International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 3665 Twitter Sentimental Analysis for Predicting Election Result using LSTM Neural Network Dipak Gaikar 1 , Ganesh Sapare 2 , Akanksha Vishwakarma 3 , Apurva Parkar 4 1 Asst. Professor, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Mumbai, Maharashtra 2 B.E. student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Mumbai, Maharashtra 3 B.E. student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Mumbai, Maharashtra 4 B.E. student, Dept. of Computer Engineering, Rajiv Gandhi Institute of Technology, Mumbai, Maharashtra ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - In recent years, social media has emerged as a powerful widespread technology and caused a huge impact on the public debate and communication in the society. More recently, micro-blogging services (e.g., twitter) and social network sites are presumed to have the potential for increasing political participation. Among the most popular social media sites, Twitter serves as an ideal platform for users to share not only information in general but also political opinions publicly through their networks, political institutions have also begun to use this media for the purpose of entering into direct dialogs with citizens and encouraging more political discussions. So we build a model that can analyze these data and extract sentiment that can help us determine the outcome of the election. The process consists methods such as extraction of tweets from twitter using API, data cleaning to get exact data, training the LSTM (Long Short Term Memory) classifier using labelled dataset and testing it to perform sentimental analysis for classification and then representation of result. Further, a comparison is made among the candidates over the type of sentiment by table and bar graph. Key Words: machine learning, twitter, social media, prediction, recurrent neural networks, sentiment analysis, embedding, keras 1. INTRODUCTION Over the last decade with the arrival of social media, the efforts to determine people’s point of view over a particular event or a topic have garnered a wide research interest in natural language processing (NLP) and thus introduced “sentiment analysis.” [1] Many social networking websites and micro blogging websites in today’s world has become the biggest web destinations for people to communicate with each other, to express their perspective about products or movies, share their daily life experience and present their opinion about real time and upcoming events, such as sports or movies, etc. Analysis of public’s sentiment requires a huge data. For achieving a large, diverse dataset of current public opinion or sentiments, Twitter could be used as a valuable resource that lets the users to send and read small text messages called “Tweets” [1]. Basically twitter allows users to post brief and quick real-time updates regarding various activities like sharing, forwarding and replying messages quickly which allows the quick spread of news or information. The wide use of hash tags also makes it easy to search for tweets dealing with a specific subject, thus making it quite a convenient way to gather data. Using sentiment analysis for predicting an election’s result is practically challenging to train a classification model to perform sentiment analysis on tweet streams for a dynamic event such as an election [7]. Implementing this model have certain key challenges such as changes in the topics of conversation and the people about whom social media posts express opinions. [5] For doing this, we first created a LSTM classifier (positive versus negative versus neutral) which is a modified version of RNN (Recurrent Neural Network) for analyzing opinions about different election candidates as expressed in the tweets. We then train our model for each candidate separately. The inspiration for this separation comes from our observation that the same tweet on an issue can be positive for one candidate while negative for another. In fact, a tweet’s sentiment is candidate- dependent. To train the model we used over a 15,000 labelled tweets which are labelled as positive, negative and neutral. In the project. using twitter API, we extracted 40000 real time tweets from Jan 2019 to Mar 2019 relating to names of Indian political parties such as #BJP, #Congress, #TMC, #BSP, #Chowkidar, #BJP4India, # ChowkidarChorHai, etc. The rest of the paper is organized as follows: Section II presents the objectives for implementing the system. Section III provides a related work on sentiment analysis lately. Section IV explains the proposed model. Section V describes the methodology to determine the sentiments associated with different candidates. Section VI is for visualization and results and finally, section VII is based around conclusions. 2. OBJECTIVES  To predict the popularity of a candidate of a political party and therefore extrapolate their chances of