Machine Learning and Social Media

Social media like Twitter, Facebook, Instagram and others are rapidly gaining momentum in our daily lives, and it looks like it will only accelerate for younger generations. Therefore, many companies find that their involvement in social media is becoming more and more important.

For the vast majority of people who analyze social media for business purposes, today's work is done either through a simple spreadsheet or by personally being active on the media through answering questions and sending updates.

Using machine learning techniques to analyze social media is still something that is limited to internal use at Google and Facebook - but there are ever more researchers and other large companies experimenting with making it more widespread.

The most common technique is to automatically keep track of which topics are gaining in popularity on social media. It is similar to the "trends" panel on twitter, but where twitter focuses primarily on hashtags, machine learning techniques can automatically extract words and concepts that are gaining popularity.

What people talk about and what is popular in general is not equally interesting for all companies, but especially news agencies get great benefits from these techniques.

In fact, the same techniques are also used to algorithmically invest in equities. If the name of a company is suddenly talked about a lot, then something new and surprising has probably happened. The only thing an investment robot needs to further analyze is whether something positive or negative has happened to that business and then hurry to trade it before others on Wall Street discover it.

Assessing whether talking positively or negatively is another branch of machine learning called "sentiment analysis". In its basic form it just gives a number between -1 and +1, where +1 is very positive and -1 is very negative and 0 is neutral. You can imagine that you get the text for a review of a product, and then you have to guess how many stars the reviewer gives the product. With sentiment analysis, the algorithm provides a qualified guess based on the words and phrases used in the text.

Running sentiment analysis for the comments on the company's facebook wall can, for example help send an email to a responsible person if a very negative comment is detected. That way you can react quickly and handle the situation before a feather turns into 5 chickens as the saying goes.

An extension of sentiment analysis that is currently being researched is "opinion mining". Opinion mining expands not only to assess positively or negatively but also to the extent possible to find out who it is who is feeling a certain way and who or what the feeling is aimed at. An example is “Jonas says the hard drive is making a lot of noise”, which with opinion mining can not only be discovered as a negative attitude but also that it is “Jonas” who feels something and that it is about the “hard drive” and that the negative assessment is that it “makes a lot of noise".

Compared to the situation where you want an email when there are negative comments on social media, opinion mining directly gives the advantage of not being easily confused. Sentiment analysis in its basic form can not distinguish whether a negatively charged comment is negatively charged to the company or it is simply in response to something another user has written.

Another area where opinion mining will be used extensively in the future is the situation where large companies have a world-renowned brand to look after. They would like to use opinion mining to analyze data in many places on the Internet and be notified if their brand is used in unfortunate contexts. For this purpose, it is absolutely necessary to make an indication only if a negative attitude is directly about the brand, so as not to drown in the amount of data.

Another possibility is that IBM, through their Watson services, offers a special service called Personality Insights. It is a service where you can get text written by a user analyzed to get some kind of assessment of the personality profile. Typically, a certain amount of text is needed for it to be accurate, but many companies have forums or very active social media where there are customers who are very engaged and thus over time actually write a lot of text.

What can you then use your clients' personality profiles for? Typically, it is used to prepare more analytical personas for customers. By analyzing a number of customers and subsequently grouping them into groups with the same type of profile, one may discover a customer segment that you had not previously considered. It may also be that you can better focus your work on improving the customer experience for certain groups that may as a whole exhibit frustrations about the company's products.

As I said, it is still a bit of a research stage to use machine learning techniques to help companies understand and respond to what is happening on social media. However, as the information flow just keeps increasing and increasing, it becomes increasingly necessary to be able to automate some of the work as a person responsible for the company's social media.

If you would like to be at the forefront of the development, please contact me for a more detailed analysis of the possibilities for your particular business.