BPR: Bayesian Personalized ordering learning model for implicit preference data

The Bayesian personalized ordering learning model of BPR oriented to implicit preference data research on the formal definition of individualized ordering in the introduction of BPR Bayesian personalized sequencing BPR optimization criteria AUC optimization analogy BPR Learning algorithm based on BPR learning model matrix decomposition Adaptive K nearest neighbor BPR and other methods Relational weighted regularization matrix decomposition WR-MF maximal edge matrix decomposition MMMF Experimental evaluation data Set evaluation method experimental results and discussion on non-individualized ordering conclusion

BPR: Bayesian Personalized ordering learning model for implicit preference data Summary Introduction Related Research Personalized Sorting

A personalized sort task is to give a user a sorted list of products. This task is also referred to as a project recommendation. For example, an online store wants to recommend a sorted list of personalized products that a user wants to buy. In this paper, we study the scenario of inferring the sort from the user's implicit feedback behavior, such as historical purchase records. The problem in implicit feedback system is that there is only forward preference. The non-observed user project pair-for example, a user has not yet purchased a product-is a mix of true negative feedback (the user is not interested in buying the product) and a missing value (which the user may later purchase). Formal Definition

The

Makes u u a collection for all users, and I I for all product collections. The implicit feedback data s⊆uxi S\subseteq U \times I is used in the recommended scenario in this article (as shown in Figure 1). Examples of such feedback data are the purchase data of the online store, the browsing data of the video portal and the click Data of the site. The task of the Recommender system is to give the user a personalized ordering of all products >u⊂i2 >_u\subset i^{2}, where >u >_u needs to meet the following properties:
∀i,j∈i:i≠j⇒i>uj∨j>ui ( Totality) \forall i,j\in i:i\neq j\rightarrow i >_uj\vee J >_ui (totality)
∀i,j∈i:i>uj∧j>ui⇒i=j (antisymm etry) \forall i,j\in i:i>_uj\wedge j>_ui\rightarrow i=j (antisymmetry)
∀i,j,k∈i:i>_uj∧j>_uk⇒i>uk ( transitivity) \forall i,j,k\in i:i>\_uj\wedge j>\_uk\rightarrow i>_uk (transitivity)
For the sake of convenience, we also define:
I+u : ={|i∈i: (u,i) ∈s} i_u^+:=\left \{|i\in I: (u,i) \in S \right \}
U+i:={|u∈u