From our smartphones to cars, to regular customer service interactions, we use machine learning every day.
And machine learning’s capabilities, computational power, and use cases will only continue to grow by the day. Some you may be aware of and some may be completely new. So, just what are the top applications of machine learning? And how is machine learning used in everyday life?
Text analysis is an example of machine learning that uses Natural Language Processing (NLP) to allow machines to “understand” human language. NLP combines the study of linguistics and data science with powerful machine learning algorithms to break down language and create programs that can understand, analyze, and extract meaning from text.
Text analysis machine learning examples include:
Sentiment analysis is an ML application and form of text analysis commonly used for “social listening” or social media sentiment analysis to follow what customers are saying about brands and products online in real time.
Sentiment analysis or opinion mining is the process of analyzing text for “opinion polarity” (positive, neutral, negative, etc.) to automatically read it for the feeling and emotion of the writer.
SaaS machine learning platforms, like MonkeyLearn, provide ready-to-use sentiment analysis tools. Try out this pre-trained sentiment model on your own text to see how it works.
If you need a customized sentiment analysis model, which can be trained to the specific language and criteria of your business sign up to MonkeyLearn and follow this quick tutorial to create your own sentiment analysis model in a simple point and click interface
Text classification is a form of text analysis that includes sentiment analysis and other techniques, like topic analysis and intent detection. It’s the process of assigning “tags” or categories to text according to its content – from documents, customer service data, surveys, online reviews, and more. Models can be trained for individual business uses to automatically organize and analyze text for powerful insights.
Topic analysis put to work on Net Promoter Score (NPS) surveys, for example, can organize them by topics, like Customer Support, Ease of Use, Features, and Pricing. Once you have them organized by topic, you can perform sentiment analysis on each topic, to find out which aspects (topics) of your business are strongest and which may need some work.
This is a process called aspect-based sentiment analysis. Below is an example of the result of aspect-based sentiment analysis run on a customer comment:
I like the new update, but it seems really slow. I can’t get tech support on the phone.
Text classification allows you to run multiple analyses together to get super fine-grained results or help automate and streamline processes. Try out this pre-trained email response classifier that automatically sorts emails by topics: Interested, Not Interested, Unsubscribe, and Wrong Person, Autoresponder, or Email Bounce to save time when wading through thousands of emails.
Yet another great use of machine learning is keyword extraction (AKA keyword detection or keyword analysis), which automatically pulls the most used and most important words and phrases from a text. Keyword extraction helps summarize the content of documents, web pages, survey data, and more; find out the main topics discussed; even uncover emerging trends in the market.
Below is an example of keyword extraction used on a comment from an angry customer.
Notice how machine learning automatically joins words that go together into phrases “order number,” “slow deliver,” and “poor customer support.” Keyword extraction can be used on customer service data from internal CRM systems, surveys, Twitter, and more, to break through big data and uncover actionable insights.
Chatbots are a common example of machine learning in business that uses similar ML techniques to the above. Chatbots are computer programs – usually on a website or smartphone app – that simulate human conversation by automatically responding to questions with learned information.
Chatbots use NLP and NLU (Natural Language Understanding), like topic analysis and intent detection, to analyze, categorize, and “understand” queries, so that they can route them to the correct employee or department, respond to customer queries, even make product recommendations. Chatbots have been shown to reduce customer service costs by up to 30% by automatically solving 80% of routine customer support tickets.
Chatbot machine learning example: Reply.io
Reply uses machine learning to automate customer service support at scale, so you can integrate all of your support interactions and teams, wherever you need.
Implementing tools like chatbots into customer service channels will become more and more important for businesses, as customers have come to expect omnichannel support from the brands they use.
With the help of deep learning and convolutional neural networks that chain together multiple machine learning algorithms – to work much like the human brain does – image recognition, face detection, and image and video analysis have come a long way in recent years.
Image recognition is used by municipal governments to help keep traffic under control or by utility companies to help locate power outages. Facial detection (or facial recognition) is used by law enforcement to identify suspects or by companies to recognize employees and allow them passage into buildings.
Facial recognition machine learning example: Facebook automatic tagging
Probably the most common facial recognition system we’re all familiar with is on Facebook, where the software automatically recognizes and tags friends and family in our photos.
Self-driving cars use “computer vision,” a form of image and video analysis, but quite a bit more advanced – aiming to mimic actual human vision by analyzing and responding to data in real time.
Because the machine learning technology of driverless cars must react in fractions of a second, it takes many many millions of images, hours of video, and miles of road to train them. Google’s self-driving project Waymo has already launched its “robotaxi” service in the suburbs of Phoenix, where customers are able to hail a ride with no backup driver.
Self-driving cars machine learning example: Tesla Autopilot
Although not a complete driverless experience, Tesla Autopilot is similar to machine learning systems used in other new cars that are constantly watching for driver error and encouraging more cautious driving.
Speech recognition software uses NLP to convert natural spoken human language into a format that machines can process.
Speech recognition is an application of machine learning in the real world that we’re all familiar with – it’s what runs virtual personal assistants and smart assistants that we use every day, like Alexa, Siri, and Google Assistant. Machine learning technology allows them to recognize distinctive voices and continue to learn to the needs of individual customers.
Speech recognition machine learning example: Amazon Alexa
Alexa has become the established favorite for successful use of speech recognition machine learning. And with Alexa Skill Kit, programmers can build their own specialized uses for Alexa.
Machine learning is becoming a helpful tool for a number of financial applications and is finding a lot of success in the fight against fraud. Machine learning programs are used to scan huge sets of historical financial data to detect anomalies and automatically block transactions or notify the appropriate authority.
