Since the inception of computers, scientists have had a strong desire to make machines understand the human language. Communicating like a human is a big ask for a robot. We speak in colloquialisms and abbreviations. The same word can take on several different meanings based on subtle nuances in tone.
The pursuit of making the computer understand us is rife with challenges. Even in the context of a single language, there's quite a bit of variation. While you could potentially feed a computer a digital dictionary loaded with every word known to humankind, you can't necessarily account for every possible combination.
This led to an interdisciplinary solution that combined linguistics and computer science: Natural Language Processing (NLP). NLP has been around for a while, but as of late, has benefited from recent developments in Machine Learning and Deep Learning techniques. Machine Learning (ML) is a subfield within Artificial Intelligence that builds algorithms that enable computers to learn to perform tasks from data instead of being explicitly programmed.
There are plenty of applications that we know and use on a daily basis that are integrating recent advancements in ML, and NLP techniques into their product including Amazon's Alexa, Google's Search Engine and Microsoft Word's spell check. Other applications serve to teach machines how to perform complex natural language-related such as machine translation, sentiment analysis, text classification and text automatic completion.
The post aims to give the reader a gentle overview of NLP, ML and Deep Learning. It will provide some initial concepts to invite the reader to continue investigating and make the connection of how it can be applied in the context of customer experience and support.
Introduction to NLP, Machine Learning and Deep Learning:
Natural Language Processing (NLP):
Natural Language Processing is an area that sits at the intersection between Artificial Intelligence and Linguistics. It involves an intelligent analysis of unstructured data in the form of written language. Human communication is frustratingly vague at times; there are few rules, we all use colloquialisms, abbreviations, and don't often bother to correct misspellings. These inconsistencies make a computer's analysis of natural language difficult at best. If you have lots of data in the form of customer feedback and you want to get business insights from it automatically, then you would have to use NLP techniques.
NLP can be split into two different sets of approaches. The first approach is a rule-based approach and involves human-crafted or curated rule sets. Rules-based approaches look for linguistic terms such as love-hate-like and dislike. The presence of positive and negative words defines whether a sentence is positive or negative. The second approach uses statistical techniques using machine learning for developing algorithmic models. With machine learning, data scientists, for example, can train an algorithm to understand sentiment based on large data sets, fine-tuning the model to predict sentiment in entirely new sentences.
Machine learning is a subset of artificial intelligence and is an inclusive term that contains a considerable number of approaches from the simple to the very complex. Machine Learning is a discipline responsible for giving computers the ability to learn without being explicitly programmed. Rather than a human telling the computer to follow a limited set of pre-defined rules, we ask it to look through the data, learn to perform specific tasks and get better at it over time.
In traditional software engineering, the developers give precise (sequential) instructions to machine (CPUs) how to execute a program/piece of code. For example, when you click the 'Submit' button after filling your details on a web form, all the program does is validating that the data you've filled is in the correct format, grouping it and sending it to a server to get a response from a piece of code running on that machine.
Different ML Paradigms
It is called supervised learning because it involves the process of algorithm learning from a training dataset and can be thought of as a teacher supervising the learning process. We know the correct answers, the algorithm iteratively makes predictions on the training data and is corrected by the teacher. Learning stops when the algorithm achieves an acceptable level of performance. Once the algorithm is trained it is then possible to make predictions about new observations.
In Unsupervised Learning, the algorithm only has input data but no corresponding output variables. The goal for unsupervised learning is to model the underlying structure or distribution in the data to learn more about the data. Unsupervised models are therefore mainly concerned with detecting patterns in the training data. They are called unsupervised because there is no teacher and pre-defined correct answers. The algorithms are left to their own devices to group observations into clusters or fitting a probability distribution over observations to detect improbable outliers.
Reinforcement Learning: The algorithm (also called the agent) interacts with the environment. That is, it observes the state of the environment, performs an action and then receives a reward/penalty. It starts off taking random actions, but over time figures out the optimal strategy. One of the most common examples of reinforcement learning is a machine learning to a play a computer game or in the case of Google's Deepmind, an AI that has managed to learn how to walk, run, jump, and climb without any prior guidance.
