Sentiment Analysis of Tweets for Estimating Criticality and Security of Events

1. Introduction

Social media websites are growing at a very fast rate throughout the world. The increase in the reach of the Internet to more people every day is also helping this. It is turning out to be one of the major industries in the world and is revolutionising several existing industries such as print media and marketing. Organisations are also benefitting from this advancement as they can now enhance their processes such as understanding customer’s opinions, analysing reviews and feedbacks. However, with the increased availability of people’s data online, there is an increasing concern for people’s privacy and security. The advancement in social media access and increased available features has also started attracting potential hackers. This makes it possible for several crimes to occur such as selling private information, usage of such social sites to do secret transactions that are banned. Due to the massive size of the data available online, sometimes activities and potential threats can also go undetected. This motivated the development of Application Programming Interface that is freely available for people to use to create software that can track these crimes online. But owing to user’s privacy these tools have limits to the amount of data that they can get. There are trackers inside these tools that ban people automatically the people that try to over use it. There are many undergoing projects in many universities that endlessly utilize everything they have to build an efficient product that helps law enforcement agencies to detect crimes in a better way.

In this work we focus on applying sentimental analysis to Twitter data to obtain tweets that can possibly lead to disruptions during an event. To use the Internet and obtain Intelligence, monitoring important sites is necessary. Useful data can be obtained for law enforcement purposes from social networking sites that can be processed for obtaining insights, detecting threats, making predictions and performing many other analyses. Such analyses will provide private organizations, authorities and law enforcement agencies to make decisions and obtain a better understanding of their people. Most analyses, including this analysis use the data that are publicly visible on the web. Users always have the option to choose their data to be publicly or privately visible. The trust and privacy gets breached when an organisation or analysis uses data that is wrongfully obtained. The proposed system is a data collection and text mining application for Intelligence and Law enforcement agencies.

A three step method is proposed for obtaining useful insights and the steps are as follows:

• •

  Obtaining data from Social media sites like Twitter in the form of raw data and tweets by identifying important keywords and locations that are of interest, such as specific public events or places. The data is obtained by crawling periodically and using the APIs provided by these sites such as search and streaming APIs;

• •

  A lexicon based on SentiWordNet was generated to obtain sentiment values for words;

• •

  On the collected real time data in the knowledge base, lexicon based sentimental analysis is done to calculate the emotional sentimental value for every entity. Sentiment values are first calculated for individual words using their values in the lexicon and the sentiment for the entire sentence is obtained by evidence based combination function and sentiment normalisation as described by Jurek et al. (2015).

The tweets that are found to be negative are further analysed to measure their intensity and how they may influence the selected public event. By measuring the number of people who have negative opinion tweets about the event against the total number of people who tweeted about the event, we provide 3 levels of possible threat or disorder for the event that is about to take place. Thus, the relationship between the number of people who tweet with a negative sentiment about an event and the actual disruption or disorder that happens during an event can be obtained. Summarization of the steps is as follows:

• •

  To collect and monitor real time data, using deep data collection by identifying public events or places of interest;

• •

  To perform lexicon based sentimental analysis on the collected data to determine emotional values;

• •

  To classify the obtained emotional values into possible levels of disorder that will happen during the event.