Powered by artificial intelligence (AI), search and recommendation algorithms shape our interaction (and satisfaction) with online platforms. Developed to predict user choices, preferences and behaviours, its purpose is to improve the overall user experience of websites, apps, smart assistants and other types of computer programmes.

From Google to Amazon and Netflix, today’s biggest online retailers and service providers are making use of this class of machine learning to improve business conversion and retention rates: pushing products, boosting repeat sales and keeping customers happy and engaged.

How do search and recommendation algorithms work?

Where search algorithms work to retrieve relevant data, information and services in reaction to a user query, recommendation algorithms suggest similar alternate results based upon the user’s search (or purchase) history. 

Put simply, search algorithms assist users in finding exactly what they want, while recommendation algorithms help users find more of what they like.

Search algorithms

A search algorithm locates specific data within a larger collection of data. According to Internet Live Stats, on average, Google processes over 100,000 search queries per second. That’s an immense demand on a system, but while 98% of all internet users frequent a search engine monthly, it’s imperative that they be built to produce accurate results quickly and efficiently. 

Basic site-search has quickly become an essential feature of almost any website, while the search function is considered a fundamental procedure in computing overall (extending to coding, development and data science). Intended to keep all types of users happy and informed, search algorithms step in to get the right resources in front of the right people.

All search algorithms operate via a search key (or bar) and work by returning a success or failure status based on the entered information. They break a query down into separate words, and using text-matching, link those words to matching titles and descriptions in the data sets. 

Different search algorithms vary in terms of performance and efficiency, depending on how they are used and the available data. Some of the more commonly used search algorithms include:

• linear search algorithm
• binary search algorithm
• depth-first search algorithm
• breadth-first search algorithm

More complex algorithms can identify typing and auto-correct mistakes, as well as offering synonym recognition. Advanced algorithms can produce more refined results, factoring in popular answers, product rankings and other key metrics.

Google’s search algorithm

Google attributes its success to meticulous testing and complex search experiments. A combination of its infamous crawling “spider” bots, data-driven indexing and rigorous ranking system enables the search engine to meet its exemplary standards of relevance and quality. 

Analysing everything from sitemap to content, images and URLs used, Google is able to identify the best pages to signpost your query in a fraction of a second. The search engine even boasts a freshness algorithm that, in response to trending topics and keywords, shows users the most up-to-date online articles available in realtime. 

Good search engines boast another important feature: related results. This can make the difference between a bounce and purchase as customers are encouraged to keep browsing the site. This is where recommendation algorithms become useful.

Recommendation algorithms

Recommendation algorithms rely on data science to filter and recommend personalised suggestions (whether that be related search results or product recommendations) based on a user’s previous actions. Recommendation algorithms can generally be separated into two types: content-based filtering and collaborative filtering.

Content-based filtering

These algorithms factor in information (such as keywords and attributes) of both the user and the chosen item or product profile to generate recommendations. By utilising the customer’s personal data (such as gender, occupation and more), content-based filtering algorithms are able to assess popular products by age-group or locale, for example. Similarly, by analysing the product characteristics, the system can recommend other items with similar attributes. The more people that use the platform, the more data can be mined and improve the specificity of the suggestions.

The Netflix recommendation engine

Netflix states that recommendation algorithms are at the core of its product. In fact, 80% of viewer activity is driven by personalised recommendations from its engine. Netflix began experimenting with data as early as 2006 to improve the accuracy of its preference algorithms. As a result, the Netflix recommendation engine tracks numerous data points, from browsing behaviours to binge-watching habits, and filters over 3,000 titles at a time using 1,300 recommendation clusters based on user preferences. The platform has taken data beyond rating prediction and into personalised ranking, page generation, search, image selection, messaging, marketing, and more.

Collaborative filtering

These algorithms accumulate data from all users on a platform and work like a word-of-mouth recommendation. By comparing datasets, such as purchase or rating information, the algorithms help the platform identify kindred customer profiles and recommend other products or services favoured by these ‘similar users’.  