Deep learning is a subset of machine learning that uses neural network architectures inspired by the biological neural networks of the human brain. Neural networks are a specific set of algorithms that have revolutionized machine learning and are used in all ML paradigms.
They teach computers to do what comes naturally to humans: learn by example. It's the technology behind driverless cars, enabling them to spot the difference between a pedestrian and a lamppost. Deep learning requires large amounts of data (which is no short supply considering we generate an estimated 2.6 quintillion bytes daily) and has come on leaps and bounds in recent years due to advancements in computing power, reducing the time it takes to train a deep learning neural networks. Deep learning has now reached the point where it can outperform humans in a range of tasks like classifying images and it will continue to improve over time.
Combination of NLP, Machine Learning and Deep Learning:
As you can see from the Venn diagram above you can see how NLP is related to ML and deep learning neural networks. Deep learning is one of the techniques used in machine learning alongside other techniques such as regression and K means clustering. Machine learning is often used for NLP tasks that use techniques such as deep learning neural networks. Below you'll see a table outlining how different deep learning algorithms are applied to NLP.
What Machine Learning NLP Advances Mean for CX Professionals
By capitalising on the advancements in NLP & ML, companies gain access to a whole new world of possibilities. Whether it's through automatically reading and classifying new support tickets, getting ahead of bad press, or discovering new business insights from unstructured data Machine Learning stands to be a powerful tool the customer experience professional can integrate into their workflow.
Following are the four waysto know how NLP, ML and Deep Learning are transforming CX:
1. Text Analytics for Business Insight: For any analytics platform, the objective is to allow the insights from data analysis to be as clear as possible. Advanced text analytics platforms should enable various sources of unstructured data to be viewed and measured using relevant scores to visualise sentiment and topic analysis across every part of the customer journey in real time.
Chattermill easily integrates with existing platforms used to collect and manage various aspects of customer experiences, such as CRM, Customer Support, App Store Reviews, Online Surveys and CEM systems to provide a combined view of all customer feedback across the entire customer journey.
Advanced text analytic platforms leverage deep learning neural networks to identify patterns that signify positive or negative sentiment in vast volumes of unstructured data. Here, machine learning and deep neural networks process unstructured text and classify it automatically in real time. The technology works to highlight fundamental topics affecting customer loyalty, such as product attributes, online experiences and customer support. Customer experience teams can interpret the bespoke insights to inform business decisions and prioritise resources to areas that can have the biggest impact on customer experience.
For example, NPS surveys allow customers to leave open ended feedback and often mention multiple categories that touch upon different business functions. Hidden within the feedback is customer sentiment about exactly how they feel and most often, customers talk about things that matter most to them in either a positive or negative manner.
Identifying themes and tagging feedback by sentiment is near impossible at scale when humans do the work. A single agent can only process 1,000 pieces of feedback a day consistently. For this reason, forward-thinking customer experience teams are embracing ML-powered text analytic tools built to handle massive unstructured data sets. Adding neural network algorithms that can pull accurate business insights at scale with human-level accuracy into their strategic planning.
2. Improve Support Experience:
As your organisation grows, getting a deeper understanding of the issues that impact your customer is critical to a smooth scaling process and routing customers to the best person is vital. With hundreds of thousands of tickets surfacing daily across multiple countries for some support teams, customer support must ensure that agents are empowered to resolve them as accurately and quickly as possible. When an agent opens a ticket, the first thing they need to do is determine the issue type out of thousands of possibilities.
Imagine you're a customer support agent you've got multiple customers writing to you and phoning you simultaneously. Your prerogative is to deliver answers they need quickly and accurately while providing a fantastic customer experience. Now layer on the need to track each inbound help request with contextual 'topic tags' for each ticket. Different agents monitor different cases, and finding the right pairing and routing to correct personnel can make or break the customer experience.
