Recommendation System Application research: Online Bookstore

In addition to sorting and active search, the recommendation system is also an important way for users to browse the goods, can help users find similar and interested products, increase the amount of traffic, the visitor into the buyer, guide the user to buy. The resulting value is to enhance the user's shopping experience and user viscosity, and increase the order amount, such as amazon30% orders from the recommendation system.

The information produced by Recommender Systems is more natural than precision marketing advertising, and it can also be used to sell goods. But if the referral system's information is too dense or inaccurate, it will be suspected of being spam. If Dangdang has logged in to view the Web information architecture: Design a large web site, sidebar and the bottom of the following several forms of recommendation system:

There are duplicate data in the recommendation system, such as 1, 2, and 6 recommending the essence of "about Face 3 interactive design", and individuals have purchased the book. Favorite songs can be repeatedly played, the books have been purchased can not be repeated recommendation, but the user's purchase record as the main recommendation basis. Repeated data can weaken the recommended effectiveness, and too much of the recommended form and data will dilute the recommended accuracy considerably.

The individual was once a loyal user, has been under the Dangdang 27 orders, purchased 90 books, the recommendation system for the purchase of help is almost negligible, and no personal purchase records to use. In the shopping cart page ready to place an order when the login Dangdang, the page will jump to the homepage, in the next order page Click Return will often appear orders lost.

Dangdang's delivery progress time on average 4-5 days, after placing orders can view the progress of delivery. According to the personal design experience, it is generally obvious that users feel slow or the operation of many steps to use the progress bar, a little relief of the user's sense of waiting, and excellent clearly inform the user of the day the goods will be delivered.

Because of the dislike when delivery is too slow, instead of being logged in, view the Web information architecture: Design a large web site book page, highlighting the "purchase of this product customers also bought", a total of 100 books recommended, about 50 are personally interested or have been bought in Dangdang. Duplication of data is well avoided because of the main use of a form of recommendation.

"Buyers of this product have also bought" than "Browse this product customers have also seen" more access to the user's trust, the use of Item-to-item collaborative filtering system, through the cosine of the calculation of the similarity of goods to recommend, compared to the traditional collaborative filtering (collaborative filtering) , solve the cold start problem, not overly rely on the user's browsing and purchase records, when the user first view the book can also recommend.

The mouse stays in Dangdang recommended books on the book cover pictures and prices, while excellence shows the score of books, evaluation number and price, in accordance with the process of customer purchase, book scoring more attractive users to browse and create the desire to buy.

Excellent mail with book recommendations, illustrations provide useful information, and will not be personal as spam, see interested data can be viewed or directly purchased, very convenient. When you invite comments in text form, you'll open it when you receive the first copy, and the other 30 content messages are deleted as spam messages.

Two sites can be more accurate ads, not according to the label of the book display ads, the purchase of computer-related books are mostly male, Dangdang display label for women's bra ads, excellent nets show high heels and women's underwear ads, may be able to distract the male's attention and clicks, but will not increase the advertising conversion rate, Displays the digital product the advertisement to be more suitable in this kind of book page.

Duplication and spam damage the user experience, the recommended information is accurate and useful, not too much to display.

Source: http://daichuanqing.com/index.php/archives/1882