InData Labs notes the greatest merits of collaborative filtering systems as:

• capability of accurately recommending complex items (such as films, books or clothing) without requiring an “understanding” of the item itself
• basing recommendations on more personalised ‘similar’ users, without needing a more comprehensive knowledge of all products or all users of the platform
• ability to be applied to any domain and provide more versatile cross-domain recommendations

Why are search and recommendation algorithms so essential?

The number of digital purchases continues to climb each year, cementing e-commerce as an indispensable function of the global retail framework. And, in the world of online shopping, customers want accuracy, ease of use and appropriate suggestions.

For online businesses and service providers, some of the key benefits to using a search or recommendation algorithm include:

• improve the relevance of search results and reduce the time it takes to find specific products and services
• boost key metrics, including web visits and purchase rate, plus improve overall user loyalty and customer satisfaction
• aid in the selection process for an undecided customer, encourage them to interact with more products and and enter other potential purchases into their field of vision
• obtain data to target the right people with personalised ads and other digital marketing strategies to encourage users to frequent the website or platform

The quality of search and recommendation systems can significantly impact key business conversions, such as lead generations, customer sentiment scores and closing sales.

Amazon’s AI algorithms

As the leader of the global e-commerce market, Amazon is an almost-unrivalled product discovery and purchase platform, thanks to its optimal machine learning model. Built upon comprehensive ranking systems, the company’s A9 search algorithm analyses sales data, observes historical traffic patterns and indexes all product description text before a customer search query even begins, ensuring the best products are placed in front of the most likely buyers. 

The platform’s combination of intelligent recommendation tools forms the personalised shopper experience that has become so popular with consumers.

Get ahead with AI

Develop specialist skills spanning data analytics, neural networks, machine learning and more with University of York’s 100% online MSc Computer Science with Artificial Intelligence. 

This intensive online course, flexibly designed for remote learning, is geared to equip you for a range of roles in computer science and software development. One of the biggest trending careers in today’s jobs market, secure your space in this highly skilled, in-demand and lucrative field.

From visual confirmation of rare diseases to securing smartphones, facial detection and recognition technologies have become embedded in both the background of our daily lives and the forefront of solving real-world problems. 

But is the resulting impact an invasive appropriation of personal data, or a benchmark in life-saving security and surveillance? Wherever you stand on the deep-learning divide, there is no denying the ways in which this ground-breaking biometric development is influencing the landscape of artificial intelligence (AI) application.

What is facial detection and recognition technology?

Facial detection and recognition systems are forms of AI that use algorithms to identify the human face in digital images. Trained to capture more detail than the human eye, they fall under the category of ‘neural networks’; aptly-named computer softwares modelled on the human brain, built to recognise relationships and patterns in given datasets.

Key differences to note

Face detection is a broader term given to any system that can identify the presence of a human face in a visual image. Face detection has numerous applications, including people-counting, online marketing, and even the auto-focus of a camera lens. Its core purpose is to flag the presence of a face. Facial recognition, however, is more specialised, and relates specifically to softwares primed for individual authentication. Its job is to identify whose face is present.

How does it work?

Facial recognition software follows a three-part process. Here’s a more granular overview, according to Toolbox:

Detection

A face is detected and extracted from a digital image. Through marking a vast array of facial features (such as eye distance, nose shape, ethnicity and demographic data, and even facial expressions), a unique code called a ‘faceprint’ is created to identify the assigned individual.

Matching

This faceprint is then fed through a database, which utilises several layers of technology to match against other templates stored on the system. The algorithms are trained to capture nuance and consider differences in lighting, angle and human emotion.

Identification

This step depends on what the facial recognition software is used for — surveillance or authentication. The technology should ideally produce a one-to-one match for the subject, passing through various complex layers to narrow down options. (For example, some software providers even analyse skin texture along with facial recognition algorithms to increase accuracy.)

Biometrics in action

If you’re an iPhone X user, you’ll be familiar with Apple’s Face ID authentication system as an example of this process. The gadget’s camera captures a face map using specific data points, allowing the stored user to unlock their device with a simple glance.

