Despite significant successes achieved in knowledge discovery, traditional machine learning methods may fail to obtain satisfactory performances when dealing with complex data, such as imbalanced, high-dimensional, noisy data, etc. The reason behind is that it is difficult for these methods to capture multiple characteristics and underlying structure of data. In this context, it becomes an important topic in the data mining field that how to effectively construct an efficient knowledge discovery and mining model. Ensemble learning, as one research hot spot, aims to integrate data fusion, data modeling, and data mining into a unified framework. Specifically, ensemble learning firstly extracts a set of features with a variety of transformations. Based on these learned features, multiple learning algorithms are utilized to produce weak predictive results. Finally, ensemble learning fuses the informative knowledge from the above results obtained to achieve knowledge discovery and better predictive performance via voting schemes in an adaptive way. In this paper, we review the research progress of the mainstream approaches of ensemble learning and classify them based on different characteristics. In addition, we present challenges and possible research directions for each mainstream approach of ensemble learning, and we also give an extra introduction for the combination of ensemble learning with other machine learning hot spots such as deep learning, reinforcement learning, etc.
Product quantization is now considered as an effective approach to solve the approximate nearest neighbor (ANN) search. A collection of derivative algorithms have been developed. However, the current techniques ignore the intrinsic high order structures of data, which usually contain helpful information for improving the computational precision. In this paper, aiming at the complex structure of high order data, we design an optimized technique, called optimized high order product quantization (O-HOPQ) for ANN search. In O-HOPQ, we incorporate the high order structures of the data into the process of designing a more effective subspace decomposition way. As a result, spatial adjacent elements in the high order data space are grouped into the same subspace. Then, O-HOPQ generates its spatial structured codebook, by optimizing the quantization distortion. Starting from the structured codebook, the global optimum quantizers can be obtained effectively and efficiently. Experimental results show that appropriate utilization of the potential information that exists in the complex structure of high order data will result in significant improvements to the performance of the product quantizers. Besides, the high order structure based approaches are effective to the scenario where the data have intrinsic complex structures.
Recently, many online Karaoke (KTV) platforms have been released, where music lovers sing songs on these platforms. In the meantime, the system automatically evaluates user proficiency according to their singing behavior. Recommending approximate songs to users can initialize singers’ participation and improve users’ loyalty to these platforms. However, this is not an easy task due to the unique characteristics of these platforms. First, since users may be not achieving high scores evaluated by the system on their favorite songs, how to balance user preferences with user proficiency on singing for song recommendation is still open. Second, the sparsity of the user-song interaction behavior may greatly impact the recommendation task. To solve the above two challenges, in this paper, we propose an informationfused song recommendationmodel by considering the unique characteristics of the singing data. Specifically, we first devise a pseudo-rating matrix by combing users’ singing behavior and the system evaluations, thus users’ preferences and proficiency are leveraged. Then wemitigate the data sparsity problem by fusing users’ and songs’ rich information in the matrix factorization process of the pseudo-ratingmatrix. Finally, extensive experimental results on a real-world dataset show the effectiveness of our proposed model.
Entity linking (EL) systems aim to link entity mentions in the document to their corresponding entity records in a reference knowledge base. Existing EL approaches usually ignore the semantic correlation between the mentions in the text, and are limited to the scale of the local knowledge base. In this paper, we propose a novel graphranking collective Chinese entity linking (GRCCEL) algorithm, which can take advantage of both the structured relationship between entities in the local knowledge base and the additional background information offered by external knowledge sources. By improved weighted word2vec textual similarity and improved PageRank algorithm, more semantic information and structural information can be captured in the document. With an incremental evidence mining process, more powerful discrimination capability for similar entities can be obtained.We evaluate the performance of our algorithm on some open domain corpus. Experimental results show the effectiveness of our method in Chinese entity linking task and demonstrate the superiority of our method over state-of-the-art methods.
The task of next POI recommendations has been studied extensively in recent years. However, developing a unified recommendation framework to incorporate multiple factors associated with both POIs and users remains challenging, because of the heterogeneity nature of these information. Further, effective mechanisms to smoothly handle cold-start cases are also a difficult topic. Inspired by the recent success of neural networks in many areas, in this paper, we propose a simple yet effective neural network framework, named NEXT, for next POI recommendations. NEXT is a unified framework to learn the hidden intent regarding user’s next move, by incorporating different factors in a unified manner. Specifically, in NEXT, we incorporatemeta-data information, e.g., user friendship and textual descriptions of POIs, and two kinds of temporal contexts (i.e., time interval and visit time). To leverage sequential relations and geographical influence, we propose to adopt DeepWalk, a network representation learning technique, to encode such knowledge. We evaluate the effectiveness of NEXT against other state-of-the-art alternatives and neural networks based solutions. Experimental results on three publicly available datasets demonstrate that NEXT significantly outperforms baselines in real-time next POI recommendations. Further experiments show inherent ability of NEXT in handling cold-start.
