Session Based Recommendations Using Recurrent Neural Networks - Long Short-Term Memory

Dobrovolny, Michal; Selamat, Ali; Krejcar, Ondřej

doi:10.1007/978-3-030-73280-6_5

Cited by 6 publications

(1 citation statement)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Circulating neural networks have achieved great success in sequence modeling and are tried to be applied in session recommendations [12][13][14], And achieved better performance than the conventional methods. GRU is the recurrent neural network that is most used in session recommendations [15][16][17], The underlying recurrent neural network RNN was applied in earlier methods based on recurrent neural networks [18], While the newer recurrent-based neural methods use long, short-term memory networks [19][20][21]. The method based on recurrent neural network has a rough transformation relationship between projects.…”

Section: Recurrent Neural Networkmentioning

confidence: 99%

Research on session recommendation system based on graph neural network

Wang,

Yu,

Niu

2024

AETR

View full text Add to dashboard Cite

before the graph neural network session recommendation system in order to ensure that the model can not smooth, and ignore the project long distance dependence, and most of the recommended method ignore the project sequence relationship, usually the session in the last project as the user's final interest, and lead to the session understanding is not comprehensive. In view of the above problems, this paper proposes a graph neural network recommendation method with sequence network. The main idea of this method is to regard the session as a graph structure, in which the nodes represent the items that the user interacts with, and the edges represent the interaction between the user and these items, such as click, purchase, etc. Then, the graph neural network can learn the embedding vector of each behavior in the graph. Comparing these behavioral vectors with the corresponding item vectors shows the user's interest in different items in the current session, and a self-attention mechanism is introduced into the graph neural network to distinguish long-distance dependencies in the learning sessions without oversmoothing. The sequence network can then be used to introduce bidirectional gating recurrent units and soft attention mechanisms when the vector representation of the item is fused into the vector representation of the session. In addition, combining contrast learning techniques to improve the expression of long tail items and indirectly improve the effect of the session recommendation model. The experimental results show that the recommendation algorithm of adding sequence network can reduce the complexity and improve the performance.

show abstract