Huaqing Min

Product aspect extraction supervised with online domain knowledge

One of the most challenging problems in aspect-based opinion mining is aspect extraction, which aims to identify expressions that describe aspects of products (called aspect expressions) and categorize domain-specific synonymous expressions. Although a number of methods of aspect extraction have been proposed before, very few of them are designed to improve the interpretability of generated aspects. Existing methods either generate multiple fine-grained aspects without proper categorization or categorize semantically unrelated product aspects (e.g., by unsupervised topic modeling). In this paper, we first examine previous studies on product aspect extraction. To overcome the limitations of existing methods, two novel semi-supervised models for product aspect extraction are then proposed. More specifically, the proposed methodology first extracts seeding aspects and related terms from detailed product descriptions readily available on E-commerce websites. Next, product reviews are regrouped according to these seeding aspects so that more effective textual contexts for topic modeling are built. Finally, two novel semi-supervised topic models are developed to extract human-comprehensible product aspects. For the first proposed topic model, the Fine-grained Labeled LDA (FL-LDA), seeding aspects are applied to guide the model to discover words that are related to these seeding aspects. For the second model, the Unified Fine-grained Labeled LDA (UFL-LDA), we incorporate unlabeled documents to extend the FL-LDA model so that words related to the seeding aspects or other high-frequency words in customer reviews are extracted. Our experimental results demonstrate that the proposed methods outperform state-of-the-art methods.

Personalized search for social media via dominating verbal context

With the rapid development of Web 2.0 communities, there has been a tremendous increase in user-generated content. Confronting such a vast volume of resources in collaborative tagging systems, users require a novel method for fast exploring and indexing so as to find their desired data. To this end, contextual information is indispensable and critical in understanding user preferences and intentions. In sociolinguistics, context can be classified as verbal context and social context. Compared with verbal context, social context requires not only domain knowledge to build pre-defined contextual attributes but also additional user data. However, to the best of our knowledge, no research has addressed the issue of irrelevant contextual factors for the verbal context model. To bridge this gap, the dominating set obtained from verbal context is proposed in this paper. We present (i) the verbal context graph to model contents and interrelationships of verbal context in folksonomy and thus capture the user intention; (ii) a method of discovering dominating set that provides a good balance of essentiality and integrality to de-emphasize irrelevant contextual factors and to keep the major characteristics of the verbal context graph; and (iii) a revised ranking method for measuring the relevance of a resource to an issued query, a discovered context and an extracted user profile. The experimental results obtained for a public dataset illustrate that the proposed method is more effective than existing baseline approaches.

Empirical analysis: stock market prediction via extreme learning machine

Abstract How to predict stock price movements based on quantitative market data modeling is an attractive topic. In front of the market news and stock prices that are commonly believed as two important market data sources, how to extract and exploit the hidden information within the raw data and make both accurate and fast predictions simultaneously becomes a challenging problem. In this paper, we present the design and architecture of our trading signal mining platform that employs extreme learning machine (ELM) to make stock price prediction based on those two data sources concurrently. Comprehensive experimental comparisons between ELM and the state-of-the-art learning algorithms, including support vector machine (SVM) and back-propagation neural network (BP-NN), have been undertaken on the intra-day tick-by-tick data of the H-share market and contemporaneous news archives. The results have shown that (1) both RBF ELM and RBF SVM achieve higher prediction accuracy and faster prediction speed than BP-NN; (2) the RBF ELM achieves similar accuracy with the RBF SVM and (3) the RBF ELM has faster prediction speed than the RBF SVM. Simulations of a preliminary trading strategy with the signals are conducted. Results show that strategy with more accurate signals will make more profits with less risk.

Folksonomy-based personalized search by hybrid user profiles in multiple levels

Recently, some systems have allowed users to rate and annotate resources, e.g., MovieLens, and we consider that it provides a way to identify favorite and non-favorite tags of a user by integrating his or her rating and tags. In this paper, we review and elaborate on the limitations of the current research on user profiling for personalized search in collaborative tagging systems. We then propose a new multi-level user profiling model by integrating tags and ratings to achieve personalized search, which can reflect not only a users likes but also a his or her dislikes. To the best of our knowledge, this is the first effort to integrate ratings and tags to model multi-level user profiles for personalized search.

NMFE-SSCC

Collective classification (CC) is a task to jointly classifying related instances of network data. Enabling CC usually improves the performance of predictive models on fully-labeled training networks with large amount of labeled data. However, acquiring such labels can be difficult and costly, and learning a CC classifier with only a few labeled data can lead to poor performance. On the other hand, there are usually large amount of unlabeled data available in practical. This naturally motivates semi-supervised collective classification (SSCC) approaches for leveraging the unlabeled data to improve CC from a sparsely-labeled network. In this paper, we propose a novel non-negative matrix factorization (NMF) based SSCC algorithm, called NMF-SSCC, to effectively learn a data representation by exploiting both labeled and unlabeled data on the network. Our idea is to use matrix factorization to obtain a compact representation of network data which uncovers the class discrimination of the data inferred from the labeled instances and simultaneously respects the intrinsic network structure. To achieve this, we design a new matrix factorization objective function and incorporate a label matrix factorization term as well as a network regularization term into it. An efficient optimization algorithm using the multiplicative updating rules is then developed to solve the new objective function. To further boost the predicting performance, we extend the proposed NMF-SSCC method into an ensemble scheme, called NMFE-SSCC, in terms of building a classification ensemble with a set of NMF-SSCC collective classifiers using different constructed latent graphs. Each NMF-SSCC classifier is learnt from one latent graph generated with various latent linkages for effectively label propagation. Experimental results on real-world data sets have demonstrated the effectiveness of the new methods.

