Hi Munawar, 

If you're asking whether features extraction may help your process, the answer is "That depends on your data".

On the other hand, if you are seeking a method for feature extraction, then my advice is to do the following: 

After converting the documents into bag-of-words, you could apply, for example, latent semantic analysis in order to extract the features out of your data. 

HTH.

Samer

There are many tasks in sentiment analysis field, but if you are working in polarity classification and you want a list of linguistic features, here some of them on pages 31- 33:

https://www.cs.uic.edu/~liub/FBS/SentimentAnalysis-and-OpinionMining.pdf

I explored different methods to extract features from text in sentiment analysis in my thesis (see below).

Of course, you can try analytical methods to extract features such as InfoGain (see my thesis). However, if you have many features (many words in BOW) you would get bad classification results using InfoGain+SVM or InfoGain+NaiveBayes since InfoGain extracts only a few features and their impact can be insignificant. You can use sledgehammer as I did in my thesis and simply extract features depending on their frequency (count frequency of words and take a part of the most frequent). I wouldn't throw away stop words since they can convey emotional meaning, for example, according to Pennebaker (see below) and improve classification significantly.

You can extract other features than BOW. For example, as Bahadorreza Ofoghi recommended considering syntax of analyzed texts. You can consider other stylometric, deictic or grammatical features (see the thesis).

http://weka.sourceforge.net/doc.dev/weka/attributeSelection/InfoGainAttributeEval.html

Article The Psychological Functions of Function Words

Thesis Opinion Mining and Lexical Affect Sensing

I forgot to mention. There is probably no necessity to lemmatize words. "Lemmatize or not" depends on the corpus you process and I would skip this step. I verified this in the thesis.

You may also be interested in this paper:

Irony detection in Twitter: The role of affective content DIH Farías, V Patti, P Rosso - ACM Transactions on Internet Technology (TOIT), 2016

http://dl.acm.org/citation.cfm?id=2930663