Digital technology has accelerated social networking and brought revolutionary changes. Twitter Facebook and YouTube are the most familiar platforms for social communication, marketing and information distribution. Unhappily, these networks have been invaded by spammers who misuse these online social networking platforms with misinformation, fake news, rumors, malicious links, unsolicited messages, comments etc. Therefore, spam detection in social networks has become a novel framework for distribution of information, sentiments, and news. Our research expresses the experimental study on spam identification from text data. Extensive researches have been done on spam detection field from English texts or other languages. But detection of spam from malicious Bangla text content still needs a lot of attention. In this experimental research, we have used Multinomial Naïve Bayes (MNB) classifier, a supervised machine learning algorithm with feature extraction to detect spam from Bangla text at the sentence level. Our proposed system identifies spam based on the polarity of each sentence correlated with it. Finally, our experiment shows that the model has an accuracy of 82.44% in detecting spam Bangla text content.