A collaborative multi-Trends sentiment classification technique is used in the false news detection strategy, enabling sentiment classifiers to be trained on many tweets simultaneously. Even in situations when labeled data is limited, our approach uses sentiment information from several tweets to train more reliable and efficient sentiment classifiers for each Trend by utilizing web opinion data mining techniques. In order to do this, we divided the sentiment classifier for every Trend into two components: one that is specific to that Trend and one that is universal. Our objective is to apply sophisticated data mining techniques to automatically extract the salient qualities of subjects from the vast amount of user evaluations available online. This method, which is based on two observations, gives end users a filtered subset of information and permits early trend classification. We examine and test a set of simple, language-independent criteria based on the social propagation of trends in order to classify these trends. While Greedy & Dynamic Blocking Algorithms customized for each Trend can capture the distinct sentiment expressions inside each Trend, the global model gathers general sentiment knowledge gained from several tweets. Through efficient data mining, we enhance the performance of Trend-specific sentiment classifiers by combining both labeled and unlabeled samples. Additionally, this method incorporates tweet similarities as regularization for Trend-specific sentiment classifiers, making it easier to share sentiment data among similar tweets. We investigate sentiment expression-based and text-based Trends similarity measures. Two efficient techniques are also introduced to solve the model of our approach. Experiments on benchmark datasets show that our approach dramatically improves multi-Trends sentiment classification performance and outperforms baseline methods.