W6 L1 Flashcards
Recommendation Engines (RE) runs on
machine learning algorithms
pros RE and how (3)
RE enhance customer engagement, increase average order value, improve up/cross-selling opportunities by showing relevant productsthat are viewed or purchased together on the homepage and category pages.
Although there are multiple types, recommendation systems (filters) are usually categorized in two main kinds
(1) Collaborative <user> filtering models and (2) Content <item-product> based models.</item-product></user>
Recommendation Systems: 1 Content (Item) Based Filtering
example
“You bought this (X) you may also be interested in that (Y) – that is similar”
Content Based filter recommends
s a new product similar to purchased product
Content (Item) Based Filtering what is meant by preidentified
Similarities are pre-identified by the firm and/or analysts (comedy movie is considered as similar with another comedy movie)
Content (Item) Based Filtering disadvantage preidentified
What if she is a fan of the main actor – and mainly after his movies ?Do you see the problem with pre-identified similarity?
Meta Path is
A Meta Path is a sequence of relations between object types, which defines a new composite relation
why use meta paths
To improve the item (content) based recommendations and link with user based matching
3 steps meta path (or content based fileting)
(1) Pre-defined set of product characteristics: e.g. genre, writer, actors
(2) Calculate similarity between products > (3) Select the products that are most similar to the consumed products
pros content bassed filtering
Simple and intuitive. Useful even if we know nothing about the user (at cold start) and if we have little/no data on past purchases.
cons content bassed filtering
S: Not model based (is high product similarity a good recommendation?) Product characteristics hard to define – what is a similar product?
Collaborative (User Based) Filtering example
“People LIKE you also bought (saw) this….”
Collaborative (User Based) Filtering pros 2
: (1) Simple and intuitive (2) Takes into account user behavior
Collaborative (User Based) Filtering cons 3
(1) Not useful for users and products with few data (sparsity)
(2) Not useful for new users and products (cold start) (3) Computation time is large with thousands of users and products
Meta-Path based approaches combine
user feedback data with additional information, such as item or user attributes and
relationships. Therefore, they can emulate collaborative filtering, content-based filtering,
User-based framing
emphasizes the similarity between customers (e.g., “People who like this also like…”);
Item-based framing
g instead emphasizes similarities between products (e.g., “Similar to this item”).
Framing the same recommendation as USER-BASED (vs. item-based) can increase
recommendation click-through rates
Framing the same recommendation as USER-BASED (vs. item-based) can increase recommendation click-through rates. Espacially when: 3
(1) The customer is less experienced (2) attractiveness of the product is higher (3) and the user perceive the similarity with others as high
Recommenders: The Effect on Sales Diversity
Common Wisdom:
Recommender systems will help consumers discover new products by lowering their search costs
Recommenders: The Effect on Sales Diversity: a critical view
Collaborative filters (CFs)—will decrease sales diversity since they only reinforce the popularity of already well-known products
Use of collaborative filters (CFs) is associated with a decrease in sales diversity relative to no recommendations. when is this negative effeect stronger? and why
If the recommender is based on purchase data (rather than view data) this negative effect is stronger.
why: Similar users still end up exploring the same kinds of products, resulting in concentration bias at the aggregate level.