Internal business controls allow payment systems to recognize only authorized administrators and hold payments back until the proper approval has been initiated. Real-time learning and analysis with constant database updates can catch fraud right at the point-of-sale or right as a withdrawal request occurs, helping save businesses millions of dollars.
Fraud detection with machine learning example: Danske Bank
As both the number of frauds and their sophistication increases, Danske Bank realized their previous systems weren’t working: they were only detecting fraud 40% of the time with up to 1,200 false positives a day.
With the implementation of machine learning systems with deep learning, they were able to increase true positives by 50%, allowing them to focus on actual fraud cases, and they now expect to eventually decrease false positives by up to 80%
Using text analysis tools and predictive analysis, companies are able to read through thousands of resumes in just minutes to pull possible employee matches and recognize previously unused talents in existing employees.
HR management and acquisition with machine learning use case: Eightfold AI
Developers of the Talent Intelligence Platform, are one of the companies at the forefront of using AI and machine learning for Human Resources management, talent acquisition, employee development, and diversity hiring applications.
Eightfold AI promises to “unlock the hidden capabilities of your employees” and better match prospective employees with open positions, by fitting work requirements with relevant skills and removing inherent human biases that lead to a lack of diversity in hiring practices.
OCR (optical character recognition or optical character reader) is the machine learning-aided recognition and conversion of typed, printed, or handwritten text into machine-encoded text. It uses a combination of hardware and software that scan written text onto a screen or into a database, etc. for further data analysis.
It can be used to convert written text and drawings into tables and graphs, in conjunction with machine translation to translate books in real time, or to scan physical checks and deposit them with banking apps.
OCR with machine learning example: Dropbox
Dropbox says that 10-20% of the data stored by their users are photos of documents. Previously users would have to manually read these documents for the information they need. However, with the help of OCR, these PDFs, handwritten documents, etc., are machine comprehensible and can be searched like a database.
Machine learning language translation or, simply, machine translation (MT) has made huge strides in recent years. Advancements in artificial neural networks and the ability to store and process huge amounts of data allow machine translation to learn and access learned data much more quickly.
These advancements allow businesses to translate important documents in a trusted, cost-effective manner and give international travelers deeper access to previously confusing menu items.
Example of machine learning language translation: Google Translate
The Google Translate app uses NLP, OCR, and speech recognition to allow the translation of printed or spoken language, and Google Chrome language detection and translation can be used as an add-on to automatically translate web pages and emails.
With the collection and constant analysis of user behavior, demographic, and product attribute data, online product sales platforms use machine learning to recommend products that users may be interested in – in addition to or in place of – other products they have purchased or viewed.
Among other techniques, they do this with the use of: Explicit ratings – items that users have rated highly (5 stars, for example) on the platform Implicit ratings – items that users have searched frequently or shown interest in by clicking on product photos, descriptions, etc. Product similarity – displaying products similar to products a user has already purchased or shown interest in
Example of machine learning product recommendations: Amazon
Machine learning applications in healthcare are growing rapidly and are proving to be some of the most productive and practical use cases of ML. Machine learning programs can scan a patient’s healthcare records and compare them to thousands of other records in search of similarities that may indicate that the patient is prone to certain diseases.
Example of machine learning medical diagnosis: UC Irvine Machine Learning Repository
Using a huge repository of digitized images of breast masses, UCI researchers were able to advance and automate breast cancer diagnoses.
Promoter.io uses NPS surveys as a metric to gauge customer loyalty. Previously these surveys would have to be annotated, calculated, and analyzed manually, which would take hundreds of employee hours and often produce less than accurate results.
However, with the help of MonkeyLearn’s keyword analysis and sentiment analysis tools, they were able to train custom models to their specific needs, that performed at their exacting criteria. This allowed them to extract data from their surveys that could have been left behind with other analyses to get the results they needed.
When hosting and displaying millions of restaurant photos – relating to different aspects of a restaurant – one can imagine it would be nearly impossible to classify them manually. Furthermore, in the case of Yelp, these photos need to be easily categorized, so they’re searchable by users and automatically placed into the relevant section of their site.
Yelp used deep learning to create a customer image classifier, so that uploaded images were automatically placed in the appropriate category.
Some of the ML model training could be done with existing data and internal systems. For example, images of menus tend to have captions with the word “menu.” However, many thousands of others were first manually tagged by experts, in order to train the model.
Uber’s dynamic pricing algorithms determine where more drivers may be needed, due to inclement weather, rush hour, large events, etc. Uber’s “Geosurge” program uses machine learning regression analysis, for example, to predict which areas of a city will be busier at any given time – and encourage more drivers to move to that area, with the incentive of surge pricing rates. Although customers end up paying more, this benefits them, as well, because there otherwise may not have been a ride available at all.
Machine learning is here to stay, of course, and its use cases, efficiency, and power will only continue to grow. As commonplace as some of the above examples are, many of them were unimaginable as recently as just a decade or so ago.
As machine learning models learn from more and more data, their accuracy and speed of learning will accelerate at rates previously unseen.
In the near future, deep learning and artificial neural networks will be able to create machine learning models that will be born with the collected knowledge of their ancestors, as it were, and will be able to learn new tasks much more quickly and with much less training data.
SaaS machine learning platforms allow you to harness the power of machine learning, without the need for a computer engineering or data science background.
MonkeyLearn is a SaaS text analysis platform that uses powerful machine learning techniques to analyze all manner of text for real-world, data-driven insights.
Perform social listening to find out what customers are saying about your brand on Twitter, Facebook, Instagram, or all over the web. Put aspect-based sentiment analysis to work on your CRM data or surveys to automatically categorize feedback by topic and find out where it’s positive and where it’s negative.
Sign up for a free MonkeyLearn account to see how it works or schedule a demo and we’ll walk you through it.
December 4th, 2020