Despite the demonstrated success of numerous correlation filter (CF) based tracking approaches, their assumption of circulant structure of samples introduces significant redundancy to learn an effective classifier. In this paper, we develop a fast manifold regularized context-aware correlation tracking algorithm that mines the local manifold structure information of different types of samples. First, different from the traditional CF based tracking that only uses one base sample, we employ a set of contextual samples near to the base sample, and impose a manifold structure assumption on them. Afterwards, to take into account the manifold structure among these samples, we introduce a linear graph Laplacian regularized term into the objective of CF learning. Fortunately, the optimization can be efficiently solved in a closed form with fast Fourier transforms (FFTs), which contributes to a highly efficient implementation. Extensive evaluations on the OTB100 and VOT2016 datasets demonstrate that the proposed tracker performs favorably against several state-of-the-art algorithms in terms of accuracy and robustness. Especially, our tracker is able to run in real-time with 28 fps on a single CPU.
Indefinite kernel support vector machine (IKSVM) has recently attracted increasing attentions in machine learning. Since IKSVM essentially is a non-convex problem, existing algorithms either change the spectrum of indefinite kernel directly but risking losing some valuable information or solve the dual form of IKSVM whereas suffering from a dual gap problem. In this paper, we propose a primal perspective for solving the problem. That is, we directly focus on the primal form of IKSVM and present a novel algorithm termed as IKSVM-DC for binary and multi-class classification. Concretely, according to the characteristics of the spectrum for the indefinite kernel matrix, IKSVM-DC decomposes the primal function into the subtraction of two convex functions as a difference of convex functions (DC) programming. To accelerate convergence rate, IKSVM-DC combines the classical DC algorithm with a line search step along the descent direction at each iteration. Furthermore, we construct a multi-class IKSVM model which can classify multiple classes in a unified form. A theoretical analysis is then presented to validate that IKSVM-DC can converge to a local minimum. Finally, we conduct experiments on both binary and multi-class datasets and the experimental results show that IKSVM-DC is superior to other state-of-the-art IKSVM algorithms.
Traditional route planners commonly focus on finding the shortest path between two points in terms of travel distance or time over road networks. However, in real cases, especially in the era of smart cities where many kinds of transportation-related data become easily available, recent years have witnessed an increasing demand of route planners that need to optimize for multiple criteria, e.g., finding the route with the highest accumulated scenic score along (utility) while not exceeding the given travel time budget (cost). Such problem can be viewed as a variant of arc orienteering problem (AOP), which is well-known as an NP-hard problem. In this paper, targeting a more realistic AOP, we allow both scenic score (utility) and travel time (cost) values on each arc of the road network are time-dependent (2TD-AOP), and propose a memetic algorithm to solve it. To be more specific, within the given travel time budget, in the phase of initiation, for each population, we iteratively add suitable arcs with high scenic score and build a path fromthe origin to the destination via a complicate procedure consisting of search region narrowing, chromosome encoding and decoding. In the phase of the local search, each path is improved via chromosome selection, local-improvement-based mutation and crossover operations. Finally, we evaluate the proposed memetic algorithm in both synthetic and real-life datasets extensively, and the experimental results demonstrate that it outperforms the baselines.
In speech recognition research, because of the variety of languages, corresponding speech recognition systems need to be constructed for different languages. Especially in a dialect speech recognition system, there are many special words and oral language features. In addition, dialect speech data is very scarce. Therefore, constructing a dialect speech recognition system is difficult. This paper constructs a speech recognition system for Sichuan dialect by combining a hidden Markov model (HMM) and a deep long short-term memory (LSTM) network. Using the HMM-LSTM architecture, we created a Sichuan dialect dataset and implemented a speech recognition system for this dataset. Compared with the deep neural network (DNN), the LSTM network can overcome the problem that the DNN only captures the context of a fixed number of information items. Moreover, to identify polyphone and special pronunciation vocabularies in Sichuan dialect accurately, we collect all the characters in the dataset and their common phoneme sequences to form a lexicon. Finally, this system yields a 11.34% character error rate on the Sichuan dialect evaluation dataset. As far as we know, it is the best performance for this corpus at present.