As your company scales, this model of customer support is broken because it asks too much of your agents. The goal of customer support teams is to facilitate the best end-to-end experience possible for users. Customer support teams can benefit from leveraging tools use machine learning and natural language processing (NLP) techniques. ML can use historical customer service data with keyword analysis to route incoming support tickets to the appropriate person.
For example, this technology can analyse tickets to learn why a customer needs support in the first place. From there, messages (tickets, social media message, even voicemails) are sent to the appropriate person. This means customers can bypass the frustrations associated with transferring calls to rep after rep. By making the life of the agent more simple, you help them focus on what truly matters: delivering amazing, personalised customer experiences.
The AI would automatically identify, tag, and route incoming tickets on your customer support platform based on their corresponding category or priority. This removes the need to manually evaluate each inquiry and make a call on who is the most appropriate contact. Helping customer support representatives improve their speed and accuracy, resulting in a better customer experience.
So far at Chattermill when working with clients we have seen some promising results! Ticket resolution times have reduced while delivering service with similar or higher levels of customer satisfaction. By empowering customer support agents to achieve quicker and more accurate solutions, Chattermills deep learning algorithms make support experience more enjoyable for customers.
3. Workforce Analytics & Voice of the Employee:
Within the workplace, ML stands to play a critical role. Listening to the Voice of the Employee by systematically collecting, managing and acting on the employee feedback on a variety of valuable topics is essential. We've long been relying on gut feelings and corporate ideologies to make hiring decisions, but more and more, organisations are turning toward data to inform who is promoted and how to allocate rewards.
Leveraging advances in ML, NLP and Deep Learning and merging into existing approaches allows organisations to get a pulse on employee sentiment quickly. VoE programs typically use a variety of methods to collect and analyse employee feedback, including surveys like eNPS, exit interviews, performance evaluations, employee forums, social networks, or focus groups. As such, employers can act on their employee feedback data by integrating those channels into text analytics platforms and make changes that improve retention rates, increase engagement, and boost performance.
4. Social Media & Brand Monitoring:
Every second, 3.3 million new posts appear on Facebook and almost half a million on Twitter. What if you wanted to keep track of all those times your brand was mentioned? Another benefit of Machine Learning and Natural Language Processing techniques is that it can help Customer Experience teams overcome information overload and analyse the massive pool of data from social media, where there stands to be a ton of information about your brand, product and support quality.
The idea is, you can extract, and filter data found on social media and feed into machine learning algorithms to identify business insights. Not only does this help determine things like press mentions or links to your site, but it also can help interpret how customers feel about your product/service.
NLP using sentiment analysis can help uncover messages that signify whether a customer is angry, satisfied, or needs help. It might also identify new leads, by picking up on messaging from people looking for a service like yours. For example, NLP could be used to pick out key phrases like Ã??Ã?Â¢??need accounting software.Ã??Ã?Â¢?? And, if your company makes accounting software, you have an opportunity to present your solution. In any case, as your company scales, a little help untangling the mess of social data and adding machine learning allows teams to automate and scale feedback processing activities.
Market leaders have been aware of the advantages of using open-ended data for a while. However, previously it was only open to small businesses where a decision maker can single-handedly read and analyse all of the comments. Large companies have resorted to manual tagging where several agents would read a piece of feedback and categorise it.
Unfortunately, manual tagging systems are costly and hard to maintain. A single agent can only process 1,000 pieces of feedback a day consistently and require training and constant quality control. Adding new languages requires a whole new set of workers or adding an expensive translation step which adds a new layer of complexity for CX teams.
Tools that utilise recent advances in deep learning open up new possibilities and solve the complex challenges that come with scale. CX pros can now centralise the collection of feedback across multiple channels and analyse structured and unstructured feedback at scale with human-level accuracy powered by state of the art neural network algorithms. Gone are the days of sifting through customer feedback and manual tagging. Advances in machine learning technology will be critical to the customer experience professional in the future as it enables them to pull detailed reports from datasets faster, with higher accuracy, that reveal trends and the potential to improve on the customer's experience. CX pros can now be proactive rather than reactive and avoid customer problems through advanced text analytic capabilities.