Jiguang Liang

CONR: A Novel Method for Sentiment Word Identification

Sentiment word identification (SWI) is of high relevance to sentiment analysis technologies and applications. Currently most SWI methods heavily rely on sentiment seed words that have limited sentiment information. Even though there emerge non-seed approaches based on sentiment labels of documents, but in which the context information has not been fully considered. In this paper, based on matrix factorization with co-occurrence neighbor regularization which is derived from context, we propose a novel non-seed model called CONR for SWI. Instead of seed words, CONR exploits two important factors: sentiment matching and sentiment consistency for sentiment word identification. Experimental results on four publicly available datasets show that CONR can outperform the state of-the-art methods.

Sentiment Classification Based on AS-LDA Model

Abstract We address the task of sentiment classification - identification of the polarity of the subjective document in this paper. We introduces a sentiment classification method called AS LDA. In this model, we assume that words in subjective documents consists of two parts: sentiment element words and auxiliary words which are sampled accordingly from sentiment topics and auxiliary topics. Sentiment element words include targets of the opinions, polarity words and modifiers of polarity words. Experimental results demonstrate that our approach outperforms Latent Dirichlet Allocation (LDA).

Feature Selection for Sentiment Classification Using Matrix Factorization

Feature selection is a critical task in both sentiment classification and topical text classification. However, most existing feature selection algorithms ignore a significant contextual difference between them that sentiment classification is commonly depended more on the words conveying sentiments. Based on this observation, a new feature selection method based on matrix factorization is proposed to identify the words with strong inter-sentiment distinguish-ability and intra-sentiment similarity. Furthermore, experiments show that our models require less features while still maintaining reasonable classification accuracy.

Text Document Latent Subspace Clustering by PLSA Factors

Text documents are often high dimensional and sparse, it is a great challenge to discover the clusters among the unlabelled text data, because there are no obvious clusters by common distance measure. In this paper we present a latent subspace clustering method to find text clusters. In our algorithm, we use latent factors extracted by probability latent semantic analysis (PLSA) to generate latent clustering subspaces, and then use the distance between sample and each latent clustering subspace as similarity for text clustering. On some text document datasets our method shows effective implementation for text clustering.

Dependency Expansion Model for Sentiment Lexicon Extraction

In this paper, we present a sentiment lexicon building method called dependency expansion method (DEM), which exploits the relations described in dependency trees between sentiment words and degree adverbs. By taking advantage of the observation that degree adverbs modify sentiment words, two extraction rules are made, through which sentiment words and degree adverbs can be effectively expanded. We evaluate performance on two product reviews corpora in Chinese. Experimental results show that our DEM can effectively extract sentiment words to support the review opinion analysis.

Leveraging Latent Sentiment Constraint in Probabilistic Matrix Factorization for Cross-domain Sentiment Classification

Sentiment analysis is concerned with classifying a subjective text into positive or negative according to the opinion expressed in it. The performance of traditional sentiment classification algorithms rely heavily on manually labeled training data. However, not every domain has the labeled data because the labeling work is time-consuming and expensive. In this paper, we propose a latent sentiment factorization (LSF) algorithm based on probabilistic matrix factorization technique for cross-domain sentiment classification. LSF works in the setting where there are only labeled data in the source domain and unlabeled data in the target domain. It bridges the gap between domains by exploiting the sentiment correlations between domain-shared and domain-specific words in a two-dimensional sentiment space. Experimental results demonstrate the superiority of our method over the state-of-the-art approaches.

RTPMF: Leveraging User and Message Embeddings for Retweeting Behavior Prediction☆

Abstract Understanding retweeting mechanism and predicting retweeting behavior is an important and valuable task in user behavior analysis. In this paper, aiming at providing a general method for improving retweeting behavior prediction performance, we propose a probabilistic matrix factorization model (RTPMF) incorporating user social network information and message semantic relationship. The contributions of this paper are three-fold: (1) We convert predicting user retweeting behavior problem to solve a probabilistic matrix factorization problem; (2) Following the intuition that user social network relationship will affect the retweeting behavior, we extensively study how to model social information to improve the prediction performance; and (3) We also incorporate message semantic embedding to constrain the objective function by making a full use of additional the messages’ content-based and structure-based features. The empirical results and analysis demonstrate that our method significantly outperform the state-of-the-art approaches.

An EMM-based Approach for Text Classification

Abstract In this paper, a classification method named explicit Markov model is applied for text classification. Currently some machine learning technologies, such as support vector machine (SVM), have been discussed widely in text classification. However, these methods consider that any two features are independent and ignore the language structure information. Hidden Markov model is a powerful tool for sequence tagging problems. This paper presents a new method called explicit Markov model (EMM) which is based on HMM for text classification. EMM make better use of the context information between the observation symbols. Our experiments are conducted on three datasets: Reuters 21578 R8 dataset, WebKB and Fudan University Chinese text classification corpus. Experimental results show that the performance of EMM is comparable to SVM for text classification.

Optimization-based model for determining words' sentiment orientations

Sentiment word identification (SWI) is a basic task of sentiment analysis. Traditional techniques become unqualified because they need seed sentiment words which may lead to low robustness. This paper presents an optimization-based framework by incorporating sentiment contextual information instead of seed words. Specifically, we exploit two sentiment phenomena: (1) sentiment matching: polarities of the document and its most component sentiment words are the same, and (2) sentiment consistency: polarities of two frequently co-occurring words are the same. Empirical results demonstrate that our models significantly outperform the existing approaches.

Sentiment Word Identification with Sentiment Contextual Factors

Sentiment word identification (SWI) refers to the task of automatically identifying whether a given word expresses positive or negative opinion. SWI is a critical component of sentiment analysis technologies. Traditional sentiment word identification techniques become unqualified because they need seed sentiment words which leads to low robustness. In this paper, we consider SWI as a matrix factorization problem and propose three models for it. Instead of seed words, we exploit sentiment matching and sentiment consistency for modeling. Extensive experimental studies on three real-world datasets demonstrate that our models outperform the state-of-the-art approaches.