Personality-Augmented Matrix Factorization

• Personality-augmented matrix factorization is a collaborative filtering approach that integrates explicit personality measures and item attributes into the rating prediction process.
• It leverages kernel methods and a low-rank representation to generalize standard matrix factorization and effectively address cold-start scenarios.
• Empirical evaluations show a 2–6% RMSE reduction on benchmarks like MovieLens, demonstrating its practical advantage over traditional methods.

Personality-augmented matrix factorization is a framework for collaborative filtering (CF) that enriches classical low-rank matrix completion by directly incorporating user and item attributes—most notably explicit personality measures such as OCEAN scores—into the modeling process. By leveraging kernel methods and low-rank constraints, this approach generalizes standard matrix factorization, enabling the prediction function to depend not only on latent user–item parameters, but also on side information represented as real-valued vectors. This methodology addresses several practical and theoretical limits of traditional CF, including cold-start scenarios and the integration of heterogeneous auxiliary data [0611124].

1. Formal Problem Specification

Given a set of users $U = \{1, \ldots, m\}$ with observed ratings $r_{ui}$ on items $I = \{1, \ldots, n\}$ for pairs $(u, i) \in \Omega \subseteq U \times I$, each user $u$ has an associated attribute vector $x_u \in \mathbb{R}^{d_U}$ (e.g., OCEAN personality dimensions), and each item $i$ has $z_i \in \mathbb{R}^{d_I}$ (e.g., genres, keywords). The learning problem is to fit a function $f: X \times Z \to \mathbb{R}$, $f \in \mathcal{H}$ that predicts $r_{ui}$0 from $r_{ui}$1.

The regularized least-squares objective is

$r_{ui}$2

where $r_{ui}$3 is a reproducing-kernel Hilbert space (RKHS) over $r_{ui}$4 constructed as a tensor product $r_{ui}$5 with associated user and item kernels.

2. Kernel Construction and Representer Expansion

The user kernel $r_{ui}$6 and item kernel $r_{ui}$7 capture pairwise similarity between users and items via their attributes. The joint kernel over $r_{ui}$8 is given by

$r_{ui}$9

By the Kimeldorf–Wahba representer theorem, the minimizer has the finite expansion

$I = \{1, \ldots, n\}$0

Setting $I = \{1, \ldots, n\}$1 by $I = \{1, \ldots, n\}$2 (zero elsewhere), the fitted ratings matrix $I = \{1, \ldots, n\}$3 decomposes as

$I = \{1, \ldots, n\}$4

where $I = \{1, \ldots, n\}$5 and $I = \{1, \ldots, n\}$6.

3. Low-Rank Augmentation and Matrix Factorization

To enforce low-rank structure, $I = \{1, \ldots, n\}$7 is factorized as $I = \{1, \ldots, n\}$8, with $I = \{1, \ldots, n\}$9, $(u, i) \in \Omega \subseteq U \times I$0. This yields

$(u, i) \in \Omega \subseteq U \times I$1

where $(u, i) \in \Omega \subseteq U \times I$2 and $(u, i) \in \Omega \subseteq U \times I$3. The predicted rating for $(u, i) \in \Omega \subseteq U \times I$4 is $(u, i) \in \Omega \subseteq U \times I$5. This factorization recovers classical MF in the absence of side-information, while allowing smooth generalization based on user and item attributes.

Alternatively, using explicit feature maps $(u, i) \in \Omega \subseteq U \times I$6, $(u, i) \in \Omega \subseteq U \times I$7, where $(u, i) \in \Omega \subseteq U \times I$8 and similarly for $(u, i) \in \Omega \subseteq U \times I$9, the bilinear form

$u$0

with $u$1, $u$2, admits a low-rank parameterization $u$3, with $u$4, $u$5 (where $u$6).

4. Optimization Algorithms and Regularization

The learning objective for the personality-augmented MF in direct feature-mapping form is

$u$7

Equivalently, using $u$8, $u$9,

$x_u \in \mathbb{R}^{d_U}$0

For $x_u \in \mathbb{R}^{d_U}$1 (linear kernel), penalization of $x_u \in \mathbb{R}^{d_U}$2 encourages $x_u \in \mathbb{R}^{d_U}$3 to remain close to the feature subspace spanned by $x_u \in \mathbb{R}^{d_U}$4.

Optimization is typically performed via alternating-least-squares (ALS): (a) with $x_u \in \mathbb{R}^{d_U}$5 fixed, $x_u \in \mathbb{R}^{d_U}$6 is updated as $x_u \in \mathbb{R}^{d_U}$7 independent ridge regressions of size $x_u \in \mathbb{R}^{d_U}$8; (b) with $x_u \in \mathbb{R}^{d_U}$9 fixed, update $i$0 in analogous fashion. Per-iteration computational complexity is $i$1, with convergence usually achieved in 10–20 ALS sweeps. Stochastic gradient descent (SGD) is also applicable for direct minimization of the objective.

5. Kernel Choices and Feature Construction

The flexibility of the kernel choices $i$2, $i$3 allows tailoring the model to the domain-specific structure of the attributes:

• On personality (user) side $i$4:
  • Linear: $i$5, modeling linear effects of personality similarity on preference.
  • Gaussian RBF: $i$6, capturing nonlinear relationships between personality vectors.
  • Polynomial: $i$7, enabling broader nonlinear interaction patterns.
• On item side $i$8:
  • For genre or binary attribute vectors: linear or intersection kernels.
  • For features such as tags or embedding representations: RBF or histogram kernels.

The selection of kernels governs how closely the learned representations respect known user and item attributes, and the regularization parameter $i$9 controls strength of this alignment.

6. Empirical Performance and Interpretive Insights

Experiments on benchmarks such as MovieLens and BookCrossing demonstrate that side-information via the tensor-product RKHS and low-rank augmentation reduces RMSE by 2–6% compared to vanilla low-rank MF, when measured in conventional rating prediction settings. With explicit OCEAN personality feature encoding for $z_i \in \mathbb{R}^{d_I}$0, further consistent gains are observed, particularly for cold-start users. The RKHS construction permits adjustable coupling between the latent space and the measured traits through the choice of $z_i \in \mathbb{R}^{d_I}$1 and $z_i \in \mathbb{R}^{d_I}$2, allowing for empirical evaluation of how much the attributes contribute to prediction accuracy.

7. Implementation Steps

A standard procedural workflow is as follows:

1. Gather data in the form $z_i \in \mathbb{R}^{d_I}$3.
2. Specify kernels or feature maps $z_i \in \mathbb{R}^{d_I}$4 for user and item attribute vectors.
3. Initialize parameters $z_i \in \mathbb{R}^{d_I}$5 (or their equivalents) with small random values.
4. Optimize the low-rank objective using ALS or SGD.
5. Predict ratings for new (user, item) pairs via $z_i \in \mathbb{R}^{d_I}$6 [0611124].

This framework systematizes the integration of explicit personality and other side-attributes into matrix factorization, with all key operations and results justified within the structure of kernel-based low-rank learning.