Sentiment analysis and sarcasm detection are specialized areas in the field of information retrieval and natural language processing. Sentiment classification is closely correlated with sarcasm detection, where people usually adopt sarcasm to highlight their negative feeling. This paper proposes a novel multi-task deep neural networks for joint sarcasm detection and sentiment analysis (MT_SS). MT_SS train both tasks jointly using bidirectional gated recurrent unit with attention network module to obtain task-specific local feature representation while using convolutional neural networks to obtain global feature representation. The experiments on two datasets show that our proposed model outperforms the state-of-the-art approaches.

Preference-based recommendation systems analyze user-item interactions to reveal latent factors that explain our latent preferences for items and form personalized recommendations based on the behavior of others with similar tastes. Most of the works in the recommendation systems literature have been developed under the assumption that user preference is a static pattern, although user preferences and item attributes may be changed through time. To achieve this goal, we develop an Evolutionary Social Poisson Factorization (EPF $$\_$$_ Social) model, a new Bayesian factorization model that can effectively model the smoothly drifting latent factors using Conjugate Gamma–Markov chains. Otherwise, EPF $$\_$$_ Social can obtain the impact of friends on social network for user’ latent preferences. We studied our models with two large real-world datasets, and demonstrated that our model gives better predictive performance than state-of-the-art static factorization models.

Probabilistic topic models have been used to detect topic-based content presentations when facing a collection of documents. However, topic models capture the semantic information according to reasonable simplifying hypotheses, which ignore the worthwhile word-order information. This paper proposes an attention hierarchical topic modeling, which adopts attention mechanism to unify topic embedding and word embedding together into a framework to enhance the clustering effect of hierarchical Dirichlet process. Otherwise, the multi-information integration Chinese restaurant franchise is adopted to construct this model, which further combines timestamp, user, and topic label to optimize topic modeling. Extensive experiments on real-life applications show that our model outperforms several strong baselines on document modeling and classification.

A large number of electronic documents are labeled using human-interpretable annotations. High-efficiency text mining on such data set requires generative model that can flexibly comprehend the significance of observed labels while simultaneously uncovering topics within unlabeled documents. This paper presents a novel and generalized on-line labeled topic model (OLT) tracking the time development of extracted topics through a structured multi-labeled data set. Our topic model has an incrementally updated principle based on time slices in an on-line fashion, and can detect dynamic trending for labeled topics in parallel. Empirical results are presented to demonstrate lower perplexity and high performance of our proposed model when compared with other models.