Several architectures and approaches have been used to investigate sentiment analysis in recent years. In this communication, The aspect extraction experiment is carried out using different sequence labeling and in combination with the LSTMs. In this research, different pre-trained word embedding techniques used are Glove, Word2vec, FastText. Every individual and combination of different techniques were simulated till we get steady results. Data used in this experiment is Semeval-2014. The added CRF layer over the neural network architecture significantly improved the performance of all models. Finally, an exhaustive analysis is carried out by combining sequence models and different word embedding methods in association with the LSTM architecture to understand the effectiveness of the particular model. In all the models least F1 score is realized using CRF, and LSTM word2vec as 75.45%, and 77.91%, consequently LSTM CRF, BiLSTM CRF with Glove.840B exhibit 84.23%, 85.06% and BilSTM CRF FastText show 85.01%. The experimental findings reveal that a combination of CNN and RNN models with proper word embedding techniques provides optimum results for sentiment analysis.