A novel category detection of social media reviews in the restaurant industry

• Published in: Multimedia systems Vol. 29; no. 3; pp. 1825 - 1838
• Main Authors: Khan, Mohib Ullah, Javed, Abdul Rehman, Ihsan, Mansoor, Tariq, Usman
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01-06-2023; Springer Nature B.V
• Subjects: Artificial intelligence; Computer Communication Networks; Computer Graphics; Computer Science; Cryptology; Data Storage Representation; Digital media; Emotions; Multimedia Information Systems; Operating Systems; Restaurants; Role of Deep Learning Models & Analytics in Industrial Multimedia Environment; Social networks; Special Issue Paper; Deep learning; Attention mechanism; CNN; LSTM; Word embedding (Word2Vec); IIoT; Social media analytic; Opinion mining; Text classification
• ISSN: 0942-4962; 1432-1882
• DOI: 10.1007/s00530-020-00704-2

Social media platforms have enabled users to share their thoughts, ideas, and opinions on different subject matters and meanwhile generate lots of information which can be adopted to understand people’s emotion towards certain products. This information can be effectively applied for Aspect Category Detection (ACD). Similarly, people’s emotions and recommendation-based Artificial Intelligence (AI)-powered systems are in trend to assist vendors and other customers to improve their standards. These systems have applications in all sorts of business available on multiple platforms. However, the current conventional approaches fail in providing promising results. Thus, in this paper, we propose novel convolutional attention-based bidirectional modified LSTM by combining the techniques of the next word, next sequence, and pattern prediction with ACD. The proposed approach extracts significant features from public reviews to detect entity and attribute pair, which are treated as a sequence or pattern from a given opinion. Next, we trained our word vectors with the proposed model to strengthen the ACD process. Empirically, we compare the approach with the state-of-the-art ACD models that use SemEval-2015, SemEval-2016, and SentiHood datasets. Results show that the proposed approach effectively achieves 78.96% F1-Score on SemEval-2015, 79.10% F1-Score on SemEval-2016, and 79.03% F1-Score on SentiHood which is higher than the existing approaches.