Some other notable face recognition softwares include:

• Amazon Rekognition: features include user verification, people counting and content moderation, often used by media houses, market analytics firms, ecommerce sites and credit solutions
• BioID: GDPR-compliant solution used to prevent online fraud and identity theft
• Cognitec: recognises faces in live video streams, with clients ranging from law enforcement to border control
• FaceFirst: a security solution which aims to use DigitalID to replace cards and passwords
• Trueface.ai: services span to weapon detection, utilised by numerous sectors including education and security

Real-world applications

As outlined in the list above, reliance on this mode of machine learning has permeated almost all areas of society, extending wider still to healthcare and law enforcement agencies. This illustrates a prominent reliance on harvesting biometric data to solve large-scale global problems, spanning – at the extreme – to the life-threatening and severe. 

Medical diagnoses

We are beginning to see documented cases of physicians using these AI algorithms to detect the presence of rare and compromising diseases in children. According to The UK Rare Diseases Framework, 75% of rare diseases affect children, while more than 30% of children with a rare disease die before their fifth birthday. With 6% of people slated to be impacted by a difficult to diagnose condition in their lifetime, this particular application of deep learning is imperative.

Criminal capture

It was recently reported that the Metropolitan Police deployed the use of facial recognition technology in Westminster, resulting in the arrests of four people. The force announced that this was part of a ‘wider operation to tackle serious and violent crime’ in the London borough. The software used was a vehicle-mounted LFR system, which enables police departments to identify passers-by in real-time by scanning their faces and matching them against a database of stored facial images. According to the Met Police website, other applications of face identification include locating individuals on their ‘watchlist’ and providing essential information when there is an unconscious, non-communicative or seriously injured party on the scene.

Surveillance and compliance

A less intensive example, but one that could prove essential to our pandemic reality. Surveillance cameras equipped with facial detection were used to filter face mask compliance at a school in Atlanta, while similar technology has been applied elsewhere to conduct gun control.

Implications of procuring biometric information

Of course, no form of emerging or evolving technology comes without pitfalls. According to Analytics Insight, the accuracy rates of facial recognition algorithms are notably low in the case of minorities, women and children, which is dangerously problematic. Controversy surrounding data protection, public monitoring and user privacy persists, while the generation of deepfake media (and softwares like it), used to replicate, transpose and project one individual’s face in replacement of another, gives rise to damaging – and potentially dangerous – authentication implications. Returning to the aforementioned Met Police arrests, even in this isolated sample, reports of false positives were made, sparking outcry within civil rights groups.

At the centre of this debate, however, one truth is abundantly clear; as a society, we are becoming rapidly reliant on artificial intelligence to function, and the inception of these recognition algorithms is certainly creating an all new norm for interacting with technology.

Want to learn more about facial detection softwares? 

Dive deeper into the helps and harms and real-world applications of this mode of machine learning (and more) as part of our MSc Computer Science with Artificial Intelligence. 

On this course, you’ll develop core abilities of computational thinking, computational problem solving and software development, while acquiring specialist knowledge across increasingly sought-after skill sets spanning neural networks, genetic algorithms and data analytics. You’ll even undertake your own independent artificial intelligence project.

With employer demand for this expertise at an all-time high, enrol now and be part of this thrillingly fast-paced, far-reaching and ground-breaking field.

The exponential growth of digital connection in our world is all-pervasive and touches on every aspect of our daily lives, both personally and professionally.

Today, many people would be hard-pressed to imagine life before the advent of digital technology. It has blended and integrated seamlessly into everyday living. Global interconnectivity and the ability to communicate and network instantly is now a ‘given’. This expectation has markedly transformed the human experience across all areas, and the ‘always on’ culture has led to the creation of a vast online and computerised world. 

Creatively, this has inevitably resulted in phenomenally new ways of working and interpreting the world through data. Artificial intelligence (AI), and its attendant strands of computational science, are a vital link in the chain of twenty-first century life.

Whose choice is it anyway?

As the everyday world becomes saturated with digital information, the element of choice and decision-making becomes harder to navigate. The sheer amount of data available has led to the development of programs to enable and equip end-users to make such choices bespoke. Examples can be simplistic – from choosing the best shampoo to suit your hair type or which restaurant to choose for a special night out – or, more complexly, looking for a new home in a different area.

Recommender systems are built into many technological platforms and are used for both individual and ecommerce purposes. Although choice availability appears straightforward, the process behind it is remarkably elaborate and sophisticated.

