چكيده به لاتين
Sentiment analysis aims to extract one’s positive or negative opinion towards a subject, the aggregation of these sentiments over a population represents opinion polling and has numerous applications, including political landscape or financial predictions. Nowadays, by the rapid rise of social media, individuals mostly tend to convey their opinions in the form of videos. Multi-modal sentiments analysis allows to analyze sentiment in multi-modal environments, provided by complementary data streams including acoustic, visual, and textual. While, this complementary data streams consist of extra individual dependent information, including personality, beliefs, and intentions, but the majority of researches on multimodal sentiment analysis neglect this information. In other words, the majority of current works on multi-modal sentiment analysis are general. They treat the user’s generated data equally and do not consider the user’s differences. However, there are cognitive researches that have verified the associations between personality and emotion manifestations. inspired by this we propose a framework that by employing an agglomerative tree, performs personality-based sentiment prediction by an ensemble of base classifiers. The proposed framework improves the multimodal sentiment prediction by a margin of five percent. Furthermore, the user cold start has been addressed by a personality prediction module, in which by employing on well-known multimodal sentiment datasets, annotates them by personality information.
