1. Internet recommendation from 0 - take 58 Zhuanzhuan as an example
1) Design the recommended product framework from 0
(1) Homepage recommendation : extract user portraits, and comprehensively recommend babies on the homepage according to the user’s age, gender, category preference, etc. extracted offline
(2) Recommendations on the baby details page : Buy it, buy it, watch it and watch related baby recommendations
(3) Nearby recommendation : The difference from the homepage recommendation is that the weight of the geographic location is increased. The geographic location should not only include the current geographic location, but also include common active areas, such as home, company, etc.
(4) Search recommendation : In addition to full keyword matching, you should consider recommendations such as synonyms, synonyms, error-prone words, pinyin, etc. At the product level, prompt "Are you looking for xxoo baby?"
(5) Recall recommendation : After the user exits the system, use the RFM model to push coupons or push messages for customer retention and recall
TIPS: What is the RFM model?
RFM model: According to the user's last purchase time Recency, the purchase frequency of the recent period, the purchase amount of the recent period of Monetary, a weighted score representing the user's willingness to trade.
2) Recommend strategy implementation starting from 0
a. User portrait
b. How to build a portrait
a). Read the list of applications installed by the user to build a portrait
b). User behavior log
c. Baby portrait
d. How to build a baby portrait: For 58 Zhuanzhuan, to make a baby portrait, it must be subdivided into categories, and the word frequency statistics can be combined with manual review.
e. Labeling and Personalized Recommendations
f. Classification prediction recommendation
g. Collaborative filtering recommendation
2.