How Different Companies Use Machine Learning
How big companies are applying ML to their products?
Let’s have a look at TripAdvisor. Just imagine how much data they collect and process, how much data there is in general - people interacting with different profiles of the hotels or restaurants, booking the restaurant, etc.
One of the biggest applications that are in TripAdvisor which includes data science or ML is obviously recommendations. You research hotels or restaurants, go to the profile page and then you have recommendations of other kinds of restaurants based on your interactions as well as all the other users’ interactions on the website. This is one of the main areas the personalization team was focusing on.
Another thing is reviews. As you know some of the hotels or restaurants are extremely popular and get hundreds of thousands of reviews. But there was this problem while going to some restaurant page with so many reviews, you could not read it all. You could just read some recent ones, but without fully getting the value from all the accumulated reviews existing on the page. So the idea was to group the reviews in a way that allowed you to choose the topic of the reviews. And this was done in a very smart way, by using the text analysis, natural language processing techniques and finding the ways to make these process at a big scale.
Another example is Zalando where the whole office is dedicated to data science and using data to improve the products.
Data science is one of the hottest topics right now which means that salaries are huge and nevertheless, lots of companies can’t easily hire people yet, the demand is bigger than the offer.
Machine Learning techniques
One of the things companies in different parts of the globe should do is to create value in order to enable them to use Machine Learning techniques for their products and organizations in general, create custom solutions using ML and AI.
Machine Learning techniques can be classified in different ways. There's computer vision, deep learning, text analysis, traditional machine learning, etc. But the most important thing is the domain knowledge. In order to efficiently apply ML to your product or your organization, the first thing you to have to do is to understand the business. After that comes the techniques and technologies, how you actually apply it.
Using Computer Vision for Food Compliance - The product allows to scan the food packaging and then the algorithm extracts the data about it from the database: name of the product, contents (like fat content, carbohydrates), compliance and safety.
Video Super-Resolution - Remember the CSI TV series where the detectives were enhancing videos to see faces on the bad quality pictures? This is happening in real life. There are techniques in deep learning which allow you to basically increase the resolution of the image.
Cell pathology detection on microscope images- Medicine and biology. From a digital picture of the microscope image, the computer vision techniques can identify cells and detect if there is pathology there. So the doctors have knowledge about how to identify if cells are not cancerous, for instance, or if there's some problem in the image. ML can help automate the processing of millions of these images by using the expertise from the doctors and machine algorithm.
Detecting guns in the video stream - And then there's the gun detection obviously. We can all agree that public shootings are a terrible thing. Maybe this is a solution to help avoid it at least to some level. CCTV cameras can not only monitor public places but also detect if there's a gun, sending alert to the security right away.
Car damage detection - It is a simple one. Instead of people examining cars by hand, you can take a picture and the computer will mark damage, which is much simpler.
Getting insights from community forums - Lots of companies have community pages where customers ask questions or complain about the products. This data is really valuable and not many companies are using it to get some insights about their products. So imagine there is a big company which releases a new version of hardware and suddenly lots of people are asking questions about what's wrong with this version or they have some particular problem with this specific hardware. If this community page is really big, it's hard to keep track of it and see what kind of problems this version of the hardware has. The idea here is to: 1. identify products - it's called named entity recognition in the text and then 2. classify this text, what kind of problem it has. After that, you summarize what's happening in general and know what to do with your hardware to make it work.
Managing M&A company profiles - There's a company or bank managing merger and acquisition of the companies. There are lots of data they have and don't really use in a smart way. ML can be used here to optimise and speed up the processes.
What movies to put on In-Flight Entertainment System? - Somebody's putting those movies you all enjoy watching in aeroplanes, but how they're choosing which movies to put there. Surprisingly, it isn't very analytical and there is lots of potential there to use past historical data to optimize and personalize what kind of videos or TV shows you have on planes.
Machine Learning has lots of opportunities for us. Most industries working with large amounts of data have already recognized the value of ML technology. This means that they are able to analyze bigger, more complex data and deliver faster, more accurate results – even on a very large scale. By getting insights from this data, organizations work much more efficiently.
You can watch Giorgi’s full talk here
