machine learning - algorithm to predict user's being online time and channel -
i working on building engine website wherein want increase user engagement based on different activities, music rated or article read, or promotional campaign signed for.
for first step, item similarity(for music , articles) has been computed via collaborative filtering, , stored. now, next step, want able analyze these scenarios
- (a,t) : scenario gives me particular activity , asks me when user activity
- (t,a) : scenario gives me time slot , asks me activity makes sense recommend user @ time.
- (c, t) : scenario gives me channel (like sms, email, webpage ad, app ad) , asks channel user available on right now.
as above suggests, there number of modules time, channel, interests, location, etc can combined interchangeably, determine in kind of plane can user be, , based on tell whether specific activity suggestion make sense or not.
how can go accomplishing scalably?
here few things thought of, not sure if can effective
- building item profile - create item profile stores time slots activity, channels activity , on. however, scenario useful (a, t) cases, not (t, a) cases.
- treat 1 user different users- have item similarity table, can store single user history multiple users, user1 user1_17hrs, user1_18hrs, user1_phone, user1_tab. however, blow user data store 2^n combinations
- graph db - have hunch kind of heterogenous data can changed graph user(user properties), channel(time channel property) , item nodes of graph , edges can rating item, etc. can use graph db queries select time , channel user user1 , item item1, in graph db syntax. not sure how model graph db this.
any leads, suggestions approach helpful, in advance!
Comments
Post a Comment