A method for recommending virtual items, a method for training a recommendation model, an apparatus and equipment
CN117224970BActive Publication Date: 2026-05-26TENCENT TECH SHANGHAI
View PDF 2 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TENCENT TECH SHANGHAI
- Filing Date
- 2022-06-06
- Publication Date
- 2026-05-26
AI Technical Summary
Technical Problem
Existing gradient boosting decision tree models cannot accurately capture changes in players' interests at different times when recommending items, causing the recommendation results to deviate from the players' true needs, affecting the game experience and commercial value.
Method used
An LSTM model and a multi-head self-attention mechanism are used to determine the first interest vector, and an attention mechanism is combined to determine the second interest vector. The two are then fused to generate a target interest vector for virtual item recommendation.
Benefits of technology
It improves the accuracy and efficiency of virtual item recommendations, enabling it to better adapt to changes in player interests and enhance the gaming experience and commercial value.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN117224970B_ABST
Abstract
This invention discloses a virtual item recommendation method, a recommendation model training method, an apparatus, and a device. The method includes the following steps: constructing a virtual item vector sequence and a target object vector; obtaining a first behavior sequence and a second behavior sequence of the target object; determining a first interest vector using an LSTM model and a multi-head self-attention mechanism based on the virtual item vector sequence and the first behavior sequence; determining a second interest vector using an attention mechanism based on the virtual item vector sequence, the target object vector, and the second behavior sequence; fusing the first interest vector and the second interest vector to obtain a target interest vector; and determining virtual item recommendation information based on the matching degree between the target interest vector and each virtual item vector in the virtual item vector sequence. This method can improve the accuracy of virtual item recommendations and further enhance the recommendation experience for target objects, ultimately increasing the commercial value of products. It can be widely applied in the field of computer technology.
Need to check novelty before this filing date? Find Prior Art