is Douban FM, which develops an application called Doublo and encourages users to tag music in a game. However, when users’ goal is to achieve more coincidence tags, they may just try words randomly, regardless of whether the word can represent that piece of music. Hence, we can see that, due to various reasons, users may tag items inappropriately. When an item is tagged inappropriately, it may encounter difficulties in matching users and items.

Description about Doublo

OK, I will end my topic here. Thank you for reading this piece of article and all comments are welcomed! =)

Reference:

[1] The dangers of the internet: Invisible sieve, The Economist