How to design the relevant product recommendation

Why do related product recommendation?

The details of the product are the islands that may have dug gold, as we all know.

So we make a variety of moves, finally let users come to the Product Details page. We quietly read the Devil's mantra, so that users immediately to point to the most eye-catching "add to the cart" or "buy immediately." However, most of the consumer's UV conversion rate is not more than 5% (not to mention the pv!), the vast majority of users will eventually not buy this product, it is possible that he was the big chest model map to cheat in, there may be inappropriate prices, there may be goods details do not like, It is possible that most of the praise has a bad comment that he can't accept, in short, he does not want to buy.

So the user is lost?

The role of the related product recommendation is to let the user continue to wander until he finds a favorite commodity.

Good product recommendation, is to let users can not stop the pace.

The key to related product recommendation is "related"

The key to the sales of related goods is "related", which means that the goods must be cut from an angle, or dimension, and then clustered and recommended to the user. This is similar to the offline display of goods, such as you go to a beef noodle shelf before, pick up a bag of noodles to carefully look up, perhaps this taste does not like, then you may find other flavors from the side of the shelf; maybe the word "Kang Shuai Fu" was finally found wrong, you can try to find it on the side shelf. Master Kang. " The former is based on taste and the latter is based on branding.

There are a lot of clues, such as specials, such as suits.

The display on the line will be richer, because the clues are configurable and can be sliced, unlike the offline shelves that are difficult to move.

commodity-based and user based behavior

Throughout the current major electric network recommendations, nothing but "based on goods" and "based on user behavior" two related products recommended.

Based on category, there are two main ways of "correlation" and "sales rankings." Relevant collocation, often based on complementary goods and categories, such as selling a mobile phone, a mobile phone shell, charger; sell a shirt, put a pair of trousers and socks on. Set Meal to buy 10 yuan Oh, pro. "Sales List", this must add other labels for refinement, such as "Same category", "with the brand", "with the price segment", this is the East Jingdong business details.

Based on user behavior, it is through the user's individual or group performance characteristics to recommend. In this way, Amazon is used to the fullest. Like "Guess what you Like", based on the personal characteristics of the user, such as age, sex, shopping preferences, income levels, and so on, this is not rich in data, ordinary consumers do not play at all. But there are some simpler ways. The simplest is the "recently browsed Merchandise" module, wake-up user memory, simple, good to see data. There is also "browse the product users also browse", "browsing the product of the final purchase", which is based on group browsing behavior; Buyers of the product also purchased ", which is based on group buying behavior." Pure so play, is not to play, the recommended goods may not be reliable. Whether it is browsing, buying behavior or pull the relevant category, brand and other label information to aggregate.

As for the specific algorithm, don't ask me, I do not know.

There are other recommendations everywhere, annoying ah there are wood?

All of these related recommendation Modules Plus, is really full screen merchandise, seemingly rich, you can not forget the business Detail Page's primary goal: let users buy goods. Choose too many, annoying, jump between the pages.

Therefore, do not overly recommend.

Differentiate the recommended commodity type: Similar goods, supplementary goods and friendly goods

A shirt on the Product Details page, you recommend a shirt, that is the same kind of goods; recommended a belt, it is a supplementary product; you calculate, bought a shirt users usually also bought TT, OK, this is a friendly product.

In general, the list of similar products, "Browse the product users also browse", "Browse the product of the final purchase", the recommendation is often similar products. "Relevant collocation", "purchase of this product also bought the user", recommended a supplementary product; Guess you like "the kind of recommendation is" friendly goods ".

Generally speaking, the content of the detailed page should include similar goods, supplementary goods and friendly goods, do not put all the thought of the modules are paved. So how do you use the appropriate modules? Consider the following factors.

Distinguishing between category requirements: Demand concentration and demand decentralization

Product life cycle long, new product update slow products, often the purchase demand is more concentrated, this time the relationship between commodity varieties is more stable, based on the recommendation of the category will be more reliable, at this time, like "related", "Sales List" from each dimension (category, brand, price) to split, the probability of matching users is relatively low.

And the demand for highly dispersed goods such as women's clothing, recommendations such as sales rankings are often unreliable, and the reason for using a product based on user behavior may be a bit more plausible, because the person who buys such a product is the same person and has a similar style, so the recommendation based on user browsing and buying behavior can actually be hit again. "Style" of the product attribute tag, this label can not be seen to the user. There's actually another place many people pay attention to, but so far did not see which home in the recommendation of the relevant products, is to bask in a single area, such as where there are guests who are pleased with the sun, but it is not very obvious in the sun to display the relevant products of the person. If it is a highly dispersed commodity, a product recommendation based on human factors is worth trying.

Differentiate user type: old user and new user

The new user's recommendation, the above play also enough game.

Old users of the relevant recommended play can be more rich, can have personalized product recommendations, if it is a platform for the site, you can recommend "the shop you have purchased similar goods." Of course, there is no basic ability, these still play not turn.

Product Recommended Location

Most of the site, is to put the supplementary products under the main map, while the same kind of goods, friendly products recommended in the sidebar and the bottom bar. The first goal, no doubt, is to allow users to buy, the second goal, bought a stop to buy some other things, buy more; the third goal, if this is not your goods, then look at the sidebar of other goods.

In recent time, in the South Korean website, Gmarket, 11st Product Details page design, but also a large number of added to the store link, the biggest difference is in the right column to add a floating bar, display related products, no matter how you drag the Merchant detail page, you can see this floating bar. The design of course is encouraging the business page of the jump, but will not be too repetitive it?

This requires data to illustrate the problem.

Ultimately, it depends on the data.

The above said some ideas, but right or wrong, suitable or not suitable, ultimately to look at the data. What data are you looking at?

Simply from the business jump to see the words, to see the business detailed PV in the first page is the proportion of business detail, the relevant recommended modules of the click Rate.

In addition, other data are also worth reference, business detail pv/Whole station PV, the business jump rate, but the two data by other factors interference is relatively large.

Island connected

Related recommendations are more of a basic ability, often short-sighted sites do not see the importance of it, the relevant recommendations have been particularly rough, it is difficult to do "related." Related recommendations are also more difficult, I just outlined the idea, in the practical application is to continue to be based on the data to optimize, and more complex algorithms need to continue to improve the iterative.

But the goods and commodities really need to be linked by some kind of clues, and this kind of clue whether through the commodity marking, worker configuration or algorithm matching, should establish a mechanism to let these full of treasures of the island connected, so as to more prosperous.