Sentiment analysis in conversations has gained increasing attention for the growing number of applications like human-robot interactions. Inaccurate emotion identification in existing sentiment analysis methods is due to a lack of concentration on explicit and implicit factors in sentiment detection. Hence a novel deep attention expression analysis technique has been introduced in which a deep attention network with parallel level fuzzy classifier identifies the nature of the words using sequential word N-gram functions by incorporating distributed semantics thereby the implicit and explicit nature of the sentence is identified and classified. Moreover, negations and tone in the sentence create a perplexity nature of sentiment analysis.To solve this problem,linguistic feature-based classification has been presented that utilizesa POS-based tagging in the attention layer and BOW to provide word embedding. Then, lemmatization and stemming process of words the root words are identified by maximum likeliness probability, resulting in the identification of new words with linguistic features. Furthermore, Naïve Bayes classifier and ensemble clustering with lambda function havebeen used to identify the negations and tone of the sentence. Thus, the results provided accurate detection of the positive, negative, or neutral sentiment of the sentence with a high accuracy of 96% and a precision of 97%.