|Appears in Collections:||Computing Science and Mathematics eTheses|
|Title:||Novel symbolic and sub-symbolic approaches for text based and multimodal sentiment analysis|
|Publisher:||University of Stirling|
|Abstract:||In the era of digital media, e-commerce and social networks, websites allow users to share opinions and feedback about products and services. Customers can make informed decisions by reading the experiences of other users. In addition, customer feedback can be used by the organizations to further improve the offered services. However, the quintillion bytes of data generated per day in different languages such as Persian consisting of user feedback cannot be manually read and analyzed by an individual or an organization, for gauging public opinion. Sentiment analysis is an automated process of computationally understanding and classifying subjective information in multi disciplinary ﬁelds such as products, movies, news, public opinion etc. In this thesis, we focus on developing novel methods for Persian text based sentiment analysis. We exploit the developed text-based methods to improve multimodal polarity detection. Speciﬁcally, we develop a novel hybrid framework that integrates dependency-based rules and deep neural networks for detecting polarity in Persian natural language sentences. In addition, we develop a Persian multimodal sentiment analysis framework that integrates audio, visual and textual cues to computationally understand and harvest sentiments from videos posted on social media platforms such as YouTube and Facebook.Speciﬁcally,a ﬁrst of its kind,multimodal Persian sentiment analysis dataset is developed, which is then used evaluate the proposed multimodal framework that exploits the hybrid dependency-based sentiment analysis framework and deep neural network based multimodal fusion. Extensive experimental results have proven the effectiveness of the proposed approaches as compared to state-of-the-art approaches including deep neural networks.|
|Type:||Thesis or Dissertation|
This item is protected by original copyright
Items in the Repository are protected by copyright, with all rights reserved, unless otherwise indicated.
The metadata of the records in the Repository are available under the CC0 public domain dedication: No Rights Reserved https://creativecommons.org/publicdomain/zero/1.0/
If you believe that any material held in STORRE infringes copyright, please contact email@example.com providing details and we will remove the Work from public display in STORRE and investigate your claim.