I am not sure if you are aware that the English Premier League kicked off at the weekend. There were 10 exciting games played over the course of three days. In case you are wondering, this is not a BBC Match of the Day post match report or analysis but a blog on how Google is transforming the search user experience.
Over the past months, I have spent some time studying machine learning (Scikit-Learn, machine learning module in Python) and how it applies to search engine optimisation. It is important to understand the meaning of machine learning before applying that to the search user experience.
Definition of Machine Learning:
Machine learning is simply a process whereby a computer teaches itself how to perform a task rather than being taught by humans or sticking to detailed programming. Put simply, it involves building systems or algorithms that can learn from experience.
Machine learning example for a Layman:
At first impression, the definition of machine learning seems a bit overwhelming and difficult to understand. Let’s use a real-world example to explain the meaning of machine learning. For example, a mum downloads a popular game known as Drawnimal for her toddler on an Ipad. This is the first time the child is exposed to this new game. The child learns how to play the game by trial and error or by making mistakes and learning from them and improves over time without being taught by his mum or his older brother. This is a machine learning process as the child can be viewed as the computer application and the game as the data. The mum or the older brother, in this case, could be viewed as the programmer who did not have to teach the child how to learn to play the game. The importance of machine learning or a child learning how to play the Drawnimal game independently is that the child could easily figure out how to play newer games with the same mindset than having to wait to be taught by his mum or big brother.
Supervised and Unsupervised Machine Leaning:
The two widely used approaches to machine learning are supervised and unsupervised. We will explain both concepts using a fruit case study. I love my fruits and hope you do as well? Below is an image of a variety of fruits that will aid our understanding of the two common methods in machine learning.
Supervised Learning: Supervised learning is a process where the computer algorithms learn from a training data set. In machine learning, you have an X (feature) and Y (Response). In the case of the fruit the X could be colour and the Y as the price of fruit. The model is guided or taught to group fruits according to colours such as red, yellow green and the rest.
The image above shows fruits being grouped according to colour. As new data of fruits are received by the algorithm they will be grouped according to their colours and will finally determine how the colour of fruits influences its price. The outcome could be that fruits that are yellow in colour have a bigger impact on the price of fruits than their red or green counterparts. This summarises how supervised learning works out. Some of the common models in supervised learning are regressions (Linear and Logical), supply vector machines, K-nearest neighbour and a host of others.
This takes place when algorithms are left to discover and express interesting structure to data. Using the fruit basket example, unsupervised models are allowed to group the fruits according to the best fit structure. This is like being given a bowl of fruits and asked to group them in any order of your choice. You could group them based on colour (yellow, green, red), size of the fruits (big, medium small) taste (sweet, sour and salty), shape (conical, oblong, elongated oblong) or the presence of seeds (large seeds, small seeds or seedless). These are some of the ways an unsupervised learning algorithm could decide the group or cluster of fruits. In unsupervised learning, there is no wrong or right answer as algorithms are allowed to create best fit clusters for any given data set. Common unsupervised learning algorithms include K-means clustering, neural networks and Density-based spatial clustering. A common use of unsupervised learning algorithm is in sentimental analysis. This involves determining if a review by a customer of a given brand is negative or positive. I used a free sentimental analysis tool for the sentence ‘I love going to school but I hate Maths’ and it stated that the sentiment of my sentence was 70% negative.
It is believed that Google could be applying sentimental analysis on ranking attribution for backlinks. That is, if someone links to your website and the sentence around the links are negative that is predicted to adversely affect the weight of the given link.
Machine Learning and the improvement of the search user experience:
Google utilises a variety of machine learning algorithms to ensure personalised and relevant results are generated for users. Running a simple search on Google leads to a rich array of results. Here is a personal example that wowed me over the weekend. I visited a sports map website to check the distance from Everton’s Goodison Park to Tottenham Hotspurs’ White Hart Lane. Then, I visited Google and ran a check for the capacity of Everton’s Goodison Park. My plan was to check for the capacity of Tottenham Hotspurs White Hart Lane Stadium afterwards but Google read my mind. Their sophisticated Machine Learning algorithms predicted that my next line of search could be Anfield, White Hart Lane or Selhurst Park. They personalised my experienced and saved me the hassle of running a second search on White Hart Lane’s capacity.
Machine learning is evolving and becoming more prevalent in the world of search engine optimisation. The search user experience is essential to Google’s search results. Search engines are more intelligent and are here to easily hydrate our thirst for information.