In recent years, the most exploited sources of information such as Facebook, Instagram, LinkedIn and Twitter have been considered to be the main sources of misinformation. The presence of false information in these social networks has a very negative impact on the opinions and the way of thinking of Internet users. To solve this problem of misinformation, several techniques have been used and the most popular is the sentiment analysis. This technique, which consists in exploring opinions on corpora of texts, has become an essential topic in this field. In this article, we propose a new approach, called Conversational Sentiment Analysis Model (CSAM), allowing, from a text written on a subject through messages exchanged between different users, called a conversation, to find the passages describing feelings, emotions, opinions and attitudes. This approach is based on: (i) the conditional probability in order to analyse sentiments of different conversation items in Twitter microblog, which are characterized by small sizes, the presence of emoticons and emojis, (ii) the aggregation of conversation items using the uncertainty theory to evaluate the general sentiment of conversation. We conducted a series of experiments based on the standard Semeval2019 datasets, using three standard and different packages, namely a library for sentiment analysis TextBlob, a dictionary, a sentiment reasoner Flair and an integration-based framework for the Vader NLP task. We evaluated our model with two dataset SemEval 2019 and ScenarioSA, the analysis of the results, which we obtained at the end of this experimental study, confirms the feasibility of our model as well as its performance in terms of precision, recall and F-measurement.

In recent years, the social networks that have become most exploited sources of information, such as Facebook, Instagram, LinkedIn, and Twitter, have been considered the main sources of non-credible information. False information on these social networks has a negative impact on the credibility of conversations. In this article, we propose a new deep learning-based credibility conversation detection approach in social network environments, called CreCDA. CreCDA is based on: (i) the combination of post and user features in order to detect credible and non-credible conversations; (ii) the integration of multi-dense layers to represent features more deeply and to improve the results; (iii) sentiment calculation based on the aggregation of tweets. In order to study the performance of our approach, we have used the standard PHEME dataset. We compared our approach with the main approaches we have studied in the literature. The results of this evaluation show the effectiveness of sentiment analysis and the combination of text and user levels to analyze conversation credibility. We recorded the mean precision of credible and non-credible conversations at 79%, the mean recall at 79%, the mean F1-score at 79%, the mean accuracy at 81%, and the mean G-Mean at 79%.