In this era of information explosion, recommendation systems play a key role in helping users to uncover content of interest among massive amounts of information. Pursuing a breadth of recall while maintaining accuracy is a core challenge for current recommendation systems. In this paper, we propose a new recommendation algorithm model, the interactive higher-order dual tower (IHDT), which improves current models by adding interactivity and higher-order feature learning between the dual tower neural networks. A heterogeneous graph is constructed containing different types of nodes, such as users, items, and attributes, extracting richer feature representations through meta-paths. To achieve feature interaction, an interactive learning mechanism is introduced to inject relevant features between the user and project towers. Additionally, this method utilizes graph convolutional networks for higher-order feature learning, pooling the node embeddings of the twin towers to obtain enhanced end-user and item representations. IHDT was evaluated on the MovieLens dataset and outperformed multiple baseline methods. Ablation experiments verified the contribution of interactive learning and high-order GCN components.

The information explosion is strongly impacting recommendation systems. The dual tower neural network model can quickly filter and score tens of millions of items in the recall phase and assemble them into a ranking recommendation. This model has attracted considerable interest from researchers and industry. However, the user and item towers operate independently in the dual tower model, so it cannot interact with information features, and its recommendation accuracy is less-than-optimal. The application of graph neural networks to recommendation systems has developed rapidly, enabling item recommendation through a high-order learning mechanism based on feature interaction between user information and item information. This paper proposes an interactive higher-order dual tower recommendation algorithm, the Interactive Higher-order Dual Tower (IHDT) model. Our model is based on an interactive learning mechanism, which combines the user's feature information into the item encoder and the item’s feature information into the user encoder simultaneously. The resulting higher-order interaction connectivity of graph neural networks is used to obtain a better representation of users and items. Finally, the model algorithm is compared with a benchmark algorithm on four public evaluation indicators on a public dataset. The experimental results show that the model performance is superior to other algorithms. The dataset file of this paper used is available at https://pan.baidu.com/s/1AEh-XNP_nJhxqrKuK0CgeA, with password: a2b4

Orientation-patterned gallium phosphide (OP-GaP) and orientation-patterned gallium antimonide (OP-GaSb) have demonstrated excellent potential for broadband second harmonic generation (SHG), with a wide range of fundamental wavelengths and temperature tunability, making them promising candidates for use in various applications. In this paper, an analytical demonstration of forward and backward broadband SHG is presented, delivering the fundamental wavelengths in 0.7–12.5 and 2.4–9.8 µm, respectively, through the utilization of OP-GaP and OP-GaSb with quasi-phase-matching (QPM). The temperature tuning ranges for OP-GaP and OP-GaSb are −195–176 and −192–126 °C, respectively. OP-GaP and OP-GaSb can obtain forward broadband QPM SHG when the conditions of QPM and group-velocity matching (GVM) are satisfied simultaneously. The bandwidths of OP-GaP and OP-GaSb with lengths of 10 mm are found to be 776 and 484 nm, respectively, in the forward GVM fundamental wavelength for forward broadband SHG at 25 °C. The bandwidths for backward broadband SHG in the forward GVM fundamental wavelength are also calculated, but the results showed that OP-GaP and OP-GaSb can only achieve forward broadband QPM SHG. Variations of the GVM fundamental wavelength have also been demonstrated as a function of temperature.