Recommendation System With Dual Collaborative Filter Usage Matrix

摘要

Example apparatus and methods perform matrix factorization (MF) on a usage matrix to create a latent space that describes similarities between users and items and between items and items in the usage matrix. The usage matrix relates users to items according to a collaborative filtering approach. A cell in the usage matrix may store a value that describes whether a user has acquired an item and the strength with which the user likes an item that has been acquired. The latent item space may reflect true relationships between items represented in the usage matrix and those relationships may be proportional to the strength in the usage matrix. The strength of the relationship may be encoded using continuous data that measures, for example, the amount of time a video game has been played, the amount of time content has been viewed, or other continuous or cumulative engagement measurements.

主权项

1. An apparatus, comprising:
a processor; a memory to store recommendation data associated with a recommendation of an item to acquire; a set of logics that produce a latent space item model Q upon which the recommendation may be based; and an interface to connect the processor, the memory, and the set of logics; the set of logics comprising:
a first logic that:
accesses a collaborative filtering based user-item usage matrix M that stores a first type of data and stores a second different type of data, where the first type of data and the second type of data are provided from one or more collaborative filtering processes, where the first type of data stores values lmn that describe whether a user m likes an item n, and where the second type of data stores values Smn that describe the strength with which m likes n;a second logic that:
establishes a first probability that m likes n as Pr(lmn=true)=σ(umTvn), where the vector um is a representation of m in Q, where the vector vn is a representation of n in Q, and where σ(x) is the logistic sigmoid;establishes a second probability that the relationship between m and n has a strength smn as Pr(smn)=(smn;αumTvn,λ−1), where (x;μ,λ−1) is a Gaussian function with mean μ and variance λ−1, and where α is a scalar, andproduces a likelihood function L for M based on joint probabilities Pr (lmn) and Pr(smn) for two or more (m,n) pairs, where L encodes values for both the first type of data and the second type of data; anda third logic that:
learns one or more item vectors and a by performing matrix factorization on M using L, andproduces Q from the one or more item vectors, where Q preserves true relationships between items represented in M, and where relationships between items in Q are proportional to values in the second type of data.