The science behind the experience

The recommender system is an information filtering system run by machine learning algorithms programmed to anticipate and predict user interest, user preferences and ratings in relation to products and/or information browsed online.

Currently, there are three main types of recommender systems:

1. Content-based filtering. This is driven and influenced by user behaviour, and picks up what has been previously or currently searched for. Keyword-dependent, it seeks out the filtering approach and patterns regarding items in order to inform decision making.
2. Collaborative filtering recommender. This uses a more-advanced approach in that similar users are selected based on their choice of similar items. Collaborative filtering methods are centred on analysing the importance of user interactions on like-for-like user items and common selections, enabling comparisons to be made. 
3. Hybrid recommender. An amalgam of the two previous types, this system creates a hybrid once recommended items have been generated. 

The optimal functionality of recommendation engines depends upon information and raw data extracted from user experience and user ratings. When combined, these facilitate the building of user profiles to inform ecommerce targets and aims.

Multiple commonly accessed corporations and e-markets are highly visible and instantly recognisable on the online stage. Household names such as Amazon and Netflix are brands that immediately spring to mind. These platforms invest massively in state-of-the-art operations and big data collection to constantly improve, evolve and calibrate their commercial aims and marketing.

Computer architecture and system software are predicated on a myriad of sources and needs, and rely heavily on machine learning and deep learning.These two terms are often considered interchangeable buzzwords, but deep learning is an evolution of the former. Using programmable neural networks, machines have the ability to make accurate and precise decisions without human intervention. Within the machine learning environment, the term ‘nearest neighbour’ is an essential classification algorithm – not to be confused with its traditional association in the pre-computer era.

Servicing enabling protocols, technologies and real-world applications requires in-depth skills and knowledge across multiple disciplines. By no means an exhaustive list, familiarity with, and indeed specialist awareness of, the following terms are integral to the optimisation of recommendation algorithms and the different types of recommendation models:

• Matrix factorization. This refers to the collaborative filtering algorithms used in recommender systems. New user items are decomposed into the product of two lower-dimensionality, rectangular matrices. Mathematical modelling further splits these entities into smaller entries in order to discover the features or information leading to interactions between different users and items. Once alerted by the search engine, matrix factorization generates product recommendations.
• Cold-start problem. This is an issue which presents in both supervised and unsupervised machine learning and is frequently addressed.
• Cosine similarity. Needing to determine the nearest user to provide recommendations, this is an approach to measure similarities between two non-zero vectors.
• Data sparsity. Many commercial recommender systems are based around large datasets. As such, the user-item matrices used in collaborative filtering could be large and sparse. Therefore, such data sparsity could present a challenge in terms of optimal recommendation performance.
• Data science. IBM’s overview offers a comprehensive explanation, and introduction to, the employment of data science within its use of data mining and complex metadata.
• Programming languages. Globally used programming languages include Scala, Perl, SQL, C++ and Python. Python is one of the foremost languages used. Managed by Grouplens Research at the University of Minnesota, MovieLens makes use of Python in collaborative filtering. Its programme predicts film ratings based on user profiles, plus user ratings, and overall user experience. 

What’s happening with social media?

In recent years, recommender systems have become integral to the continued growth of social media. Due to the nature of the interconnected online community across locations and social demographics, a higher volume of traffic is both generated and triggered by recommendations enforced by likes and shares.

Online shopping has exploded as a result of the global pandemic. Websites such as Meta (formerly Facebook) and Etsy have been joined by new e-businesses and ‘shop fronts’, all of which incorporate the latest recommender technology. Targeted focus centres on growing user profiles by analysing purchase history and the browsing of new items. The aim is to both attract new users and retain existing ones. These embeddings are made possible through the use of recommender systems.

Careers in artificial intelligence and computer science      

Professionally relevant associations such as the Institute of Electrical and Electronics Engineers  (IEEE), and digital libraries such as Association for Computing Machinery (ACM), exist to provide further knowledge and support to those working in this fascinating field. 

Whichever specialisation appeals – computer science, software development, programming, AI-oriented solutions development – the many pathways are leveraged to build a rewarding career. There are many in-demand roles and no shortage of successful and creative organisations in which to work as evidenced in 50 Artificial Intelligence Companies to watch in 2022.

