.png){kind=icon}
.png){kind=icon}
Return
MOCA: Multi-Objective, Collaborative, and Attentive Senitiment Analysis
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2019.2891019.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Document-level sentiment analysis aims to predict the overall ratings of texts (e.g., reviews) written by users for items. Most current works model this problem as a supervised learning task, i.e., classification or regression. Recent studies argue that user preferences and item characteristics also have significant influences on ratings that are modeled by learning user-item-specific text embeddings based on neural networks. However, these studies only use the explicit influence observed in the texts and fail to model the implicit influence that cannot be observed in the texts. To this end, in this paper, we propose a multi-objective, collaborative, and attentive framework called MOCA for document-level sentiment analysis. Our MOCA has three important characteristics: 1) attentive model for explicit influence; MOCA applies a bidirectional recurrent neural network with attention mechanism to learn user-item-specific text embeddings for exploiting the explicit influence the of users and items; 2) collaborative model for implicit influence; MOCA devises a new neural collaborative filtering model based on multilayer perceptron to capture the implicit influence that is implied in the highly personalized interactions between the users and items; and 3) multi-objective optimization; MOCA models this problem as both classification and regression tasks and simultaneously optimizes the two objectives to reinforce one another. The experimental results show that our MOCA significantly outperforms other state-of-the-art techniques on three real-world datasets collected from IMDB and Yelp.
Keywords:
Document-level sentiment analysis
semantic learning
attentive model
neural collaborative filtering
multi-objective optimization
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.