Time filter

Source Type

Hefei, China

Chen Y.,Hefei University of Technology | Li X.,IFLYTEK Research | Li L.,Wuhan University of Technology | Liu G.,Hefei University of Technology | Xu G.,University of Technology, Sydney
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2016

The pervasive employments of Location-based Social Network call for precise and personalized Point-of-Interest (POI) recommendation to predict which places the users prefer. Modeling user mobility, as an important component of understanding user preference, plays an essential role in POI recommendation. However, existing methods mainly model user mobility through analyzing the check-in data and formulating a distribution without considering why a user checks in at a specific place from psychological perspective. In this paper, we propose a POI recommendation algorithm modeling user mobility by considering check-in data and geographical information. Specifically, with check-in data, we propose a novel probabilistic latent factor model to formulate user psychological behavior from the perspective of utility theory, which could help reveal the inner information underlying the comparative choice behaviors of users. Geographical behavior of all the historical check-ins captured by a power law distribution is then combined with probabilistic latent factor model to form the POI recommendation algorithm. Extensive evaluation experiments conducted on two real-world datasets confirm the superiority of our approach over state-of-the-art methods. © Springer International Publishing Switzerland 2016.

Ding J.,Hefei University of Technology | Chen Y.,Hefei University of Technology | Li X.,IFLYTEK Research | Liu G.,Hefei University of Technology | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2016

Personalized Recommendation has drawn greater attention in academia and industry as it can help people filter out massive useless information. Several existing recommender techniques exploit social connections, i.e., friends or trust relations as auxiliary information to improve recommendation accuracy. However, opinion leaders in each circle tend to have greater impact on recommendation than those of friends with different tastes. So we devise two unsupervised methods to identify opinion leaders that are defined as experts. In this paper, we incorporate the influence of experts into circle-based personalized recommendation. Specifically, we first build explicit and implicit social networks by utilizing users’ friendships and similarity respectively. Then we identify experts on both social networks. Further, we propose a circle-based personalized recommendation approach via fusing experts’ influences into matrix factorization technique. Extensive experiments conducted on two datasets demonstrate that our approach outperforms existing methods, particularly on handing cold-start problem. © Springer International Publishing Switzerland 2016.

Xia X.-J.,Anhui University of Science and Technology | Ling Z.-H.,Anhui University of Science and Technology | Jiang Y.,IFLYTEK Research | Dai L.-R.,Anhui University of Science and Technology
Speech Communication | Year: 2014

This paper presents a hidden Markov model (HMM) based unit selection speech synthesis method using log likelihood ratios (LLR) derived from perceptual data. The perceptual data is collected by judging the naturalness of each synthetic prosodic word manually. Two acoustic models which represent the natural speech and the unnatural synthetic speech are trained respectively. At synthesis time, the LLRs are derived from the estimated acoustic models and integrated into the unit selection criterion as target cost functions. The experimental results show that our proposed method can synthesize more natural speech than the conventional method using likelihood functions. Due to the inadequacy of the acoustic model estimated for the unnatural synthetic speech, utilizing the LLR-based target cost functions to rescore the pre-selection results or the N-best sequences can achieve better performance than substituting them for the original target cost functions directly. © 2014 Elsevier B.V. All rights reserved.

Ding H.,National School of Technology | Pan J.,IFLYTEK Research | Shen M.,University of Konstanz | Shen M.,South China University of Technology
2015 IEEE International Conference on Information and Automation, ICIA 2015 - In conjunction with 2015 IEEE International Conference on Automation and Logistics | Year: 2015

Objective measures are favored and widely used by many researchers in evaluating the quality of noise-suppressed speech. A good and reliable objective measure should have property that it could evaluate speech quality in consistent and well correlated with subjective ratings. In this paper, several widely used objective measures are applied to the speech signals with the Chinese languages including Mandarin and Cantonese. The correlations between objective measure outputs and perceptual-subjective ratings are reported and analyzed. The experimental results show that the correlation with the language types of Mandarin and Cantonese are lower than the one with English and objective measures behave differently in Mandarin, Cantonese and English. Detail discussion and conclusion are presented as well. © 2015 IEEE.

Du J.,Microsoft | Hu Y.,IFLYTEK Research | Jiang H.,York University
IEEE Transactions on Audio, Speech and Language Processing | Year: 2011

In this paper, we apply the well-known boosted mixture learning (BML) method to learn Gaussian mixture HMMs in speech recognition. BML is an incremental method to learn mixture models for classification problems. In each step of BML, one new mixture component is estimated according to the functional gradient of an objective function to ensure that it is added along the direction that maximizes the objective function. Several techniques have been proposed to extend BML from simple mixture models like the Gaussian mixture model (GMM) to the Gaussian mixture hidden Markov model (HMM), including Viterbi approximation for state segmentation, weight decay and sampling boosting to initialize sample weights to avoid overfitting, combination between partial updating and global updating to refine model parameters in each BML iteration, and use of the Bayesian Information Criterion (BIC) for parsimonious modeling. Experimental results on two large-vocabulary continuous speech recognition tasks, namely the WSJ-5k and Switchboard tasks, have shown that the proposed BML yields significant performance gain over the conventional training procedure, especially for small model sizes. © 2006 IEEE.

Discover hidden collaborations