Mixing social media and AI to help humanity

Introduction

Social media is huge! A huge percentage of people use some form of social media, and the number is growing by the day. Intelligent algorithms are all around us too and are being used by a lot of internet companies to personalize and model our liking and habits to provide us with a more curated social experience. That has been the face of huge criticism in recent years, more on that later. In this write-up, I am focusing on how researchers across the world are leveraging the power of social media and AI to provide services to humans like never before.

What is social media NLP?

Social media NLP is a huge domain of research in itself. In this domain, researchers focus on mining data (mostly textual, sometimes multimedia data) that is publicly available (in the wild), and then analyzing the contents of the data, to provide some form of services. These services can be anywhere between smart city management, social media monitoring for malicious behavior, and even crisis management.

Social media NLP for smart city traffic management

Urban traffic management is a huge problem in modern-day cities and is taking a toll on the economy and the environment, along with the mental health of the citizens. A big part of everyday tweets generated is about how bad the traffic is. These tweets range from a simple mention of the bad traffic scenario and can even go as far as dissenting against the transport authorities and local governments. Along with that, there are local police departments that are very active on Twitter. Local traffic PDs are constantly updating the current traffic scenario in the area via their tweets to keep their citizens updated. These tweets can be scrapped using Twitter APIs and cleaning and used to provide traffic management services.

At the core of it, the problem is to analyze the tweets and extract important heuristics to understand the traffic scenario of an area. There can be lots of ways to look into the technicality of this problem. One common way of dealing with the problem from a classification perspective is to train a classification model to classify the tweets into one of the pre-defined classes, where each class represents a traffic scenario. The classification simply indicates what type of traffic congestion the tweet is talking about. Researchers have also used neural networks, and deep learning techniques to perform the classification with a good deal of success. From a sentiment analysis perspective, there are also very useful works, where the researchers have analyzed the sentiment of the tweets, and then used the sentiment information to extract interesting heuristics about the current traffic scenarios.

These heuristics collected from the Twitter data can be fed into smart algorithms which can be used to re-route or manage traffic flows at congested areas of the city. Modern research in these domains has shown that the time spent by the vehicles in congestion can be significantly reduced using twitter-NLP backed smart algorithms.

Social media NLP for social monitoring.

There have been lots of accusations and discussions in recent years about how social media is causing division in society and causing people with varied ideas and different communities to hate each other. There is a dire need of the hour to monitor such hateful comments and sentiments from speredding in social media. Social media NLP can come in handy here too!

Hate speech detection is a very useful domain of research that is useful to monitor social media for hateful content. hateful contents are very dangerous in a social media setting, and these models are useful to track them down and delete them. This can be represented as a classic case of sentiment analysis, where the 3 levels of sentiments are hateful, neutral, and non-hateful. Researchers have been very successful in identifying these hateful comments among social media texts.

However, there is more to this problem than just sentiment analysis. Take, for example, the tweets generated in India. A huge percentage of the population writes their tweets in their local language using the English alphabets. Some of the hateful sentiments are hidden in the form of sarcasm or past references that can be hard to detect by machine learning models. Researchers are implementing various techniques to detect these hidden hate-comments. One approach is to build a dictionary of the most common abuses in a given Indian language, written using the English alphabets. Datasets of hate-speech are also being built in different languages to support the cause. Hate Speech masquerading as sarcasm remains a huge area to explore.

Social media NLP for disaster management

Social media text analysis can help us identify disasters in an area, faster than traditional news and media. “Social media sensors”, as they are called, can be used to monitor and identify tweets that inform us about disasters. These are usually very helpful in cases where the disaster can not be predicted, such as flash floods and earthquakes. This research is not necessarily be restricted to just text data and can also be extended to use text, audio, and multimedia data also. Another challenge is in this area is to be able to predict these disasters with a very low amount of data. This is to ensure, that those areas which have very limited social media usage, can also be served with these technologies.

Conclusion

Social media is probably going to be around for a long time, and with the research that is being produced, I am hopeful, that we would be able to provide a lot of social services using social media. I would encourage researchers to contribute more to this domain, so that a lot of social problems that we face today, might be eradicated in the coming days.

Originally published at https://www.linkedin.com.