Esports has grown quickly in recent years, and the business has produced a ton of specifications-based data that is simple to obtain. Because of the aforementioned traits, data mining and deep learning techniques can be used to direct participants and create winning strategies. As one of the world’s most famous e-sports, Dota 2 has grown over time by luring many players from all over the world, and it recently held its tenth professional international tournament. A Dota2 game consists of a drafting and a playing phase. The drafting phase can influence the outcome of the game. However, players frequently struggle to select the strongest lineup to participate. In order to address this gap, we present in this article an NLP and Deep Learning hybrid model as an alternative approach for optimal drafting. We report state-of-the-art results on the problem of predicting the outcome of the match in the hero select phase of the game. Hero vectors are produced by the model using the Continuous Bag of Words (CBOW) in the Word2vec model. The context of a word in a sentence can be predicted by the CBOW model. The model used in this article suggests a predictive tool for players that can provide a more reliable method for the drafting phase. Accordingly, a word is changed into a hero, a phrase into a lineup, and a word vector into a hero vector. The improved Long short-term memory (LSTM) model is used to solve the influence of timing relationships affecting selection on the prediction of winning rate.