Further your learning in this fast-paced field

If you’re looking for a university course offering up-to-date theoretical and practical knowledge with holistic, pedagogical and real-world expertise, then choose the University of York’s online MSc Computer Science with Artificial Intelligence course and take your next step towards a fulfilling and stimulating career.

Even beyond the fully autonomous robots that appeared in the second half of the 20th Century, human beings have long been fascinated by automata. From Da Vinci’s mechanical knight  (designed around 1495) to Vaucanson’s “digesting duck” of 1739, and John Dee’s flying beetle of 1543 (which caused him to be charged with sorcery), the history of robotics stretches well before 1941 when Isaac Asimov first coined the term in one of his short stories.

Robotics combines computer science and mechanical engineering to construct robots that are able to assist humans in various tasks. We’re used to seeing industrial robots in manufacturing, for example, in the construction of cars. And, robots have been, and continue to be, particularly useful in heavy industry to help with processes that could be dangerous for humans resulting in injury or even death. Robotic arms, sometimes known as manipulators, were originally utilised in the handling of radioactive or biohazardous materials which would cause damage to human tissues and organs with exposure.

ABB Robotics is one of the leading multinational companies that deals in service robotics for manufacturing. Robotics applications include welding, heavy lifting, assembly, painting and coating, bonding and sealing, drilling, polishing, and palletising and packaging. These are all heavy-duty tasks that require a variety of end effectors, but robotics technology has progressed considerably and can also be seen in healthcare carrying out sophisticated medical procedures.

With the global pandemic, advances in robotics for surgery have been invaluable, allowing surgeons to remotely control the procedure from a safe distance. Now this technology is being developed further to allow surgeons to log in to operating theatres anywhere in the world, using remote control systems on a tablet or a laptop. They can then carry out surgery with the assistance of medical robotics on site. The technology was created by Nadine Hachach-Haram, who grew up in war-torn Lebanon, and there is no doubt it will be put to good use in locations where there is medical inequality or conflict or both.    

Mobile robots are also fairly common in various industries but we have yet to see them take over domestic chores in a way that perhaps the inventors of Roomba (the autonomous robotic vacuum cleaner) may have hoped they would. And yet, research has shown that people can tell what kind of personality a Neato Botvac has simply by the way that it moves. The study has provided interesting insights into human-robot interaction. Another study in 2017 demonstrated that anthropomorphic robots actually made people feel less lonely. People who work alongside collaborative robots, whether in the military or in manufacturing, also tend to express affection for their “cobots”.

Building on this affection that it seems people can and do develop for robots, Amazon has created Astro, a rolling droid that incorporates AI and is powered by Alexa, Amazon’s voice assistant. Astro also offers a detachable cup holder if your hands are too full to carry your coffee into the next room. Amazon has revolutionised automation with the use of Kiva robots in its warehouses and data science in its online commerce. So, it will be interesting to see if it can succeed where others have failed in making home robots popular.

What’s the difference between artificial intelligence and robotics?

While artificial intelligence refers only to the computer programs that process immense amounts of information in order to “think”, robotics refers to a machine designed to carry out assistive tasks that don’t necessarily require intelligence. Artificial intelligence has given us neural networks built on machine learning, which mimic the neural pathways of the human brain. It seems like an obvious next step would be to integrate these powerful developments with robotic systems to create intelligent robots.

Tesla’s self-driving cars contain neural networks that use autonomy algorithms to support the car’s real-time object detection, motion planning, and decision-making in complicated real-world situations. This level of advanced architectures and programming in the use of robotics is now being proposed with the Tesla Bot, a humanoid robot that Elon Musk says is designed to eliminate dangerous, repetitive, and boring tasks. Andrew Maynard, Associate Dean at the College of Global Futures, Arizona State University voices caution with regard to “a future that, judging by Musk’s various endeavours, will be built on a set of underlying interconnected technologies that include sensors, actuators, energy and data infrastructures, systems integration and substantial advances in computer power.” He adds that “before technology can become superhuman, it first needs to be human – or at least be designed to thrive in a human-designed world.”

