How Personalized Recommender System Influence Users’ Attitude？

How Personalized Recommender System Influence Users’ Attitude？

Feng, Yuan (1155081930)

Nowadays, this is no longer the “geeks time” when people are sensitive to information and always require more message. What most users want to do now, is just migrate the real life into Internet, they show far more interest in SNS friend speaking than the latest technology trends. As the composition of Internet users changing, for Internet product, service object shall change accordingly. And that’s may be one reason why the personalized recommendation emerged and mattered.

From Affective to Conative and Cognitive

Customer’s attitude can be influenced by multiple factors, as what we’ve learned in last few classes, the mental position a person takes towards a topic, person, or an event that influences an individual’s feelings, perceptions, learning processes, and behaviors. There exist three steps or dimensions in this field, for personalized recommender system, this function basically follows the sequence from affective to conative and cognitive.

The strongest proof is what you may already have discovered that when you open a app, it used to contain various types of information, while now is full of one or a few specific types, usually entertainment gossip, or funny making jokes of socialization. Like the daily recommendation songs in your Netease cloud music APP (see Figure 1), and “something that you may interested” part in Taobao, usually listing something that you searched before or just bought the similar products. These are all the very achievement of personalized recommendation system.

Figure 1. Netease cloud music’s personal recommendation

Generally speaking, personalized recommendation based on user characteristics and preferences, through the collection, analysis and define its users’ behavior in history, understand what kind of person the user is, and what are their behavior preference, what they buy, share, produce, or interactive feedback and so on. Finally, it will understand and obtain user preferences that comply with the rules of the platform features, therefore recommend information of interest to users.

Besides, target customer segmentation is also important, as the view of demographic aspect, different age and work group of consumers may show different curiosity and acceptance level towards personalized recommendation, younger generations may feel less offend when they realize their behaviors have been “observing and calculated” their behaviors, and hence grow interest in one specific APP, while white collar may find recommended interesting news truly help relief the stress since all-day work really don’t need extra serious common information.

Five Key Elements of Personalized Recommendation

However, how exactly is personalized recommendation influence users’ attitudes, does it make things better or worse, this will be a complicated question to answer. Take Today’s Headline, which is a typical APP that using personalized recommendation system to attract users, as an example. (see Figure 2)

Figure 2. Today’s Headline’s personal recommendation

Basically, this personalized recommendation system should contain five key elements, which can be entry point of variables of user’s satisfaction degree:

1.     Producers, as in people who produce the original content, could be either user (UGC) or professionals (PGC), if production needs to be paid, then it is referred to the occupationally-generated Content (OGC).

2.     Content, without the production, in this case, content, there will no users visit to platform. Personalized recommendation require content as a fundamental basis.

3.     Consumption platform, a platform providing content for consumers to access, as in website or APPs like Netease cloud music.

4.     Consumer, meaning the users who enter the platform for content.

5.     Feedback, refers to the interactive behavior of the contents when consumers in consumption platform. In the recommended such as Today’s Headline news app, product or collect news and information, and display it in the client. Every time users click on one of those news and read the detailed content formed a feedback. Similarly, a click on a tag at the top of the navigation, add or delete a channel, collect, offline, share or repeat click an article, these behaviors all can be regarded as feedback.

Today’s Headline will use technical method according to the information, to set up the regular user interest model and the recent interest model after a period of time. Then the model will be used to trial and error, make adjustment again and again according to the behavior variance, prompting the model rising up and more and more closer to the real user preferences.

The Machine Will Not Do Evil

In the same time, practical experience reflects that information’s diversity will be sharply cutoff once started the personal recommendation, and the content will be intensive, even quality of which is more difficult to control, since the machines are not responsible for content quality, but only responsible for the potential audience and popularity of content.

Mechanized judgment inevitably lack of " personality " or "style", since the content of the different media has different style, which reflected special personality, previously content is mostly completed by artificial, so all the content is with a unique style.

However, recommendation algorithm is not person, it can just automatically according to existing data to recommend, which perhaps is the reason why most of content algorithms recommended looks similar.

     The system cannot determine the current popular events in a short time, tend to be popular had formed after the machine to discover. That is to say, screening machines have several dimensions, and there also exist invisible factors. No doubt that machine will not do evil, it just continuing make optimization around one goal that put forward by designer. But customers’ goals will sometime have deviation, leading machine to enlarge this deviation.

References

Yu Li, Liu Lu, Li Xue Feng. (2004). The Personalized Recommendation Algorithm under User Multiple Interest Study. Chinese science and technology periodical database, 117-125.

Lu Qian Yun. (2014). Mobile Terminal News APP Orientation Analysis Under the Media Ecological. Journal of northwest institute of adult education, 1-3.

Xiao Fei. (2013). Analysis of Audience’s Use and Satisfaction on News APP. The dissemination and copyright, 12-53.

Liu Jian Guo, Zhou Tao, Guo Qiang, Wang Bing Hong. (2009). Overview of the Evaluated Algorithms for the Personal Recommendation Systems. Complex systems and complexity science, 132-67.

Brief Talk on personalized recommendation. (n.d.). Retrieved October 21, 2016, from woshipm.com Web site: http://www.woshipm.com/pd/321344.html

RSS and Embarrassment of Personalized Recommendation. (n.d.). Retrieved October 21, 2016, from leiphone.com Web site: http://www.leiphone.com/news/201406/s-embarrassed.html