Location-based social networks provide services that allow users to share their locations with friends. To attract users and keep them active, social networks or venue holders may offer some awards. But some users make fake check-ins to achieve these awards. These cheaters cause monetary loss and decrease the accuracy of venue recommendations. In this paper, we study users of Foursquare, a popular location-based social network. Behaviors of cheaters and normal users are discussed. Two types of connections are defined to construct graphs of these users. And we propose a method to find cheaters using community structure of the constructed graphs. Our results verify that this cheater detection method is effective and costs little.

Classical supervised machine learning techniques have been explored for semantically annotating unstructured textual data such as consumers' comments archived at social media websites to extract business intelligence. However, these techniques often require a large number of manually labeled training examples to produce accurate annotations. Several active learning approaches that are designed based on probabilistic sequence models have been explored to minimize the number of labeled training examples for semantic annotation tasks. Recent research has shown that large-margin classifiers are viable alternatives to automated semantic annotation, given their strong generalization capabilities and the ability to process high-dimensional data. However, the existing active learning methods that are designed for probabilistic sequence models cannot be easily adapted and applied to large-margin classifiers. The main contribution of this paper is the development of novel active learning methods for large-margin classifiers to fill the aforementioned research gap. In particular, we propose an innovative perspective of taking active learning as a search of optimal parameters for large-margin classifiers. A rigorous evaluation involving two benchmark tests and an empirical test based on real-world data extracted from Amazon.com reveals that the proposed active learning methods can train effective classifiers with significantly fewer training examples while achieving similar annotation performance, compared to a typical state-of-the-art classifier that only uses several labeled training examples. More specifically, one of our proposed active learning methods can reduce the number of training examples by 19.74% at the 68% level of F1 when compared to the best baseline method, as evaluated based on the Amazon data set. Our research opens the door to the application of intelligent semantic annotation techniques to support real-world applications such as automatically analyzing consumer comments for customer relationship management.

This paper proposes a model to calculate the average speed of transmission of intervehicle communication (IVC) messages in a general traffic stream on highways in the early stage of deploying distributed traffic information systems (DTIS). The model helps explain the relationship between average IVC message speed and traffic parameters such as equipped vehicle density, traffic flow speed, and traffic direction. Simulation results are used to verify the correctness of the model. This model needs a much shorter calculation time than a simulation. Moreover, the theoretical analysis helps provide more insightful explanations of the phenomenon for IVC performance analysis. The results of this paper would help people better understand the design criteria for DTIS.

With the proliferation of its applications in various industries, sentiment analysis by using publicly available web data has become an active research area in text classification during these years. It is argued by researchers that semi-supervised learning is an effective approach to this problem since it is capable to mitigate the manual labeling effort which is usually expensive and time-consuming. However, there was a long-term debate on the effectiveness of unlabeled data in text classification. This was partially caused by the fact that many assumptions in theoretic analysis often do not hold in practice. We argue that this problem may be further understood by adding an additional dimension in the experiment. This allows us to address this problem in the perspective of bias and variance in a broader view. We show that the well-known performance degradation issue caused by unlabeled data can be reproduced as a subset of the whole scenario. We argue that if the bias-variance trade-off is to be better balanced by a more effective feature selection method unlabeled data is very likely to boost the classification performance. We then propose a feature selection framework in which labeled and unlabeled training samples are both considered. We discuss its potential in achieving such a balance. Besides, the application in financial sentiment analysis is chosen because it not only exemplifies an important application, the data possesses better illustrative power as well. The implications of this study in text classification and financial sentiment analysis are both discussed.

This paper presents a hybrid approach to automatic incident detection (AID) in transportation systems. It combines time series analysis (TSA) and machine learning (ML) techniques in light of the fault diagnosis theory. In this approach, the time series component is to forecast the normal traffic for the current time point based on prior (normal) traffic. The ML component aims to detect incidents using features of real-time traffic, predicted normal traffic, as well as differences between the two. We validate our approach using a real-world data set collected in previous research. The results show that the hybrid approach is able to detect incidents more accurately [higher detection rate (DR)] and faster (shorter mean time to detect) under the requirement of a similar false alarm rate (FAR), as compared with state-of-the-art algorithms. This paper lends support to further studies on combining TSA with ML to address problems related to intelligent transportation systems (ITS).

