Conference Paper

Aspect and sentiment unification model for online review analysis

DOI: 10.1145/1935826.1935932 Conference: Proceedings of the Forth International Conference on Web Search and Web Data Mining, WSDM 2011, Hong Kong, China, February 9-12, 2011
Source: DBLP

ABSTRACT User-generated reviews on the Web contain sentiments about detailed aspects of products and services. However, most of the reviews are plain text and thus require much effort to obtain information about relevant details. In this paper, we tackle the problem of automatically discovering what aspects are evaluated in reviews and how sentiments for different aspects are expressed. We first propose Sentence-LDA (SLDA), a probabilistic generative model that assumes all words in a single sentence are generated from one aspect. We then extend SLDA to Aspect and Sentiment Unification Model (ASUM), which incorporates aspect and sentiment together to model sentiments toward different aspects. ASUM discovers pairs of {aspect, sentiment} which we call senti-aspects. We applied SLDA and ASUM to reviews of electronic devices and restaurants. The results show that the aspects discovered by SLDA match evaluative details of the reviews, and the senti-aspects found by ASUM capture important aspects that are closely coupled with a sentiment. The results of sentiment classification show that ASUM outperforms other generative models and comes close to supervised classification methods. One important advantage of ASUM is that it does not require any sentiment labels of the reviews, which are often expensive to obtain.

Download full-text

Full-text

Available from: Alice Oh, Oct 31, 2014
3 Followers
 · 
319 Views
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: User-generated content diffusion on social networks has triggered an explosive attention in various disciplines. Within tourism activities, social media has growth in the past years rapidly through regular social network sites, or thematic social network sites such as TripAdvisor. The present study aims to provide a deeper insight into this matter, having as starting point the thought that clients posts good or bad reviews, regarding to different aspects of their experience; and, that a client who has a good experience in restaurant tends to revisit it and recommended it to friends, as opposite if the experience was bad they tell this to friend and recommend not visit. To assess customers' reviews of restaurants, data was gathered on TripAdvisor of Top 10 restaurants in two island context Azores and Hawaii. All the comments were studied carefully and categorized in set of dimensions that measured how the entirety of a meal was perceived: sight, hearing, smell, taste and touch. As the results showed, food is the most decisive variable adopted in the UGC. Additionally, our findings support the notion that the overall quality of the meal reflects a lot more than flavor or taste of the food. To these elements, we need to add visual effect, freshness of the ingredients, and healthiness of the meal, among others as main contents spread on SNS. Thus, results reinforce the literature relative to the social media and ads to the knowledge of the contents created and shared by tourists relative to restaurant experience as a whole.
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Traditional topic models, like LDA and PLSA, have been efficiently extended to capture further aspects of text in addition to the latent topics (e.g., time evolution, sentiment etc.). In this paper, we discuss the issue of joint topicsentiment modeling. We propose a novel topic model for topic-specific sentiment modeling from text and we derive an inference algorithm based on the Gibbs sampling process. We also propose a method for automatically setting the model parameters. The experiments performed on two review datasets show that our model outperforms other stateof-the-art models, in particular for sentiment prediction at the topic level.
    The 30th ACM/SIGAPP Symposium on Applied Computing (SAC’15), Salamanca, Spain; 04/2015
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Most existing topic models focus either on extract-ing static topic-sentiment conjunctions or topic-wise evolution over time leaving out topic-sentiment dynamics and missing the opportunity to provide a more in-depth analysis of textual data. In this paper, we propose an LDA-based topic model for analyzing topic-sentiment evolution over time by modeling time jointly with topics and sentiments. We derive inference algorithm based on Gibbs Sampling process. Finally, we present results on reviews and news datasets showing interpretable trends and strong correlation with ground truth in particular for topic-sentiment evolution over time.
    IEEE International Conference on Data Mining (ICDM’2014), Shenzhen, China; 12/2014