Personalized search is a biased phrase that looks beautiful but is very difficult to practice. Here you can see the beauty of smallness, big data, the collision between the public and the long tail, and the power that technology gives to business.
Text/Tianxia.com reporter Fan Tingting
"Personalized factors have been added to Tmall search results. Different buyer IDs or different cookies may result in different search results due to different previous purchasing or browsing behaviors. Personalized algorithms are used in The goal of making full use of search results and marketing resources is to allow consumers to see the products most relevant to them, and ultimately achieve a Tmall with thousands of people,” Tmall CEO Xiaoyaozi posted on August 2. A post on Weibo in the afternoon explained the reason why Taobao users “showed different results when searching for the same keywords” in recent days.
This Weibo is like a butterfly effect, and its influence can radiate to the entire Taobao sellers, large and small. In this regard, a reporter from "Tianxia Online Business" interviewed Gui Jiaoqi, the person in charge of Yitao Search, and made a detailed analysis of personalized search.
Currently, personalized search is still in the testing phase. The platform randomly selects about 5% of the traffic as test samples to further optimize Taobao search through various indicators such as click-through rate and conversion rate after personalized search. It is expected that preliminary statistical results will be available within three to four months.
What is personalized search
Personalized search, as the name suggests, is to incorporate personalized elements into the search process.
Guijiaoqi explains the principle of personalized search from two aspects: On the one hand, from the user's perspective, everyone has different preferences, which is reflected in online shopping behavior. This preference can be reflected in many aspects, including price preference, category preference, regional preference, usage preference, quality preference, etc. In particular, different purchasing abilities will lead to users’ preferences for different unit prices or brands. It just so happens that Taobao has the ability to dig into these preferences to a certain extent and then further segment users according to different dimensions.
On the other hand, from a search perspective, the platform can do two things: firstly, explore the needs of users, and then clarify their consumption directions, including a classification derived from a mixture of various consumer preference information; secondly, It is the classification of goods, such as whether a product is high-end or low-end, what age group the brand is positioned for, etc. The platform can classify goods according to various dimensions through data collection. Once these two things are done, when customers enter Taobao to shop, the platform will be able to connect them with the products. If the consumer is what we commonly call a "diaosi type" customer who usually likes to buy discounted products and tends to have low prices, then the search results page presented to him will be biased towards low-priced products. If it is observed that a consumer is over 30 years old, has relatively strong spending power, and focuses on the pursuit of brands, then the platform will try to place products that match his consumption habits as high as possible in the relevant pages he searches for.
We can use an example to connect the relationship between the two: if the system detects that a certain consumer is more concerned about new products, Taobao will accordingly present him with more popular products. So how should new products be defined? Products can generally be divided into two categories, standard and non-standard. Standard categories will not be updated very quickly, such as 3C products, which basically have a clear time to market, and it is easier to define new products; for non-standard categories, only brands will have new products, such as some clothing brands. Therefore, it is relatively difficult to capture new products of non-standard products. It is difficult to achieve absolute accuracy from both technical means and sellers' perspective, but this will be a process of gradual improvement.
At present, Yitao's mining efficiency can reach about 80%, which means that more than 80% of the new products considered by the platform are indeed new products. Through this targeted display of search results, the conversion rate will be higher than normal.
In this example, it actually reflects two aspects of personalized search, namely the division of users and the division of products, as well as the corresponding relationship between the two.
To achieve this effective correlation, on the one hand, it is necessary to cooperate with the category, and on the other hand, it is also necessary for merchants to have more accurate descriptions of the products when they are put on the shelves. The searchers themselves will also do in-depth digging.
For sellers, personalized search helps increase the value of traffic. Originally, after consumers entered the Taobao page, the result display page for searching for the same keywords was the same. In this case, on the one hand, consumers cannot find the products they really need quickly, and on the other hand, the display of merchants is also a waste of traffic.
