At present， how to capture the dynamic changes and dependencies of user behaviors is an important problem in the field of sequential recommendation， which mainly faces challenges such as large behavior event space and complex sequential dependencies of behaviors. To address the above challenges， a sequential recommendation algorithm based on the learning of latent states of behavioral sequences and their dependency relationships was proposed. Firstly， the low-dimensional representation of the latent states of behavioral sequences was obtained by using the maximum pooling hierarchical structure. Then， the dependencies between the latent states were captured and described by graph neural network in order to achieve the learning of user behavior change patterns， which led to more accurate sequential recommendation effect. Experimental results show that compared with the recent Hierarchical Gating Network （HGN） baseline algorithm on the IPTV， New York City （NYC） and Tokyo （TKY） datasets， the proposed algorithm improves the performance evaluation metric recall by 30.03%， 29.48% and 33.75% respectively， and obtains 37.20%， 43.47% and 40.34% relative improvements on Normalized Discounted Cumulative Gain （NDCG） metric， respectively. And the ablation experimental results demonstrate the effectiveness of dependency learning of sequential states. Therefore， the proposed algorithm is especially suitable for solving the problems with sparse behaviors in single time slice and complex behavioral dependencies in sequential recommendation.

Focusing on the issue that the user preferences change with time in the real system, and a user ID may be shared by multiple members of a family, a dynamic recommendation algorithm for the group-users who contained multiple types of members and have preferences varying with time was proposed. Firstly, it was assumed that the user's historical behavior data were composed of exposure data and click data, and the current member role was discriminated by learning the role weights of all types of members of the group-user at the present moment. Secondly, two design ideas were proposed according to the exposure data to construct a popularity model, and the training data were balanced by adopting the inverse propensity score weighting. Finally, the matrix factorization technique was used to obtain the user latent preference factor varying with time and the item latent attribute factor, and the inner products of the former and the latter were calculated to obtain the Top- K preference recommendations of the user which vary with time. Experimental results show that the proposed algorithm not only outperforms the benchmark method at least 16 moments in 24 moments a day on three metrics of Recall, Mean Average Precision (MAP), and Normalized Discounted Cumulative Gain (NDCG), but also shortens the running time and reduces the time complexity of calculation.

Block Diagonal Representation (BDR) model can efficiently cluster data by using linear representation, but it cannot make good use of non-linear manifold information commonly appeared in high-dimensional data. To solve this problem, the improved Block Diagonal Representation based on Neighbor Graph (BDRNG) clustering algorithm was proposed to perform the linear fitting of the local geometric structure by the neighbor graph and generate the block-diagonal structure by using the block-diagonal regularization. In BDRNG algorithm, both global information and local data structure were learned at the same time to achieve a better clustering performance. Due to the fact that the model contains the neighbor graph and non-convex block-diagonal representation norm, the alternative minimization was adopted by BDRNG to optimize the solving algorithm. Experimental results show that:on the noise dataset, BDRNG can generate the stable coefficient matrix with block-diagonal form, which proves that BDRNG is robust to the noise data; on the standard datasets, BDRNG has better clustering performance than BDR, especially on the facial dataset, BDRNG has the clustering accuracy 8% higher than BDR.

Aiming at how to merge the nodes' attributes and network structure information to realize the classification of social network nodes, a social network node classification algorithm based on graph encoder network was proposed. Firstly, the information of each node was propagated to its neighbors. Secondly, for each node, the possible implicit relationships between itself and its neighbor nodes were mined through neural network, and these relationships were merged together. Finally, the higher-level features of each node were extracted based on the information of the node itself and the relationships with the neighboring nodes and were used as the representation of the node, and the node was classified according to this representation. On the Weibo dataset, compared with DeepWalk model, logistic regression algorithm and the recently proposed graph convolutional network, the proposed algorithm has the classification accuracy greater than 8%; on the DBLP dataset, compared with multilayer perceptron, the classification accuracy of this algorithm is increased by 4.83%, and is increased by 0.91% compared with graph convolutional network.

The existing wireless network performance optimization methods are mainly based on the correlation analysis between indicators, and cannot effectively guide the design of optimization strategies and some other interventions. Thus, a Canonical Causal Inference (CCI) algorithm was proposed and used for wireless network performance optimization. Firstly, concerning that wireless network performance is usually presented by numerous correlated indicators, the Canonical Correlation Analysis (CCA) method was employed to extract atomic events from indicators. Then, typical causal inference method was conducted on the extracted atomic events to find the causality among the atomic events. The above two stages were iterated to determine the causal network of the atomic events and provided a robust and effective basis for wireless network performance optimization. The validity of CCI was indicated by simulation experiments, and some valuable causal relations of wireless network indicators were found on the data of a city's more than 30000 mobile base stations.

The analysis and study of readers' emotion is helpful to find negative information of the Internet, and it is an important part of public opinion monitoring. Taking into account the main factors that lead to the different emotions of readers is the semantic content of the text, how to extract semantic features of the text has become an important issue. To solve this problem, the initial features related to the semantic content of the text was expressed by word2vec model. On the basis of that, representative semantic word clusters were established for all emotion categories. Furthermore, a strategy was adopted to select the representative word clusters that are helpful for emotion classification, thus the traditional text word vector was transformed to the vector on semantic word clusters. Finally, the multi-label classification was implemented for the emotion label learning and classification. Experimental results demonstrate that the proposed method achieves better accuracy and stability compared with state-of-the-art methods.

Without any prior information about data distribution, parameter and the labels of data, not all base clustering results can truly benefit for the combination decision of clustering ensemble. In addition, if each base clustering plays the same role, the performance of clustering ensemble may be weakened. This paper proposed a selective K-means clustering ensemble based on random sampling, called RS-KMCE. In RS-MKCE, random sampling can avoid local minimum in the process of selecting base clustering subset for ensemble. And the defined evaluation index according to diversity and accuracy can lead to a better base clustering subset for improving the performance of clustering ensemble. The experiment results on two synthetic datasets and four UCI datasets show that performance of the proposed RS-KMCE is better than K-means, K-means clustering ensemble, and selective K-means clustering ensemble based on bagging.