recommendation systems: applying amazon's collaborative filtering methods to e-commerce
TRANSCRIPT
Amazon Recommendation Services 123Mua.vn Recommendation Services
Recommendation Services
Nguyen.Cao-Duc
Data Mining Team LeadE-Commerce & Services Dept.
VNG Corp.
Internal Research Document
September 19, 2012
Amazon Recommendation Services 123Mua.vn Recommendation Services
Outline
1 Amazon Recommendation ServicesBusiness ModelResearch ModelImplementation Model
2 123Mua.vn Recommendation ServicesBusiness ModelImplementation Model
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Amazon Website
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services (cont.)
Browsing Product Recommendations:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services (cont.)
Viewing Product Recommendations:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services (cont.)
Purchasing Product Recommendations:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services (cont.)
How to have such Recommendations:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services (cont.)
Other Recommendations:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Recommendation Problem
The main purpose of the recommendation system is torecommend personalized products to users of a merchant’sWeb site.
Two types of Recommendations:Content-based Filtering
Recommend items with similar content.Collaborative Filtering
Recommend items based on interests of a community ofusers.
Hybrid Content-based Collaborative FilteringCombination the two above approaches to overcome thedisadvantages of each approach.
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Content-based Recommendation
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Collaborative Filtering - Problem Description
Question: What should be the rating of Sam for Yellow?Approach: Use ratings of other users (user-based CF) orother items (item-based CF)
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
How Collaborative Filtering works?
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Similarity Computation
Vector Cosine-based Similarity:Formular:
wu,v =
∑i∈I ru,i rv ,i√∑
i∈I r2u,i
√∑i∈I r2
v ,i
wi,j =
∑u∈U ru,i ru,j√∑
u∈U r2u,i
√∑u∈U r2
u,j
where:I is the set of items that both user u and v have rated.U is the set of users who rate both item i and i .Drawbacks
Different users have their own rating scales.
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Similarity Computation (cont.)
Correlation-based Similarity:Formular:
wu,v =
∑i∈I(ru,i − r̄u)(rv ,i − r̄v )√∑
i∈I(ru,i − r̄u)2√∑
i∈I(rv ,i − r̄v )2
wi,j =
∑u∈U(ru,i − r̄i)(ru,j − r̄j)√∑
u∈U(ru,i − r̄i)2√∑
u∈U(ru,j − r̄j)2
where:I is the set of items that both user u and v have rated.U is the set of users who rate both item i and i .
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Collaborative Recommendation
Given a user u:User-based prediction:
Aggregate the ratings of other users:
Pu,i = r̄u +
∑v∈V (rv ,i − r̄v )wu,v∑
v∈V |wu,v |
where V is the set of all users have rated the item iItem-based prediction:
Simple weighted average:
Pu,i =
∑n∈N ru,nwi,n∑
n∈N |wi,n|
where N is the set of other rated items of user u
Amazon Recommendation Services 123Mua.vn Recommendation Services
Research Model
Collaborative Filtering - Drawbacks
User has to rate items to build profiles as well as item hasto be rated (cold-start problem: new user, new item, newsystem)Recommendations may not be diversed
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components
Recommendation Service Components takes Items ofKnown Interest of the given User and Similar Items Table tocreate Recommendation Items.
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Engine
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components (cont.)
Sources of Items of Known Interest with respect to a User:User shopping card activitiesUser purchasing activitiesUser favorite items profile (i.e WishList)
Popular items are items satisfied some pre-specified popularcriteria:
Number of item viewsTime on item viewNumber of item purchasings
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components (cont.)
Each cell in the Similar Items Table associates a commonalityindex (CI) to indicate the relatedness of that item with thepopular item. The relatedness of two items i , j could beexpressed via:
Two items have been purchased togetherTwo items have been rated similarly
or the value of wi,j =∑
u∈U (ru,i−r̄i )(ru,j−r̄j )√∑u∈U (ru,i−r̄i )2
√∑u∈U (ru,j−r̄j )2
or the similarity between two items using content-basedfilteringor . . . combinations of all above with some controlledparameters.
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components (cont.)
Similar lists are combined appropriately by a weightingscheme representing the relative importance of popularitems with respect to the items of known interests.Weighting scheme of similar item lists:
Rating of the user to the popular item.User purchased multiple copies of the popular itemTime user spend on the popular itemRecent purchasing items are weighted more than earlierpurchasing items
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Service Components (cont.)
The combined sorted list of similar items lists may need bemodified to remove certain items:
Items already purchased or rated by user or have beenviewed by user and its content has not changed.Items not in any product groups registered by user.
The combined sorted list of similar items lists may need bemodified to add certain items:
Items user has considered to purchase but did notpurchase.Items user has viewed but its content has changed afterthat.
The recommendation result may be transfered to the enduser by different types of transmission methods (view onsite, email, mobile message, chat message, etc.)
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
123Mua.vn Website
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Paid services
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Recommendation Services
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Pageview Traffic
Amazon Recommendation Services 123Mua.vn Recommendation Services
Business Model
Pageview Traffic (cont.)
We want to have:
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Engine
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Recommendation Engine (cont.)
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Items of Known Interest
Sources of Items of Known Interest with respect to a User:User category browsing or item viewing activitiesUser shopping card activities
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Popular Items
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Popular Items (cont.)
Popular items are items satisfied some pre-specified popularcriteria:
Number of page views on an item and/or category of theitem and/or shop of itemTime on an item and/or category of item and/or shop ofitemBounce Rate on that itemExit Rate on that item
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Popular Items (cont.)
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Similar Item List
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Similar Item List (cont.)
Sources of Similar Item List by targeting to certain itemsbased on:
Number of page views on an item and/or category of theitem and/or shop of itemTime on an item and/or category of item and/or shop ofitemBounce Rate on that itemExit Rate on that itemor items follow certain business objectives such as itemsis Up within a period of time.
Sources of Similar Item List with respect to a Popular item:Items of the same category and/or shop
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Similar Item List (cont.)
Each item in the Similar Item List associates a commonalityindex (CI) to indicate the relatedness of that item with thepopular item. The relatedness of two items i , j could beexpressed via:
The value of wi,j =∑
u∈U (ru,i−r̄i )(ru,j−r̄j )√∑u∈U (ru,i−r̄i )2
√∑u∈U (ru,j−r̄j )2
where:ru,i represents interest level of user u towards item ior the similarity between two items using content-basedfilteringor . . . combinations of all above with some controlledparameters.
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Weighting Scheme of Similar Item Lists
Similar item lists are combined appropriately by aweighting scheme representing the relative importance ofpopular items with respect to the items of known interests.Weighting scheme of similar item lists of popular items:
User views the popular item multiple timesTime user spent on the popular itemRecent viewings of popular items weighted more
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
Question
How to identify popular topics from multiplerelated/independent properties?How to measure the interest of a topic viewed by a user?
Amazon Recommendation Services 123Mua.vn Recommendation Services
Implementation Model
THANK YOU *-*