Abstract
With the popularity of the Internet and mobile devices, users generate a large amount of short video content every day, including social media platforms and online video platforms. This large amount of data requires intelligent and personalised recommendations to meet user needs. Therefore, a short video recommendation algorithm based on attention mechanism and neural network is proposed. Firstly, the attention mechanism is introduced on the basis of recurrent neural networks to adaptively learn the output weights of second-order features, which solves the problem that invalid secondorder features may bring noise and adversely affect the model performance. Then, when calculating the attention, the second-order feature vectors and the target video vectors are equally sliced into a number of vectors, constituting a number of different hidden semantic spaces, so as to calculate the attention weights. The problem of weight smoothing between vector elements is improved by learning the relationship between feature vectors and target short video vectors at a fine-grained level. Experimental results show that the attention mechanism overcomes the output weighting problem of recurrent neural networks. Compared with the commonly used video recommendation algorithms, by reasonably setting the window size of the attention mechanism, the proposed algorithm exhibits higher recommendation performance under different feature vector size samples.
Original language | English |
---|---|
Pages (from-to) | 1003-1018 |
Number of pages | 16 |
Journal | Journal of Network Intelligence |
Volume | 9 |
Issue number | 2 |
Publication status | Published - 2024 |
Bibliographical note
Publisher Copyright:© 2024, J. Network Intell. All rights reserved.
Keywords
- adaptive
- attention mechanism
- output weights
- recurrent neural network
- video recommendation