Search

Electronic medical record (EMR) text word segmentation is the basis of natural language processing in medicine. Due to the characteristics of EMR, such as strong specialization, high cost of annotation, special writing style and sustained growth of terminology, the current Chinese word segmentation (CWS) methods cannot fully meet the requirements of the application of EMR. In order to solve this problem, an EMR word segmentation model based on Graph Neural Network (GNN), bidirectional Long Short-Term Memory network (Bi-LSTM) and conditional random field (CRF) is designed in this paper to improve the segmentation effect and reduce the dependence on data set. In the model, GNN based on the domain lexicon is used to learn the local composition features, Bi-LSTM is used to capture the long-term dependence and context sequence information, and CRF is used to obtain the optimal annotation sequence based on the sentence level label information. Through multi-feature interaction, the ambiguity resolution and new word recognition in the EMR word segmentation are effectively carried out. Compared with CWS tools such as Jieba and Pkuseg, as well as baseline models and state-of-the-art methods, the precision and recall rate of the model in this paper have been significantly improved.

Topic models, such as probabilistic latent semantic analysis (PLSA) and latent Dirichlet allocation (LDA), have shown impressive success in many fields. Recently, multi-view learning via probabilistic latent semantic analysis (MVPLSA), is also designed for multi-view topic modeling. These approaches are instances of generative model, whereas they all ignore the manifold structure of data distribution, which is generally useful for preserving the nonlinear information. In this paper, we propose a novel multiple graph regularized generative model to exploit the manifold structure in multiple views. Specifically, we construct a nearest neighbor graph for each view to encode its corresponding manifold information. A multiple graph ensemble regularization framework is proposed to learn the optimal intrinsic manifold. Then, the manifold regularization term is incorporated into a multi-view topic model, resulting in a unified objective function. The solutions are derived based on the Expectation Maximization optimization framework. Experimental results on real-world multi-view data sets demonstrate the effectiveness of our approach.

An increasing number of reviews from the customers have been available online. Thus, sentiment classification for such reviews has attracted more and more attention from the natural language processing (NLP) community. Related literature has shown that sentiment analysis can benefit from Deep Belief Networks (DBN). However, determining the structure of the deep network and improving its performance still remains an open question. In this paper, we propose a sophisticated algorithm based on fuzzy mathematics and genetic algorithm, called evolutionary fuzzy deep belief networks with incremental rules (EFDBNI). We evaluate our proposal using empirical data sets that are dedicated for sentiment classification. The results show that EFDBNI brings out significant improvement over existing methods.

Heterogeneous data sources and multi-label are two important characteristics of protein function prediction. They describe protein data from two different aspects. However, it is of considerable challenge to integrate multiple data sources and multi-label simultaneously for predicting protein functions, especially when there are only a limited number of labeled proteins. In this paper, we propose a generative model with hypergraph regularizers algorithm, called GMHR, for predicting proteins with multiple functions. The GMHR algorithm integrates all data sources that are available, including protein attribute features, interaction networks, label correlations, and unlabeled data. Experimental results on the real-world datasets predicting the functions of proteins demonstrate the superiority of our proposed method compared with the state-of-the-art baselines.

Presently, many users are involved in multiple social networks. Identifying the same user in different networks, also known as anchor link prediction, becomes an important problem, which can serve numerous applications, e.g., cross-network recommendation, user profiling, etc. Previous studies mainly use hand-crafted structure features, which, if not carefully designed, may fail to reflect the intrinsic structure regularities. Moreover, most of the methods neglect the attribute information of social networks. In this paper, we propose a novel semi-supervised network-embedding model to address the problem. In the model, each node of the multiple networks is represented by a vector for anchor link prediction, which is learnt with awareness of observed anchor links as semi-supervised information, and topology structure and attributes as input. Experimental results on the real-world data sets demonstrate the superiority of the proposed model compared to state-of-the-art techniques.

A representative sample of a social network is essential for many internet services that rely on accurate analysis. A good sampling method for social network should be able to generate small sample network with similar structures and distributions as its original network. In this paper, a sampling algorithm based on graph partition, sampling based on graph partition (SGP), is proposed to sample social networks. SGP firstly partitions the original network into several sub-networks, and then samples in each sub-network evenly. This procedure enables SGP to effectively maintain the topological similarity and community structure similarity between the sampled network and its original network. Finally, we evaluate SGP on several well-known datasets. The experimental results show that SGP method outperforms seven state-of-the-art methods.

Deriving a representative sample from a big social network is essential for many Internet services that rely on accurate analysis of big social data. A good sampling method for social network should be able to generate small sample networks with similar structures as original big network. In this paper, we propose SGP, a new big social network sampling algorithm based on graph partition. In SGP, original network is firstly partitioned into several sub-networks that will be sampled evenly. This procedure enables SGP to effectively maintain the topological similarity and community structure similarity between the sampled network and its original network. We have evaluated SGP on several well-known data sets. The experimental results show that SGP outperforms six state-of-the-art methods.

One popular solution to deal with large-scale relational networks is to derive a representative sample from huge relational networks. We expect this sample could represent the origin relational network well so that the sampled network can be used for simulations and further analysis instead of the origin one. In this paper, we propose a network stratified sampling algorithm using topologically divided stratums for large relational networks, which can maintain the topological similarity well between sampled network and original network. In addition, we evaluate our algorithm on several well-known data sets. The experimental results show that our algorithm outperforms the previous methods.

The key idea of active learning is that it can perform better with less data or costs if a machine learner is allowed to choose the data actively. However, the relation between labeling cost and model performance is seldom studied in the literature. In this paper, we thoroughly study this problem and give a criterion called as cost-performance to balance this relation. Based on the criterion, a cost-driven active SSC algorithm is proposed, which can stop the active process automatically. Empirical results show that our method outperforms active SVM and co-EMT.

The hierarchy clustering methods based on vertex similarity have the advantage that global evaluation can be incorporated for community discovery. Vertex similarity metric is the most important part of these methods. However, the existing methods perform not well for community discovery compared with the state-of-the-art algorithms. In this paper, we propose a new vertex similarity metric based on local flow model, called Local Flow Metric (LFM), for community discovery. LFM considers both the number of connecting paths and local edge density which are essential measures in community structure. Compared with the existing metrics of vertex similarity, LFM outperforms substantially in community discovery quality and the computing time. Furthermore, our LFM algorithm is superior to the state-of-the-art algorithms in some aspects.

We describe an area-adaptive multi-level layout for visualising social networks. This area-adaptive layout can visualise social networks according to the display area and community structure, which could reasonably utilise the display area and enhance the community features. The whole process consists of two parts: graph multi-layered compression and top-down multi-level layout. The multi-layered compression process groups vertices to form clusters and then abstracts the clusters as new vertices to define a new graph and is repeated until the graph size falls below some threshold. Based on the compressed graph, we optimise the display area to top-down positions all vertices. We have evaluated our layout on several well-known data sets. The experimental results show that our layout outperforms the state-of-the-art methods.

One challenge of visualizing the relational networks in computer screen is the scale of relational networks is too large. One solution is deriving a representative sample from a huge real graph. The purpose is to select a set of vertices and edges in graph so that the induced graph obeys some general characteristics, and so the sampled graphs can be used for simulations and further analysis. In this paper, we propose a stratified sampling algorithm using topologically divided stratums for large relational networks. In addition, we evaluate our algorithm on several well-known datasets. The experimental results show that our algorithm outperforms the previous methods.