With about 500 million registered users, 60 million daily visitors, and 48 thousand items sold per minute, Alibaba’s Taobao, who just celebrated its tenth birthday, must have the largest online shopping database in China. If every user purchased one more item when visiting Taobao because of the recommendations generated by data-powered algorithms, the sales number would be huge.
At the 2013 Innovation Summit on Big Data today, Yang Tao, a data scientist at Alibaba, said their approaches did help Taobao retailers sell more items and users find more they’d like to purchase.
One patented approach is a recommender built on top of data sets of user groups. It divides Taobao users into thousands of groups based on user profile information and what they have bought; for instance, middle-aged rich ladies who are living in third-tier cities. In each group there are several users who are expert at certain category of goods or good at exploring less-known Taobao stores. Algorithms based on the above-mentioned settings would help surface goods to group members based on the purchase history of the expert users.
Other users would benefit that recommendations are categories they always pay attention to, or goods of higher quality or better designed they otherwise wouldn’t know about. For retailers, the direct benefit is it boosts sales. In a test with six groups of female users who like purchasing clothing, it turned out that both the average value per order and value of per visit increased by more than 20%, according to Mr. Yang.
Taobao’s data are also used to predict sales at Juhuasuan, Taobao’s group-buying platform. As there are limited placements on Juhuasuan and staff took much time reviewing and selecting goods, an algorithms-powered model now is in place to forecast sales within a certain period of time, in 5 minutes or in an hour. So far the accuracy is as high as 80%, Alibaba claims.
Taobao doesn’t require users to fill in with detailed personal information, but it came up with an identification system based on profile information users had submitted and purchase history. The system concludes that there are 20 million college student users on Taobao, so goods would be recommended to them accordingly. It also can tell whether the kids of a user are old enough to go to middle school or not.
Some offline stores such as supermarkets showed interest in cooperating with Alibaba on data mining. As most supermarket visitors are also Taobao & Tmall users, supermarkets hope to leverage their online purchase history in addition to the offline to push personalized e-coupons and the like.
Other areas Mr. Yang’s team is working on include improving the recommender that is powering Alimama, the cross-web contextual advertising network and how to use the social data from Sina Weibo in which Alibaba made strategic investment. Alibaba now offers data analysis services to retailers, including paid offerings. Mr. Yang said they’ll develop more data products for retailers.