Recommender Systems

In recent lectures, we gained very useful knowledge about recommender systems, including fundamental knowledge, identifying the differences and similarities between different recommender systems, predicting ratings, how to evaluate recommender systems, etc.

First, we have reached a consensus that recommender systems are very useful and important in our daily life, since we make decisions every day, and we usually need the comments and feedback from others to help us make choices. This is also the reflection of human’s social attributes.

Since the user demands appeared, the recommender systems also appeared, like Facebook, Twitter, Amazon, Weibo, etc. Each recommender systems have their own design and more or less different from others. Let’s use Facebook and Twitter as an example.

The most direct and can be found immediately difference between Facebook and Twitter, in my opinion, is that they have different connection rules of users. Facebook requires the connections must me mutual, while twitter don’t.

In Facebook, users can share pictures, links, likes, and other things, while twitter can share them in a limited words.

Most important, twitter focuses on instant information, people can get the hot event at the first time in Twitter. Facebook mainly meet the demand of user’s daily life.

However, the two systems have the same goal, to help users communicate easily and widely. We can get more information in the below graph.

Computer Can Be More Suitable than People in Social Inference

In recent lectures, we have learned about a number of concepts in social psychology. According to the lecture, social psychology is about understanding people’s behavior in a social context.

In this blog, I would like to focus on the theory-Social Inference. Social inference refers to the way we process social information to form impressions of people and make judgments about them.

People must obtain sufficient information in order to make social judgments. In social situations, there are many ways access to information, such as people’s manners, social role, or relationship between group members. For example, a student can know a new classmate through his or her look, clothes, communication skills and so on.

However, people’s previous experience will affect his collection of information. For example, a student may know a friend from the same school as the new classmate, then he may have a previous impression from his friend. That means, his friend may be very nice, and he may think the new classmate is nice too.

Previous experience of people are likely to draw a wrong judgment. Because, first, the previous experience itself may be incorrect; second, people may not be aware of their previous experience, which may affect him unconsciously.

What’s more, people may make choices of collected information. This process may cause errors. First, people may judge the relationship between collected information based on previous experience. Second, people may not notice the errors in the information. The most important, people may tend to grab the most significant case information, ignoring the information from majority of people.

On the contrary, computer can avoid the problem of subjective, filter the entire data without previous experience, and draw the conclusion more accurate. So Computer Can Be More Suitable than People in Social Inference.

“Like” Is Really Like?

This week, Rosanna talked about the comment collection, analysis and some clustering models. As we all know, comments are very important to social network analysis, but how about “Like”? Does it has the same effectiveness?

When we analyzes comments, we can select the comments that have some effective information. But how can we select the effective “Like” since they all look the same. That is the most difficult problem.

A new study reported in “New York Times” shows that around these problem: the user’s “Like” of a article on Facebook can often lead to other users to follow a “Like”, even if the article actually written was not so good. However, “Dislike” will not lead other users to follow. That is, the “Like” is contagious.

Researchers with a website has cooperated to conduct an experiment. During the five months research, users can give ”Like” or “Dislike” to the articles. They divided the users into two groups, experimental group and control group. The experimental group members will give a “Like” or “Dislike” randomly by the system, when the article published. At the same time, the control group members will have no changes by system. As a result, the number of the “Like” of the experimental group was almost twice than the the control group.

Through this experiment, we can find out that “Like” is a kind of group behavior, and users may be affected by herd mentality, and then click “Like” or leave some active comments rather than depending on theirs own judgments or feelings.

Maybe that is why some websites will hide the score or “Like” numbers for some time.

The Importance of Contents in Social Media

With the popularity of the internet access devices and the development of computer technology, it has entered the web2.0 era, a time for the social media. Nowadays, almost everyone uses social media, and it has become a part of our daily lives.

After the course of the last few weeks, I have learned a more precise definition of social media, which refers to the internet and mobile technology based channels of communication in which people share content with each other.

More importantly, I have begun to pay more attention to the UGC(User Generated Content) because of the course. After combining reasons given by Rosanna and my own summary, I think there are following reasons that make the UGC important.

First of all, base on the social attributes of social medias, the relationship between users can directly affect the user activity and user stickiness. The UGC just can show the relationship between users truly.

Then, UGC may play a decisive role on the user's purchasing behavior. When people make a decision on whether buy something or not, they will refer to the advice of others, and the advice impact will be greater if the advice from someone they knew.

Let us see a research from Dr. Kelly McGuire, SAS worldwide hotel software executive director, and Dr. Breffni Noone, assistant professor of University of Pennsylvania.

They surveyed the consumers purchasing behavior on booking hotel among the different prices, rating and UGC. According the results(in graph 1), they drew the conclusion that hotel reviews(UGC) are the most important indicators of the quality and value assessment for consumers. Consumer buying behavior seriously affected by the user comments. Hotel User Reviews (rather than consumer rating) is the biggest factor affecting hotel quality, value and ultimately purchasing behavior. Even though poor rating hotels reduce the prices, they can not create any added value.