And if you switch to personalized search, it will form a display of search results for thousands of people. Then customers can see more products that match their expectations, and the store will have more display opportunities, which means it will improve The value of traffic.
For stores with a relatively clear store style, they may be able to attract higher-quality target customers without spending a lot of money on promotion, which saves time and cost for both merchants and buyers.
Therefore, the most important purpose of personalized search is to provide customers with similar products based on the past purchasing preferences of different groups of people, and to use more precise positioning and more relevant search results to increase the click-through rate of products and the store's reputation. Conversion rate.
Difficulties in Personalized Search
But Guijiaoqi also admitted that although personalized search sounds beautiful, there are many problems in actual operation.
First of all, you will encounter many difficulties when distinguishing customers. Identifying user intent can be a very tangled process. Not to mention the differences among users based on various dimensions such as age, gender, region, income, etc., there are still many situations where two or more people share the same account on Taobao, such as male and female friends, such as parents and children. This situation causes their purchasing behavior to sometimes appear "contradictory" to the system. If the system recognizes that you are a female, but you buy a male jacket, the system will be "dumbfounded". What personalized search can do is that when you search for "jeans" by default as a woman, a large number of products displayed in the results are women's jeans. Unless you type "jeans for men", then the system will know that you are for "jeans for men". Men buy products and it will display the products according to your wishes.
Therefore, for the platform, there are many factors that need to be considered. If the search results are too rigid, it will cause user disgust. Because not all buyers have clear user behaviors, most purchasing behaviors may be relatively random. If 20% to 30% of user intentions can be identified, in Guijiaoqi's view, it can be regarded as the achievement of a preliminary goal.
Secondly, classifying goods is also a huge project. How to define whether a product is cheap and good-looking or high-end? Is a piece of clothing suitable for office hours or weekends? When positioning a product, you must also consider its usage scenarios. At present, positioning in these different dimensions will be completed through cooperation between merchants and platforms.
Therefore, personalized search is an interactive process. In the process of incorporating personalized elements, Taobao also hopes that more buyers will participate. If there are highly loyal customers who are willing to describe their consumption habits and are willing to help the platform optimize search rules, that is the best way. The more engaged the user is, the better the degree of personalization will be achieved.
Gui Jiaoqi gave an analogy to the reporter of "Tianxia Online Business". When everyone plays Weibo, the pages they see are different. What is displayed on your page is your customized content, and the information you see is all Is what you want to see. The same goes for personalized search. It is hoped that through the linkage between buyers and sellers and the platform, we can ultimately achieve an efficient information acquisition result, so that consumers can have a better shopping experience.
In the current testing phase, merchants may not feel that much. Gui Jiaoqi believes that at least until the end of this year, 60% of the traffic will not change significantly. For the remaining traffic, stores can fully demonstrate their personalization. However, this personalization factor still needs to consider the quality of the product itself and the service, and better combine current resources with customer preferences.
Personalized search can only really work when customers have clear consumption behavior. At this time, the clearer the seller’s product positioning, the better the effect will be.
Although it is difficult for merchants to increase their traffic based on the specific rules of personalized search, stores can attract consumers with higher purchase intentions based on their positioning and product characteristics. For example, a user with an obvious consumption tendency always chooses clothing from several stores. The styles and prices of these clothing are relatively close. Then when he searches again, the platform will show him products of the same style. Therefore, personalized search will be a good channel to expand traffic for stores with clear styles.
At present, the number of online shoppers in my country has exceeded 200 million. Regarding the changes in search rules, Gui Jiaoqi said that he will take a cautious attitude. However, the trend of personalized search will not change. It is necessary to achieve the results desired by the platform without affecting the user experience. For the entire search team, this is not just a project this year, but also a project that requires long-term persistence. The goal of continuing to strive for.
Weaken the impact of positive rating on search
In addition to personalized search, the unbundling of praise rates has also been a hot topic on Taobao forums recently. There is also an inextricable relationship between praise rate and search.
