Quantitative Modeling of Recommendation Systems Driven by Dynamic Preference Logic

Jianming Sun, Shuo Yang, Bing Zhang, Zheng Dequan, Jie Piao

Keywords: Artificial Intelligence, Preference Logic, Dynamic Drive

Abstract:

With the wide application of recommendation systems in ecommerce and media scenarios, user preferences have shown
obvious dynamics and contextual dependence. However, traditional collaborative filtering, matrix factorization and deep
models mostly only fit static vectors and are difficult to explain the mechanisms behind the changes in preferences. To
this end, this paper proposes a three-layer interpretable recommendation model DPDRS based on dynamic preference
logic, achieving a transparent reasoning process from behavioral feedback to preference evolution and then to recommendation generation. The model constructs preference states that can be updated over time at the presentation layer based
on long-term preferences, short-term preferences, and preference temperatures. Introduce the preference upgrade operator in the reasoning layer to strengthen or suppress the corresponding attributes through positive and negative feedback.
In the generation layer, design a three-channel fusion structure of behavioral decision-making, attribute logic, and collaboration, and align behavioral semantics and attribute semantics with consistency loss. To adapt to different scenarios, in this paper, corresponding attribute matrices were constructed on datasets such as MovieLens-1M, and systematic comparisons were made with methods such as MAUT and
CosRec. DPDRS achieved significant improvements in indicators such as Precision@10, Recall@10 and NDCG@10. In
addition, diversity assessment, interpretability analysis, and multiple ablation experiments further verified the effectiveness of the preference escalation operator, attribute channel, consistency loss, and time decay mechanism. Overall, DPDRS has achieved interpretable dynamic recommendation based on logical reasoning, providing theoretical and practical support for the construction of a new generation of recommendation systems with the ability to model preference evolution.

Submission Number: 55