Twitter has become a popular social network. This is the potential data source to explore useful information mentioned in the social network by users. Moreover, we could detect an event occurring in real-time since Twitter owns an important characteristic called its real-time nature. This massive amount of raw data from social network can be used for industrial, social management, government policies, economic or business purpose even meteorological analysis. For instance of the disaster situations, Twitter users around the world will post many tweets related to the earthquake; these will be utilized to detect temporal occurrence as well as location information by a humanitarian organization. Because of the requirement of supporting the real-time response in urgency and the subject of the problem are social media data, this thesis proposed a novel approach to detect the temporal earthquake. Firstly, Convolution Neural Network (CNN) based method is used for filtering irrelevant to determine informative tweet and then these informative data will be processed by Concurrent Neural Network (RNN) underlying time series data for temporal event detection. Our system with the aid of CNN module which takes more advantages since it does not require any feature extraction engineering and perform better than current methods. The time earthquake detection is the charge of RNN model with Long Short Term Model (LSTM) architecture which is an excellent candidate for time series. We prove that earthquake can be detected after it happens in the level of tolerance time delay.