Affordance Research in Developmental Robotics: A Survey

Affordances capture the relationships between a robot and the environment in terms of the actions that the robot is able to perform. The notable characteristic of affordance-based perception is that an object is perceived by what it affords (e.g., graspable and rollable), instead of identities (e.g., name, color, and shape). Affordances play an important role in basic robot capabilities such as recognition, planning, and prediction. The key challenges in affordance research are: (1) how to automatically discover the distinctive features that specify an affordance in an online and incremental manner and (2) how to generalize these features to novel environments. This survey provides an entry point for interested researchers, including: (1) a general overview; (2) classification and critical analysis of existing work; (3) discussion of how affordances are useful in developmental robotics; (4) some open questions about how to use the affordance concept; and (5) a few promising research directions.

Event Relationship Analysis for Temporal Event Search

There are many news articles about events reported on the Web daily, and people are getting more and more used to reading news articles online to know and understand what events happened. For an event, (which may consist of several component events, i.e., episodes), people are often interested in the whole picture of its evolution and development along a time line. This calls for modeling the dependent relationships between component events. Further, people may also be interested in component events which play important roles in the event evolution or development. To satisfy the user needs in finding and understanding the whole picture of an event effectively and efficiently, we formalize in this paper the problem of temporal event search and propose a framework of event relationship analysis for search events based on user queries. We define three kinds of event relationships which are temporal relationship, content dependence relationship, and event reference relationship for identifying to what an extent a component event is dependent on another component event in the evolution of a target event (i.e., query event). Experiments conducted on a real data set show that our method outperforms a number of baseline methods.

A Study of Reinforcement Learning in a New Multiagent Domain

RoboCup Keepaway is one of the most challenging multiagent systems (MAS) where a team of keepers tries to keep the ball away from the team of takers. Most of current works concentrate on the learning of keeper, not the learning of taker, which is also a great challenge to the application of reinforcement learning (RL). In this paper, we propose a task named takeaway for takers and study the learning of them. We employ an initial learning algorithm called Update on Steps (UoS) for takers and demonstrate that this algorithm has two main faults including action oscillation and reliance on designers experience. Thereafter we present a novel RL algorithm called dynamic CMAC advantage learning (DCMAC-AL). It makes use of advantage(lambda) learning to calculate value function as well as CMAC to generalize state space, and creates novel features based on Bellman error to improve the precision of CMAC. Empirical results show that takers with DCMAC- AL can learn efficiently.

A Unified Framework for Metric Transfer Learning

Transfer learning has been proven to be effective for the problems where training data from a source domain and test data from a target domain are drawn from different distributions. To reduce the distribution divergence between the source domain and the target domain, many previous studies have been focused on designing and optimizing objective functions with the Euclidean distance to measure dissimilarity between instances. However, in some real-world applications, the Euclidean distance may be inappropriate to capture the intrinsic similarity or dissimilarity between instances. To deal with this issue, in this paper, we propose a metric transfer learning framework (MTLF) to encode metric learning in transfer learning. In MTLF, instance weights are learned and exploited to bridge the distributions of different domains, while Mahalanobis distance is learned simultaneously to maximize the intra-class distances and minimize the inter-class distances for the target domain. Unlike previous work where instance weights and Mahalanobis distance are trained in a pipelined framework that potentially leads to error propagation across different components, MTLF attempts to learn instance weights and a Mahalanobis distance in a parallel framework to make knowledge transfer across domains more effective. Furthermore, we develop general solutions to both classification and regression problems on top of MTLF, respectively. We conduct extensive experiments on several real-world datasets on object recognition, handwriting recognition, and WiFi location to verify the effectiveness of MTLF compared with a number of state-of-the-art methods.

Laplacian regularized locality-constrained coding for image classification

Feature coding, which encodes local features extracted from an image with a codebook and generates a set of codes for efficient image representation, has shown very promising results in image classification. Vector quantization is the most simple but widely used method for feature coding. However, it suffers from large quantization errors and leads to dissimilar codes for similar features. To alleviate these problems, we propose Laplacian Regularized Locality-constrained Coding (LapLLC), wherein a locality constraint is used to favor nearby bases for encoding, and Laplacian regularization is integrated to preserve the code consistency of similar features. By incorporating a set of template features, the objective function used by LapLLC can be decomposed, and each feature is encoded by solving a linear system. Additionally, k nearest neighbor technique is employed to construct a much smaller linear system, so that fast approximated coding can be achieved. Therefore, LapLLC provides a novel way for efficient feature coding. Our experiments on a variety of image classification tasks demonstrated the effectiveness of this proposed approach.