Traffic demand prediction is of great importance for traffic management, yet it is also a challenging problem since the traffic data are usually with complex spatial-temporal dependencies and nonlinear relationships. In this paper, to better characterize and utilize the spatial and temporal features, we propose a Multi-dimensional Graph Convolutional Network (M-GCN) to capture dynamic spatial-temporal dependence of traffic data. M-GCN captures the explicit spatio-temporal dependencies by establishing a spatial adjacency graph and a temporal adjacency graph, and captures the hidden spatiotemporal dependencies by establishing a spatial adaptive graph and a temporal adaptive graph. We leverage graph convolutional networks for complex spatial and temporal dependencies modeling and design a gated fusion module to obtain the interactive spatial-temporal dependence. Besides, an attention mechanism is applied to alleviate the error propagation problem in long-term traffic prediction. Experimental results on two real-world traffic datasets demonstrate the superiority of M-GCN. The proposed M-GCN outperforms baseline methods by up to 4.6% improvement in MAE on the dataset TaxiNYC, and the training time and predicting time of $\mathrm{M}-\mathrm{GCN}$ are reduced by nearly half.

Semi-supervised cluster ensemble usually introduces a small amount of supervision in the first stage of cluster ensemble, i.e., ensemble generation, by performing many runs of semi-supervised clustering algorithms. However, it is neither efficient in terms of computational complexity, nor flexible in a dynamic learning environment where limited supervision changes over time. In this article we propose a new framework which generates base partitions in an unsupervised manner and attributes different weights to each cluster of the base partitions. The weighting scheme considers both the internal validation measures of clustering and the degrees of satisfaction of pairwise constraints. A weighted co-association matrix based consensus approach is then applied to achieve a final partition. To handle high-dimensional data, we generate base partitions using k-means with both random sampling and random subspace techniques. The new framework retains a high accuracy, and is efficient since it avoids performing semi-supervised clustering in ensemble generation and the complexity of the weighting scheme is independent of the number of instances in a dynamic environment. It is more adaptive than the traditional approach because it does not require rerunning semi-supervised clustering algorithms when the limited supervision changes. Empirical results on 12 datasets demonstrate that it is also more robust to noisy constraints.

As people acquire much more personal information as a result of personal and work activities, the management of these information becomes a serious problem and an important research issue. Modeling personal desktop activities and identifying them are two basic problems for supporting activity-based operations. To the best of our knowledge there is no literature on formalizing and identifying desktop activity from personal information management perspective. There are a number of challenges to this work, including the fact that people exhibit personalized behaviors, have individual interests, needs and resources, no available experimental data set, etc. In this paper, we perform a user experiment to learn about user desktop activities in a personal information management context. We collected information access activities in a naturalistic setting and propose a conceptual activity model by analyzing features of user behaviors at their desktop computers. We present an effective and efficient method of automatically identifying desktop activities. To evaluate performance of our method, we develop a prototype system to collect real users activities, and evaluate our methods for identifying activities. The results verify the effectiveness and efficiency of our methods.

Entity search is a new search pattern that return related entities to users rather than amounts of web pages containing mass and messy information. It is also a challenging research topic because it is difficult to understand the meaning of users' input and identify the entities from the messy web pages. In this paper, we propose an entity search pattern based on online encyclopedias and define it as MMK search(Multi-modifier Search), which means the input text by people only includes one kernel concept and multiple modifiers. We propose a solution framework to solve this kind of search, and propose a method to identify expected entities based on well-utilized online encyclopedias. To evaluate the methods, we create an experimental data set and a baseline under the help of participants, the results verified the effectiveness of our methods.

Tables and spreadsheets on the Internet often contain valuable information, but are created by people who have different individuation. As a result, the similar data are often issued with different structures. This limits the integration of such tables. This paper aims to overcome this problem by automatically analyzing the structure area and propose the method transforming the tables into formal relational tables. We propose the methods on identifying structure area, modeling the table structure based on tree and methods to generate the 1NF schema of the original table. We proved the correctness of the method in semantic and the experiment results with tables from different areas demonstrate the effectiveness of our method.

Recently there have been growing interests in the applications of wireless sensor networks such as traffic tracking, environmental surveillance, and network monitoring. In these applications, the exploration of the relationship and linkage of sensing data with other data sources can be naturally expressed by the external join, where the sensory tuples join with an external table at the base station. However, executing such kind of join queries in a highly distributed and resource-constraint sensor network is a challenging task. In this paper, we propose a partition-based algorithm called NEJA (in-Network External Join Algorithm) for the external join processing in sensor networks. NEJA organizes the sensory data of the network through an optimized "value-to-storage" mapping, according to which each storage point stores the tuples that belong to the same subrange on the joint attribute. Then the subrange of each storage point is further partitioned into unit ranges, and tuples in the same unit range wisely choose their joining point that incurs the least communication cost based on a cost metric according to the latest historical statistics. Also, NEJA adopts some optimization techniques to handle the changes of sensory data and uses approximate approaches to cut down the maintenance cost of the mechanism. The experimental results indicate that our scheme is effective in reducing the amount of transmissions for the real time external join processing, especially when the external table has a relatively large size.