Xiaodong Qiao

Multi-output least-squares support vector regression machines

Multi-output regression aims at learning a mapping from a multivariate input feature space to a multivariate output space. Despite its potential usefulness, the standard formulation of the least-squares support vector regression machine (LS-SVR) cannot cope with the multi-output case. The usual procedure is to train multiple independent LS-SVR, thus disregarding the underlying (potentially nonlinear) cross relatedness among different outputs. To address this problem, inspired by the multi-task learning methods, this study proposes a novel approach, Multi-output LS-SVR (MLS-SVR), in multi-output setting. Furthermore, a more efficient training algorithm is also given. Finally, extensive experimental results validate the effectiveness of the proposed approach.

Author-Topic over Time (AToT): A Dynamic Users’ Interest Model

One of the key problems in upgrading information services towards knowledge services is to automatically mine latent topics, users’ interests and their evolution patterns from large-scale S&T literatures. Most of current methods are devoted to either discover static latent topics and users’ interests, or to analyze topic evolution only from intra-features of documents, namely text content without considering directly extra-features of documents such as authors. To overcome this problem, a dynamic users’ interest model for documents using authors and topics with timestamps is proposed, named as Author-Topic over Time (AToT) model, and collapsed Gibbs sampling method is utilized for inferring model parameters. This model is not only able to discover latent topics and users’ interests, but also to mine their changing patterns over time. Finally, the extensive experimental results on NIPS dataset with 1,740 papers indicate that our AToT model is feasible and efficient.

A Dynamic Users’ Interest Discovery Model with Distributed Inference Algorithm

One of the key issues for providing users user-customized or context-aware services is to automatically detect latent topics, users’ interests, and their changing patterns from large-scale social network information. Most of the current methods are devoted either to discovering static latent topics and users’ interests or to analyzing topic evolution only from intrafeatures of documents, namely, text content, without considering directly extrafeatures of documents such as authors. Moreover, they are applicable only to the case of single processor. To resolve these problems, we propose a dynamic users’ interest discovery model with distributed inference algorithm, named as Distributed Author-Topic over Time (D-AToT) model. The collapsed Gibbs sampling method following the main idea of MapReduce is also utilized for inferring model parameters. The proposed model can discover latent topics and users’ interests, and mine their changing patterns over time. Extensive experimental results on NIPS (Neural Information Processing Systems) dataset show that our D-AToT model is feasible and efficient.

Multi-task least-squares support vector machines

There are often the underlying cross relatedness amongst multiple tasks, which is discarded directly by traditional single-task learning methods. Since multi-task learning can exploit these relatedness to further improve the performance, it has attracted extensive attention in many domains including multimedia. It has been shown through a meticulous empirical study that the generalization performance of Least-Squares Support Vector Machine (LS-SVM) is comparable to that of SVM. In order to generalize LS-SVM from single-task to multi-task learning, inspired by the regularized multi-task learning (RMTL), this study proposes a novel multi-task learning approach, multi-task LS-SVM (MTLS-SVM). Similar to LS-SVM, one only solves a convex linear system in the training phrase, too. What’s more, we unify the classification and regression problems in an efficient training algorithm, which effectively employs the Krylow methods. Finally, experimental results on school and dermatology validate the effectiveness of the proposed approach.

A Novel Approach for Measuring Chinese Terms Semantic Similarity Based on Pairwise Sequence Alignment

In this study, we first give a problem formulation for Chinese terms semantic similarity calculation. After that, on closer examination, we find that the traditional approach makes an implicit assumption that the order of corresponding primitive terms for two terms is roughly consistent. In other words, it doesn’t consider how the difference in the order affects the quality of correspondence. To overcome this problem, a novel approach based on pairwise sequence alignment is proposed. Finally, an experimental evaluation is conducted, and the result indicates that our approach outperforms or matches at least the traditional one in the majority of cases.

Uncovering Research Topics of Academic Communities of Scientific Collaboration Network

In order to improve the quality of applications, such as recommendation or retrieval in knowledge-based service system, it is very helpful to uncover research topics of academic communities in scientific collaboration network (SCN). Previous research mainly focuses on network characteristics measurement and community evolution, but it remains largely understudied on how to uncover research topics of each community. This paper proposes a nonjoint approach, consisting of three simple steps: (1) to detect overlapping academic communities in SCN with the clique percolation method, (2) to discover underlying topics and research interests of each researcher with author-topic (AT) model, and (3) to label research topics of each community with top N most frequent collaborative topics between members belonging to the community. Extensive experimental results on NIPS (neural information processing systems) dataset show that our simple procedure is feasible and efficient.

Chinese Scientific & Technical Vocabulary System for domain content computing

Chinese Scientific & Technical Vocabulary System (CSTVS) is a kind of knowledge organization systems for Chinese SCIENTIFIC & TECHNICAL information resources management and deep knowledge services that proposed by us. Now we provides the CSTVS as free semantic resources for research and education users all over the world. In this paper, knowledge infrastructure and related content computing tools are introduced. The related resources and tools of CSTVS in new energy vehicles domain (NEV-CSTVS) will provide as the first batch, which is used in some research projects. CSTVS will provide more semantic resources for domain content computing and that the uses of CSTVS will promote and improve the future large-scale practice.

Comparative Study on the Double-Array Structure for Large English & Chinese Lexicons

In this study, time and space efficiency of the double-array structure for large English & Chinese lexicons are comprehensively analyzed. Some important observations include: (1) both time and space efficiency are dependent of the different order of inserting the keys for Chinese lexicons, but neither for English ones; (2) on the condition that the order of inserting the keys is by characters’ numerical values, for Chinese lexicons, space efficiency is dependent of different character encoding methods, while time efficiency is not. Finally, a Chinese character encoding method based character frequency is raised, which improve further space efficiency to some extent.