An online commodity recommendation method, device and system based on an attention mechanism

What is AI technical title?
AI technical title is built by Patsnap AI team. It summarizes the technical point description of the patent document.
By using multi-round randomized attention masking and self-attention association extraction, random interaction behaviors are identified and masked. Non-random interaction behaviors are used to generate product recommendations under stability constraints, which solves the problem of interference from random interaction behaviors in online product recommendations and achieves the accuracy of recommendation results and uniformity of category distribution.

CN122388264APending Publication Date: 2026-07-14GUIZHOU INST OF TECH

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: GUIZHOU INST OF TECH
Filing Date: 2026-06-05
Publication Date: 2026-07-14

AI Technical Summary

Technical Problem

Existing online product recommendation technologies based on attention mechanisms struggle to effectively distinguish between accidental user interactions and genuine, stable preferences, resulting in recommendation results being affected by noise and failing to accurately reflect users' true product preferences.

Method used

By performing multi-round randomized attention masking on the interaction behavior sequence, a sequence of interfering behaviors is generated and self-attention association is extracted. Random interaction behaviors are identified and masked, and a behavior stability label sequence is generated. Attention bearing projection processing under stability constraints is performed using non-random interaction behaviors to generate a multi-layer product recommendation sequence with a category-dispersed structure.

Benefits of technology

It effectively eliminates noise from accidental interactions, ensuring that the recommendation results reflect users' true and stable preferences, and avoids the continuous accumulation of similar products during the stratification process, thereby improving the uniformity of category distribution in the recommendation sequence.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122388264A_ABST

Patent Text Reader

Abstract

The application provides an online commodity recommendation method, device and system based on an attention mechanism, obtains an interaction behavior sequence of a target user and a candidate commodity set; performs multi-round randomized attention mask processing on the interaction behavior sequence to generate a set of interference attention response distributions; performs self-attention association extraction on the interaction behavior sequence to obtain an original attention response distribution, and performs cross-round stability comparison on the set of interference attention response distributions to generate a behavior stability label sequence to filter out a non-accidental interaction behavior subsequence; performs attention acceptance projection processing on the candidate commodity set under the stability constraint to generate an attention acceptance degree indication of each candidate commodity; and performs hierarchical screening on the candidate commodities and simultaneously performs same-layer category dispersion arrangement to output a multi-layer commodity recommendation sequence with a category dispersion structure. The application can filter user accidental interaction noise, capture stable interest, and improve recommendation diversity and accuracy.

Need to check novelty before this filing date? Find Prior Art