Sentiment Analysis plays an important role in developing AI applications involving human language and sense. Language-specific Sentiment Lexicon is important to accelerate Sentiment Analysis. Marathi is a morphologically rich but low-resource Indic Language. The language has a very strong grammatical base showing high resemblance to the human body. The paper elaborates on issues like Lexicon basics, word embedding, and Neural Network Models in the process of Lexicon Construction. Existing databases that are useful as seed databases are enlisted. A detailed study of various methods of Lexicon Construction is done. The methods highlight that Sentiment Neural Embedding is required and contextual information plays a vital role in detecting the polarity of a single word or document. The hybrid approach which uses lexicon-based features for deep learning provides better understanding of sentiment analysis. A method is proposed for MSL construction and sentiment analysis of Marathi text. Natural Language Processing Research for Indic Language is its boom. For Marathi, the efforts are seen. Considering the depth and scope of this language, more resource creation is a must for its revolutionization.