Because of significant improvements in cost, accuracy, and coverage over dedicated traffic infrastructures, GPS-enabled mobile devices are preferred for continuous collection of traffic data. Estimating traffic states accurately from the obtained GPS data streams has great potential to increase efficiency of the existing traffic systems and to help reduce commuting time and fuel consumption. In this paper, first we propose a novel method to reasonably process GPS data by increasing weights of recent records and high velocity, rather than employing the current two extreme and popular approaches: the naive method aggregating all records with equal weights and the sliding-window (SW) sampling method preserving only the most recent records. Then, in line with the existing works, the proposed weighted approach is explored in two ways: aggregate-based and sampling-based ways. The aggregate-based way is classical but somewhat specific to the particular goal of traffic state estimation, whereas the sampling-based way is somewhat complicated but provides a universal set of samples for performing a variety of analyses. In the sampling-based way, a heuristic method is proposed to accurately estimate traffic states using preserved samples. Both ways are leveraged to evaluate performance of the novel weighed method and the heuristic method for estimating traffic states using samples. Finally, the feasibility and effectiveness of these methods is experimentally validated using a field-experiment data set (Mobile Century) and three simulated data sets.

This paper presents a new method for learning Bayesian networks from functional dependencies (FD) and third normal form (3NF) tables in relational databases. The method sets up a linkage between the theory of relational databases and probabilistic reasoning models, which is interesting and useful especially when data are incomplete and inaccurate. The effectiveness and practicability of the proposed method is demonstrated by its implementation in a mobile commerce system.

To facilitate images under the nonlinear geometric transformation T and its inverse transformation T(-1), we have developed numerical algorithms. A cycle conversion T(-1)T of image transformations is said if an image is distorted by a transformation T and then restored back to itself. The combination (CSIM) of splitting-shooting-integrating methods was first proposed in Li for T(-1)T. In this paper other two combinations, CIIM and C I# I M, of splitting integrating methods for T(-1)T are provided. Combination CSIM has been successfully applied to many topics in image processing and pattern recognition. Since combination CSIM causes large greyness errors, it well suited to a few greyness level images, but needs a huge computation work for 256 greyness level images of enlarged transformations. We may instead choose combination CIIM which involves nonlinear solutions. However, the improved combination C I# I M may bypass the nonlinear solutions completely. Hence, both CIIM and C I# I M can be applied to q(q > or = 256) greyness level images of any enlarged transformations. On the other hand, the combined algorithms, CSIM, CIIM, and C I# I M, are applied to several important topics of image processing and pattern recognition: binary images, multi-greyness level images, image condensing, illumination, affine transformations, prospective and projection, wrapping images, handwriting characters, image concealment, the transformations with arbitrary shapes, and face transformation. This paper may also be regarded as a review of our recent research papers.

Finding similar sequences in time series has received much attention and is a widely studied topic. Most existing approaches in the time series area focus on the efficiency of algorithms but seldom provide a means to handle imprecise data. In this paper, a more general approach is proposed to measure the distance of time sequences containing crisp values, intervals, and fuzzy intervals as well. The concept of distance measurement and its associated dynamic-programming-based algorithms are described. In addition to finding the sequences with similar evolving trends, a means of finding the sequences with opposite evolving tendencies is also proposed, which is usually omitted in current related research but could be of great interest to many users.

Models of a fuzzy Petri net (FPN) are defined based on generating rules of a knowledge base and the Petri nets semantics description of logical rules are presented. Two kinds of reasoning patterns are considered. First, the beliefs of initial propositions are known and their beliefs of noninitial propositions need to be computed. Second, the beliefs of initial propositions are known and the beliefs of goal propositions need to be computed. The concurrent reasoning algorithms of the two patterns based on a multi-task schedule are proposed. The application of this model and reasoning algorithm is discussed.

System synthesis is one of the most important issues in automatic control. In a synthesis process, local system properties must be preserved in the global system. Most of the previous researches focuses on such properties as liveness, reachability, reversibility, and so forth. However, little has been done on system behaviors, that is, the language of firing sequences. In this paper, we study the behavior of systems using the Petri net model synthesis, with self-loop inhibitor, as well as synchronization operations. A new system behavior property, behavior invariance, is defined. Self-loop and inhibitor operations are transformed into synchronization operations, under which a set of preservation conditions for behavior invariance are obtained. The theoretical foundation for the preservation of system behavior properties is presented and proved, and the applicability of the theory to real-world automation systems demonstrated. An example for modeling a super-elevated conveyor is discussed, illustrating the usefulness of the proposed theory in modeling and analyzing the behavior invariance of such a conveyor.

Based on the concepts of the semantic proximity, we present a definition of the fuzzy functional dependency, We show that the inference rules for fuzzy functional dependencies, which are the same as Armstrong's axioms for the crisp case, are correct and complete. We also show that dependent constraints with dull values constitute a lattice. Functional dependencies in classical relational databases and null functional dependencies can be viewed as a special case of fuzzy functional dependencies. By applying the unified functional dependencies to the relational database design, we can represent the data with fuzzy values, null values and crisp values under relational database management systems, By using fuzzy functional dependencies, we can compress the range of a fuzzy value and make this fuzzy value "clearer".