الفهرس 
Title : A Soft Computing Approach for Sentiment Analysis
Table of ContentsOnly 14 pages are availabe for public view
 |  | from | 151 |  |  |
| | | from | 151 | | |
Abstract
This thesis addresses the question of how the reusability approach could be utilized to support the development of sentiment extraction systems to different types of data (Multimedia data). To find an answer to this question, we highlighted three objectives. Firstly, developing a generic framework that compensates sentiment analysis components of problem-solving methods for all data types (video, audio, text, and image). Secondly, instantiating the suggested framework with a reusable component of tasks and methods for supporting the development of sentiment analysis systems, within the approach of a library of task-specific problem-solving methods. Thirdly, proofing the validity of the suggested library by experimenting with two different sentiment extraction applications configured from the suggested library that represent the basic data types (image, and text). The first application is extracting sentiment from images, while, the other is extracting sentiment from text. The visual sentiment analysis application utilizes both Pulse-Coupled Neural Networks (PCNN) and Neural Networks (NN) for detecting visual positive sentiment analysis. The proposed visual application, which uses the PCNN with the NN classifier achieves 96% right classification, whereas the Viola algorithm achieves 94% for the same dataset One the other hand, the text-based sentiment analysis application is accomplished by suggesting an architecture that can be used to analyze social media text data sentiments based on their clustering. The suggested architecture is composed of three main tasks namely: data cleaning, similarity finding, and randomized clustering Cuckoo search (RCCS). A formula that combines the similarity degree is suggested to improve the accuracy. As well, we utilized the power of the Cuckoo Search with the Levy flight algorithm to cluster the text data. Our architecture is used to detect the optimal or near-optimal number of clusters that best describes a text dataset. To test our model, we used the Niek Sanders tweets dataset. The proposed model achieved better performance compared with the other six algorithms. The six algorithms involved in our comparisons are K-Means, Latent Dirichlet Allocation (LDA), Scalable Multi-stage Clustering (SMSC), and Grouping Like-minded people using Interests Centers GLIC algorithm with its three different variations. According to our experiments, we claim that our model is efficient and very helpful in the sentiment analysis of social media text data.