تحليل المشاعر النصية -الصورية في الشبكات الاجتماعية باستخدام تقنية التعلم العميق

TEXT-IMAGE SENTIMENTANALYSIS IN SOCIAL NETWORKS USING DEEP LEARNING TECHNIQUE

During the recent period, social networks platforms have experienced a notable increase in importance and have garnered substantial attraction from individuals, thus becoming an indispensable part of their daily lives and routines. Rather, they have become a virtual reality through which they publish their activities and interests, and exchange opinions and viewpoints on the various events that occur in the community, which makes them contribute this information to community leadership and decision-making.
From here, the importance of the field of sentiment analysis in social networks emerged, and this field has expanded to include sentiment analysis in visual content, including, for example, images. However, the field of visual sentiment analysis is still relatively new and needs to focus research on it.
The process of analyzing sentiment in social networks is sometimes considered a challenge because the published content may be insufficient or misleading to spread and not convey any feelings through it, such as using attention-grabbing images or including hashtags in the published text. Another challenge is the difficulty in obtaining datasets to conduct research in the field of sentiment analysis in social networks, some social networks may set restrictions and not allow the download and use of their data. In addition, the data collection and classification process manually is a painstaking process that requires a lot of time and an expert team to evaluate the manual classification process.
In this thesis, we present an approach based on sentiment analysis in social networks using texts and images together to predict the overall sentiment of the published content. Two deep neural networks were used, Long short-term memory(LSTM) and Mobilenetv2, one for text classification and the other for image classification, then the results of the outputs of the two networks are combined in a new deep learning network called Dunya Visual Sentiment fusion(DVSF) to produce the predicted output of sentiment classification.
We also introduced an approach termed Persent to Labeling Multi-view Sentiment Analysis (MVSA) Dataset from Twitter with sentiment label and then publish it on Kaggle, The labeled dataset is used in training the text model (LSTM), and EmotionRoi datset from Flickr was used to train the image model(Mobilenetv2).
The findings from the training outcomes and evaluation assessments demonstrated the commendable efficacy of DVSF model in the effective classification of text-image sentiment present in social networks.