Attention mechanisms are now widely used in image captioning models. However, most attention models only focus on visual features. When generating syntax related words, little visual information is needed. In this case, these attention models could mislead the word generation. In this paper, we propose Task-Adaptive Attention module for image captioning, which can alleviate this misleading problem and learn implicit non-visual clues which can be helpful for the generation of non-visual words. We further introduce a diversity regularization to enhance the expression ability of the Task-Adaptive Attention module. Extensive experiments on the MSCOCO captioning dataset demonstrate that by plugging our Task-Adaptive Attention module into a vanilla Transformer-based image captioning model, performance improvement can be achieved.

The embedded representation and clustering tasks both play important roles in relational data analysis and mining. Traditional methods mainly employ graph structure to describe relational data, but intuitive pairwise connections among nodes are insufficient to model high-order data in the real-world, such as the relations between proteins and polypeptide chains. Hypergraphs are a generalization of graphs, and hypergraphs can well model high-order data. When modeling relational data in the real world, hypergraphs are often accompanied by node attributes, i.e. attributed hypergraphs. Besides this, how to integrate the structural information and attribute information appropriately is another important task, while has not been investigated systematically. In this paper, we propose Adaptive Hypergraph Auto-Encoder(AHGAE) to learn node embeddings in low-dimensional space. Our method can utilize the high-order relation to generate embedding for clustering. It is composed of two procedures, i.e. the adaptive hypergraph Laplacian smoothing filter and the relational reconstruction auto-encoder. It has the advantage of integrating more complex data relations compared with graph-based methods, which leads to better modeling and clustering performance. The proposed method has been evaluated on hypergraph datasets and benchmark graph datasets. Experimental results and comparison with the state-of-the-art methods have demonstrated the effectiveness of our proposed method.

Increasing evidence suggests that dysregulation of microRNAs (miRNAs) may lead to a variety of diseases. Therefore, identifying disease-related miRNAs is a crucial problem. Currently, many computational approaches have been proposed to predict binary miRNA-disease associations. In this study, in order to predict underlying miRNA-disease association types, a semi-supervised model called the network-based label propagation algorithm is proposed to infer multiple types of miRNA-disease associations (NLPMMDA) by mutual information derived from the heterogeneous network. The NLPMMDA method integrates disease semantic similarity, miRNA functional similarity, and Gaussian interaction profile kernel similarity information of miRNAs and diseases to construct a heterogeneous network. NLPMMDA is a semi-supervised model which does not require verified negative samples. Leave-one-out cross validation (LOOCV) was implemented for four known types of miRNA-disease associations and demonstrated the reliable performance of our method. Moreover, case studies of lung cancer and breast cancer confirmed effective performance of NLPMMDA to predict novel miRNA-disease associations and their association types.

Recently, deep neural networks are widely used in recommendation systems, but most of them are used to process auxiliary information of recommendation systems such as items’ descriptions and images. When it comes to how to learn a better interaction function to model the relation between user latent features and item latent features, which is the most critical step in a recommendation task, most works employ matrix factorization together with the inner product. However, it is sub-optimal because of ignoring many correlations between latent factors. As deep neural networks perform well in building more complex non-linear models, we employ deep neural networks to improve the collaborative filtering algorithm, solving the problem of implicit feedback which is the most common scene in real applications. Some recent work has contributed to finding better interaction function, but these functions are not exact enough to model comprehensive correlations among latent features. In this work, we propose the Convolutional Neural Networks based Deep Collaborative Filtering model (CNN-DCF) to solve the key problem in the recommendation system. Based on the outer product and deep neural networks, we develop a correlation extraction module that can learn high-order correlations between item latent features and user latent features. Extensive experiments on the public implicit feedback dataset Yelp show that the proposed CNN-DCF model brings significant improvements over the state-of-the-art methods.

In recent years, the proposed Deep Interest Network (DIN), Deep Interest Evolution Network (DIEN) and Deep Session Interest Network (DSIN) have further developed click-through rate prediction models. The above three models mainly focus on the evolution and development of the user’s historical behavior sequence. To a certain extent, the influence of environmental vectors on the user’s choice of the advertisement for the item to be recommended is ignored. As a result, click-through rates cannot be predicted more accurately when items have strong environmental attributes. To solve this problem, we propose a new model based on DIN, called Deep Interest Context Network (DICN). DICN combines two local activation units. It adaptively learns the user’s interest representation from the user’s historical behavior data concerning an advertisement and the context in which the advertisement is located (i.e., environmental factors). The experimental results show that DICN significantly improves the performance and model expression ability of advertisements with strong environmental attributes.