Mining Aspect-Specific Opinion using a Holistic Lifelong Topic Model

Abstract

Aspect-level sentiment analysis or opinion mining consists of several core sub-tasks: aspect extraction, opinion identification, polarity classification, and separation of general and aspect-specific opinions. Various topic models have been proposed by researchers to address some of these sub-tasks. However, there is little work on modeling all of them together. In this paper, we first propose a holistic fine-grained topic model, called the JAST (Joint Aspectbased Sentiment Topic) model, that can simultaneously model all of above problems under a unified framework. To further improve it, we incorporate the idea of lifelong machine learning and propose a more advanced model, called the LAST (Lifelong Aspectbased Sentiment Topic) model. LAST automatically mines the prior knowledge of aspect, opinion, and their correspondence from other products or domains. Such knowledge is automatically extracted and incorporated into the proposed LAST model without any human involvement. Our experiments using reviews of a large number of product domains show major improvements of the proposed models over state-of-the-art baselines.

Bias‐Sentiment‐Topic model for microblog sentiment analysis

“…4,7,9,22,25 Turney applied an unsupervised learning algorithm to classify the orientation of users' reviews. Most of these methods take sentiment words or emoticons as sentiment priors to guide sentiment analysis.…”

“…The model supposes there are balls of different colors in an urn, where each color represents each word in the vocabulary corresponding to BST. 9,23 In order to use the GPU model, the correlation matrix δ of words should be estimated in the first place. Then, the color of this ball is recorded and two balls of this color are put back into the urn, which results in the increase in the proportion of balls of this color in the urn.…”

“…Most of these methods take sentiment words or emoticons as sentiment priors to guide sentiment analysis. 4,7,9,22,25 Turney applied an unsupervised learning algorithm to classify the orientation of users' reviews. 25 Hu et al regarded emoticons, sentiment words, textual similarity, and word co-occurrence as four emotional signals, and presented a novel framework to make use of the Emotional Signals for unsupervised Sentiment Analysis (ESSA).…”

“…8 Wang et al built the JAST model to jointly model aspect, opinion, polarity, and generality. 9 Words in the model are divided into three categories: general opinion word, aspect word, and aspect-specific opinion word. JAST can simultaneously achieve aspect extraction, opinion identification, polarity classification, and separation of general and aspect-specific opinions.…”

“…Later, the GPU model containing prior knowledge was utilized in other works. 9,23 In order to use the GPU model, the correlation matrix δ of words should be estimated in the first place. This paper applies the Word2vec toolkit 12,13 to learn word embedding from massive data as the general prior knowledge and then combines it with word co-occurrence information in the test dataset to obtain matrix δ.…”