It’s an interesting perspective, and one that goes beyond the usual moral and ethical concerns of whether robots could be used for ill, a theme often explored in science fiction films like The Terminator, Chappie, and Robot and Frank. Science fiction has always provided inspiration for actual technologies but it also serves to reflect back some of our own moral conundrums. If, as Elon Musk has indicated, the Tesla Bot “has the potential to be a generalised substitute for human labour over time,” what does this mean for artificially intelligent robotics? Would we essentially be enslaving robots? And, of course, this is built on the belief that the foundation of the economy will continue to be labour.

Tesla Bot hasn’t yet reached prototype stage yet, so whether Musk’s vision becomes a reality remains to be seen. The kinematics required to support bi-ped humanoid robots, though, are extremely complex. When it comes to bipedal robot design, LEONARDO (LEgs ONboARD drOne) is a quadcopter with legs. Only 76cm tall with an extremely light structure, even with an exoskeleton, LEONARDO looks far from the humanoids that the robotics industry may believe will become embedded in our everyday lives. Yet his multimodal locomotion system solves (or perhaps simply avoids) some of the issues real-world bipedal robots experience related to weight and centre of gravity. Mechatronics is an interdisciplinary branch of engineering that concerns itself with these kinds of issues. Some would say that mechatronics is where control systems, computers, electronic systems, and mechanical systems overlap. However, others would say that it’s simply a buzzword which is interchangeable with automation, robotics, and electromechanical engineering.

Build your knowledge in artificial intelligence with an MSc

Artificial intelligence is vital to take robotics research to the next level and enable robots to go beyond the relatively simple tasks they can complete on their own, as well as the more complex tasks they support humans with. Research in areas such as machine learning, distributed artificial intelligence, computer vision, and human-machine interaction will all be key to the future of robotics. 

Inspired to discover more about how you can specialise in AI? Study an 100% online and part-time MSc Computer Science with Artificial Intelligence with the University of York.

 

What is natural language processing?

Natural language processing (NLP) is a branch of artificial intelligence within computer science that focuses on helping computers to understand the way that humans write and speak. This is a difficult task because it involves a lot of unstructured data. The style in which people talk and write (sometimes referred to as ‘tone of voice’) is unique to individuals, and constantly evolving to reflect popular usage.

Understanding context is also an issue – something that requires semantic analysis for machine learning to get a handle on it. Natural language understanding (NLU) is a sub-branch of NLP and deals with these nuances via machine reading comprehension rather than simply understanding literal meanings. The aim of NLP and NLU is to help computers understand human language well enough that they can converse in a natural way.

Real-world applications and use cases of NLP include:

• Voice-controlled assistants like Siri and Alexa.
• Natural language generation for question answering by customer service chatbots.
• Streamlining the recruiting process on sites like LinkedIn by scanning through people’s listed skills and experience.
• Tools like Grammarly which use NLP to help correct errors and make suggestions for simplifying complex writing.
• Language models like autocomplete which are trained to predict the next words in a text, based on what has already been typed.

All these functions improve the more that we write, speak, and converse with computers: they are learning all the time. A good example of this iterative learning is a function like Google Translate which uses a system called Google Neural Machine Translation (GNMT). GNMT is a system that operates using a large artificial neural network to increase fluency and accuracy across languages. Rather than translating one piece of text at a time, GNMT attempts to translate whole sentences. Because it scours millions of examples, GNMT uses broader context to deduce the most relevant translation. It also finds commonality between many languages rather than creating its own universal interlingua. Unlike the original Google Translate which used the lengthy process of translating from the source language into English before translating into the target language, GNMT uses “zero-shot translate” – translating directly from source to target.

Google Translate may not be good enough yet for medical instructions, but NLP is widely used in healthcare. It is particularly useful in aggregating information from electronic health record systems, which is full of unstructured data. Not only is it unstructured, but because of the challenges of using sometimes clunky platforms, doctors’ case notes may be inconsistent and will naturally use lots of different keywords. NLP can help discover previously missed or improperly coded conditions.  

How does natural language processing work?