Mining outliers in heterogeneous networks is crucial to many applications, but challenges abound. In this paper, we focus on identifying meta-path-based outliers in heterogeneous information network (HIN), and calculate the similarity between different types of objects. We propose a meta-path-based outlier detection method (MPOutliers) in heterogeneous information network to deal with problems in one go under a unified framework. MPOutliers calculates the heterogeneous reachable probability by combining different types of objects and their relationships. It discovers the semantic information among nodes in heterogeneous networks, instead of only considering the network structure. It also computes the closeness degree between nodes with the same type, which extends the whole heterogeneous network. Moreover, each node is assigned with a reliable weighting to measure its authority degree. Substantial experiments on two real datasets (AMiner and Movies dataset) show that our proposed method is very effective and efficient for outlier detection.
There are large and growing textual corpora in which people express contrastive opinions about the same topic. This has led to an increasing number of studies about contrastive opinion mining. However, there are several notable issues with the existing studies. They mostly focus on mining contrastive opinions from multiple data collections, which need to be separated into their respective collections beforehand. In addition, existing models are opaque in terms of the relationship between topics that are extracted and the sentences in the corpus which express the topics; this opacity does not help us understand the opinions expressed in the corpus. Finally, contrastive opinion is mostly analysed qualitatively rather than quantitatively. This paper addresses these matters and proposes a novel unified latent variable model (contraLDA), which: mines contrastive opinions from both single and multiple data collections, extracts the sentences that project the contrastive opinion, and measures the strength of opinion contrastiveness towards the extracted topics. Experimental results show the effectiveness of our model in mining contrasted opinions, which outperformed our baselines in extracting coherent and informative sentiment-bearing topics. We further show the accuracy of our model in classifying topics and sentiments of textual data, and we compared our results to five strong baselines.
Despite demonstrated success of SVM based trackers, their performance remains a boosting room if carefully considering the following factors: first, the tradeoff between sampling and budgeting samples affects tracking accuracy and efficiency much; second, how to effectively fuse different types of features to learn a robust target representation plays a key role in tracking accuracy. In this paper, we propose a novel SVM based tracking method that handles the first factor with the help of the circulant structures of the samples and the second one by a multi-kernel learning mechanism. Specifically, we formulate an SVM classification model for visual tracking that incorporates two types of kernels whose matrices are circulant, fully taking advantage of the complementary traits of the color and HOG features to learn a robust target representation. Moreover, it is fortunate that the SVM model has a closed-form solution in terms of both the classifier weights and the kernel weights, and both can be efficiently computed via fast Fourier transforms (FFTs). Extensive evaluations on OTB100 and VOT2016 visual tracking benchmarks demonstrate that the proposed method achieves a favorable performance against various state-of-the-art trackers with a speed of 50 fps on a single CPU.
In the past decade, recommender systems have been widely used to provide users with personalized products and services. However, most traditional recommender systems are still facing a challenge in dealing with the huge volume, complexity, and dynamics of information. To tackle this challenge, many studies have been conducted to improve recommender system by integrating deep learning techniques. As an unsupervised deep learning method, autoencoder has been widely used for its excellent performance in data dimensionality reduction, feature extraction, and data reconstruction. Meanwhile, recent researches have shown the high efficiency of autoencoder in information retrieval and recommendation tasks. Applying autoencoder on recommender systems would improve the quality of recommendations due to its better understanding of users’ demands and characteristics of items. This paper reviews the recent researches on autoencoder-based recommender systems. The differences between autoencoder-based recommender systems and traditional recommender systems are presented in this paper. At last, some potential research directions of autoencoder-based recommender systems are discussed.
RNA 5-methylcytosine (m5C) sites perform a major role in numerous biological processes and commonly reported in both DNA and RNA cellular. The enzymatic mechanism and biological functions of m5C sites in DNA remain the focusing area of researchers for last few decades. Likewise, the investigators also targeted m5C sites in RNA due to its cellular functions, positioning and formation mechanism. Currently, several rudimentary roles of the m5C in RNA have been explored, but a lot of improvements are still under consideration. Initially, the identification of RNA methylcytosine sites was carried out via experimental methods, which were very hard, erroneous and time consuming owing to partial availability of recognized structures. Looking at the significance of m5C role in RNA, scientists have diverted their attention from structure to sequence-based prediction. In this regards, an intelligent computational model is proposed in order to identify m5C sites in RNA with high precision. Three RNA sequences formulation methods namely: pseudo dinucleotide composition,pseudo trinucleotide composition and pseudo tetra nucleotide composition are applied to extract variant and high profound numerical features. In a sequel, the vector spaces are fused to build a hybrid space in order to compensate the weakness of each other. Various learning hypotheses are examined to select the best operational engine, which can truly identify the pattern of the target class. The strength and generalization of the proposed model are measured using two different cross validation tests. The reported outcomes reveal that the proposed model achieved 3% better accuracy than that of the highest present approach in the literature so far.