<p>Emotion recognition plays a prominent role in adaptive healthcare and human-computer interaction. Among various techniques, electroencephalogram (EEG) based emotion recognition emerged as a more reliable alternative to traditional methods such as facial expression or voice tone analysis. This work attempts to study how the history of emotional stimuli shows an impact on a person’s current emotional response, which is relatively less investigated aspect of EEG based emotion recognition. Initially, local standard deviation (LSD) is used for preprocessing EEG data, to reduce the data size while preserving crucial temporal signal fluctuations. The preprocessed EEG data is transformed into a visibility graph (VG), from which five topological properties such as modularity, number of communities, density, average degree, and differential entropy are calculated. The properties extracted are used to perform three binary classification tasks: positive vs. negative, negative vs. neutral, and positive vs. neutral, across all possible sequential combinations of the emotional video trail. The classification results indicated that highest accuracy is achieved when the emotional sequences are new, whereas repetition either in the video source or in the combination of emotional stimuli made the subjects emotionally familiar and consequently led to less distinct emotional responses. This approach used only 12 EEG channels to achieve high accuracy, when compared to earlier studies that generally required 18 or more channels. Thus, the LSDVG-based approach can be seen as a computationally efficient approach for emotion recognition and also helps in understanding the influence of past emotional history on the present emotional response of the brain.</p>