Session-based Recommendation Algorithm Based on Two-tier Network Mining User Preferences
The next recommendation is one of the challenging tasks in the field of e-commerce.In the existing serialization recommendation model,the importance of contextual interactive items to the next item to be recommended is the same,and the pro-portion of users'long-term and short-term preferences in the recommendation process is the same.An improved algorithm of users'long-term and short-term preferences based on attention network is proposed.The proposed algorithm uses a two-layer attention net-work.In the first layer of attention network structure,the neural memory network with attention mechanism combined with user his-torical session embedding is used to calculate the attention weight of items in the current sequence.In the second layer of attention network,the weighted context information is combined with the current hidden layer output of the gated loop unit to form a neural network memory unit,and then the attention mechanism is used to calculate the weight of the current memory unit,that is,the weight of short-term preference.Then,combined with the user's historical shopping information and the embedding of the current context,the weight of the long-term preference of the current context is calculated through the attention mechanism.Through the test on the open e-commerce data set JDATA,the recommendation accuracy of the proposed method has been greatly improved.
万方数据
维普