From a seller's perspective, a simple summary of past praise rates has three impacts: First, when buyers browse and shop, the praise rate, as an intuitive data, will largely affect the store's conversion rate; second, If a seller wants to participate in some of Taobao's promotional activities, such as gold coins, through trains, Juhuasuan, etc., the positive rating is an essential indicator, which has a relatively large impact on the seller; the third is search. In the search ranking, the praise rate is an important reference indicator, which has a certain impact on the presentation page of buyers' keyword searches.
But as Gui Jiaoqi said, rules also need to keep pace with the times. If the existing rules no longer conform to the current environment and become a "profitable" secret box, and nourish some black industries, So if the entire Taobao ecosystem is to develop healthily, it is inevitable to upgrade these flawed rules.
The reason why the positive review rate is used this time is precisely because a large number of professional negative reviewers have seriously damaged the interests of small and medium-sized sellers. What Taobao is doing is, on the one hand, to protect small and medium-sized sellers, and on the other hand, it is also to clean up the entire Taobao environment.
Any rule is to make the market more perfect, and this is a valuable rule. If this rule can become an accomplice of extortion, it will be eliminated. And if a large number of people are not subject to the rules, then the rules must be biased.
In fact, as early as June, in Taobao baby search and store search, the indicator "good review rate" no longer affects the search results. Starting from June 27, the positive rating will no longer be used as an indicator for gold medal sellers to enter.
Gui Jiaoqi told "Tianxia Online Business" that the biggest purpose of doing so is to make merchants pay more attention to the transaction process itself, rather than just to obtain an evaluation result.
Gui Jiaoqi uses a Buddhist motto to describe this change: "Work hard for the cause, and let the fate follow the results." That is to say, as long as the merchants do what they can, they should not insist on the results. When we do something, it is very important to think about it. The same goes for Taobao. Sellers have to serve buyers, but if they encounter consumers who are deliberately making things difficult and cause the merchant's service to deviate from the normal track, there is no need to compromise in order to get a good review.
Of course, the unbundling of sellers and positive reviews does not mean that the platform no longer considers positive reviews, but that the rules must be improved to better serve buyers and sellers. Perfect means that the platform will consider it comprehensively. If a merchant sells hundreds of products and most of them have negative reviews, then there may be problems with the product or service. However, if the seller has made a total of ten sales and received relatively concentrated negative reviews, the platform will consider whether the negative reviews are malicious. At present, Taobao has also opened a green channel, and merchants can report suspicious transactions when they encounter them.
In addition, there will be other functions on the search page, such as good store recommendations, which mainly recommend some stores that have more repeat customers, a higher repeat purchase rate, and are particularly popular among yellow diamond buyers. Buyers' preferences also have a great impact on search results. For example, some consumers like to "sort by sales volume."
While weakening the praise rate, the platform hopes that merchants will pay more attention to their own upgrades. Since the praise rate is no longer a reference value for search ranking, what is an important reference for ranking? The platform's answer is that it will pay more attention to the overall effect. , such as the product's refund rate, delivery speed, description of the details page, and Wangwang's response speed. When sorting, the platform will take all comprehensive factors into consideration and recommend these products to users. At the same time, we also hope that merchants can have better ways to retain customers when going to the details page.
Another issue is relevance. For example, if a user searches for "white casual pants for men", if the merchant's pants are indeed white, but are not marked in the attributes, then the merchant's products may not appear in this search. The search results are in. Therefore, merchants should try their best to optimize keywords, including category name, brand name, attributes, style and other aspects. This is keyword optimization.
Conclusion: Generally speaking, personalization will definitely have a certain impact on the presentation of search results pages, but it does not necessarily mean that you see a page that is different from others, it must be affected by personalization factors. There are many factors involved in the ranking of the entire search, and search optimization is also in progress. But what remains the same is that merchants must first do their own thing, clarify their positioning, and deepen and refine their market. Then no matter how the search rules change, they will be the beneficiaries.