Natural language processing can be structured in many different ways using different machine learning methods according to what is being analysed. It could be something simple like frequency of use or sentiment attached, or something more complex. Whatever the use case, an algorithm will need to be formulated. The Natural Language Toolkit (NLTK) is a suite of libraries and programs that can be used for symbolic and statistical natural language processing in English, written in Python. It can help with all kinds of NLP tasks like tokenising (also known as word segmentation), part-of-speech tagging, creating text classification datasets, and much more.

These initial tasks in word level analysis are used for sorting, helping refine the problem and the coding that’s needed to solve it. Syntax analysis or parsing is the process that follows to draw out exact meaning based on the structure of the sentence using the rules of formal grammar. Semantic analysis would help the computer learn about less literal meanings that go beyond the standard lexicon. This is often linked to sentiment analysis.

Sentiment analysis is a way of measuring tone and intent in social media comments or reviews. It is often used on text data by businesses so that they can monitor their customers’ feelings towards them and better understand customer needs. In 2005 when blogging was really becoming part of the fabric of everyday life, a computer scientist called Jonathan Harris started tracking how people were saying they felt. The result was We Feel Fine, part infographic, part work of art, part data science. This kind of experiment was a precursor to how valuable deep learning and big data would become when used by search engines and large organisations to gauge public opinion.

Simple emotion detection systems use lexicons – lists of words and the emotions they convey from positive to negative. More advanced systems use complex machine learning algorithms for accuracy. This is because lexicons may class a word like “killing” as negative and so wouldn’t recognise the positive connotations from a phrase like, “you guys are killing it”. Word sense disambiguation (WSD) is used in computational linguistics to ascertain which sense of a word is being used in a sentence.

Other algorithms that help with understanding of words are lemmatisation and stemming. These are text normalisation techniques often used by search engines and chatbots. Stemming algorithms work by using the end or the beginning of a word (a stem of the word) to identify the common root form of the word. This technique is very fast but can lack accuracy. For example, the stem of “caring” would be “car” rather than the correct base form of “care”. Lemmatisation uses the context in which the word is being used and refers back to the base form according to the dictionary. So, a lemmatisation algorithm would understand that the word “better” has “good” as its lemma.

Summarisation is an NLP task that is often used in journalism and on the many newspaper sites that need to summarise news stories. Named entity recognition (NER) is also used on these sites to help with tagging and displaying related stories in a hierarchical order on the web page.

How does AI relate to natural language processing?

Natural language processing – understanding humans – is key to AI being able to justify its claim to intelligence. New deep learning models are constantly improving AI’s performance in Turing tests. Google’s Director of Engineering Ray Kurzweil predicts that AIs will “achieve human levels of intelligence” by 2029.

What humans say is sometimes very different to what humans do though, and understanding human nature is not so easy. More intelligent AIs raise the prospect of artificial consciousness, which has created a new field of philosophical and applied research.

Interested in specialising in NLP?

Whether your interest is in data science or artificial intelligence, the world of natural language processing offers solutions to real-world problems all the time. This fascinating and growing area of computer science has the potential to change the face of many industries and sectors and you could be at the forefront. 

Find out more about NLP with an MSc Computer Science with Artificial Intelligence from the University of York.

What is deep learning?

Deep learning is a sector of artificial intelligence (AI) concerned with creating computer structures that mimic the highly complex neural networks of the human brain. Because of this, it is also sometimes referred to as deep neural learning or deep neural networks (DNNs). 

A subset of machine learning, the artificial neural networks utilised in deep learning are capable of sorting much more information from large data sets to learn and consequently use in making decisions. These vast amounts of information that DNNs scour for patterns are sometimes referred to as big data.

Is deep learning machine learning?

The technology used in deep learning means that computers are closer to thinking for themselves without support or input from humans (and all the associated benefits and potential dangers of this). 

Traditional machine learning requires rules-based programming and a lot of raw data preprocessing by data scientists and analysts. This is prone to human bias and is limited by what we are able to observe and mentally compute ourselves before handing over the data to the machine. Supervised learning, unsupervised learning, and semi-supervised learning are all ways that computers become familiar with data and learn what to do with it. 

Artificial neural networks (sometimes called neural nets for short) use layer upon layer of neurons so that they can process a large amount of data quickly. As a result, they have the “brain power” to start noticing other patterns and create their own algorithms based on what they are “seeing”. This is unsupervised learning and leads to technological advances that would take humans a lot longer to achieve. Generative modelling is an example of unsupervised learning.

Real-world examples of deep learning

Deep learning applications are used (and built upon) every time you do a Google search. They are also used in more complicated scenarios like in self-driving cars and in cancer diagnosis. In these scenarios, the machine is almost always looking for irregularities. The decisions the machine makes are based on probability in order to predict the most likely outcome. Obviously, in the case of automated driving or medical testing, accuracy is more crucial, so computers are rigorously tested on training data and learning techniques.

Everyday examples of deep learning are augmented by computer vision for object recognition and natural language processing for things like voice activation. Speech recognition is a function that we are familiar with through use of voice-activated assistants like Siri or Alexa, but a machine’s ability to recognise natural language can help in surprising ways. Replika, also referred to as “My AI Friend”, is essentially a chatbot that gets to know a user through questioning. It uses a neural network to have an ongoing one-to-one conversation with the user to gather information. Over time, Replika begins to speak like the user, giving the impression of emotion and empathy. In April 2020, at the height of the pandemic, half a million people downloaded Replika, suggesting curiosity about AI but also a need for AI, even if it does simply mirror back human traits. This is not a new idea as in 1966, computer scientist Joseph Weizenbaum created what was a precursor to the chatbot with the program ELIZA, the computer therapist.

How does deep learning work?

Deep learning algorithms make use of very large datasets of labelled data such as images, text, audio, and video in order to build knowledge. In its computation of the content – scanning through and becoming familiar with it – the machine begins to recognise and know what to look for. Like the human brain, each computer neuron has a role in processing data, it provides an output by applying the algorithm to the input data provided. Hidden layers contain groups of neurons.

At the heart of machine learning algorithms is automated optimisation. The goal is to achieve the most accurate output so we need the speed of machines to efficiently assess all the information they have and to begin detecting patterns which we may have missed. This is also core to deep learning and how artificial neural networks are trained.

TensorFlow is an open source platform created by Google, written in Python. A symbolic maths library, it can be utilised for many tasks, but primarily for training, transfer learning, and developing deep neural networks with many layers. It’s particularly useful for reinforcement learning because it can calculate large numbers of gradients. The gradient is how the data is seen on a graph. So, for example, the gradient descent algorithm would be used to minimise error function and would be represented graphically as the gradient at its lowest possible point. The algorithm used to calculate the gradient of an error function is “backpropagation”, short for “backward propagation of errors”.

One of the most used deep learning models in reinforcement learning, particularly for image recognition, Convolutional Neural Networks (CNN) can learn increasingly abstract features by using deeper layers. CNNs can be accelerated by using Graphics Processing Units (GPUs) because they can process many pieces of data simultaneously. They can help perform feature extraction by analysing pixel colour and brightness or vectors in the case of grayscale.

Recurrent Neural Networks (RNNs) are considered state of the art because they are the first of their kind to use an algorithm that lets them remember their input. Because of this, RNNs are used in speech recognition and natural language processing in applications like Google Translate.

Can deep learning be used for regression?

Neural networks can be used for both classification and regression. However, regression models only really work well if they’re the right fit for the data and that can affect the network architecture. Classifiers in something like image recognition, have more of a compositional nature compared with the many variables that can make up a regression problem. Regression offers a lot more insight than simply, “Can we predict Y given X?”, because it explores the relationship between variables. Most regression models don’t fit the data perfectly, but neural networks are flexible enough to be able to pick the best type of regression. To add to this, hidden layers can always be added to improve prediction.

Knowing when to use regression or not to solve a problem may take some research. Luckily, there are lots of tutorials online to help, such as How to Fit Regression Data with CNN Model in Python.

Ready to discover more about deep learning?

The University of York’s online MSc Computer Science with Artificial Intelligence from the University of York is the ideal next step if your career ambitions lie in this exciting and fast-paced sector. 

Whether you already have knowledge of machine learning algorithms or want to immerse yourself in deep learning methods, this master’s degree will equip you with the knowledge you need to get ahead.