‘In god we trust, everyone else please bring data’
– American Statistician Edwards Demming
Analytics can perhaps be taken to be the basis of building the foundation of a society based on reason and logic, not just feelings or gut. It not only helps in finding patterns through an iterative process but also in making predictions on never before seen data. The main trigger of such frenzy over data science and analytics can be owed to not only modern cheaper sources of data storage but also to high processing power of systems, ease of internet connectivity and enhanced security features.
There are numerous models and algorithms for implementation of machine learning, however the coverage of the article is around the future possibilities of analytics applications. It touches upon industries like the e-commerce space and research. It also covers specific potential applications – like suspicion algorithms and driverless cars – with a future outlook.
The E-commerce Space
Currently, recommendation engines – like the one used by Netflix – are using various learning techniques like clustering, decision trees and collaborative filtering for giving better suggestions. In the future, generating recommendations on the basis of what the user currently likes may be insufficient to generate best possible set of recommendations. This is because of the Law of Diminishing Marginal Utility – when a person has already bought product ‘X’, he/she may not need it or other similar products for a significant time to come, depending on the life of the product in question.
Instead, if the recommendation engine is smart enough to ‘learn’ the allied or soon to-be-required products, those would be the best recommendations to begin with. For example, camera accessories might be a good start for recommendation if a camera is bought. Leveraging the huge amount of data available from a user’s social media profile can also be done by allowing social login (that is, login through a social media platform) on seller/merchandiser’s website. So now, if a user posts on Facebook about supporting Indian-made goods, an ideal recommendation would be a khadi cloth or Patanjali product. Even if the user does not post anything, personal details like birthdays of the users, their family members or close friends can serve as an apt occasion for throwing in recommendations for gifts.
An area which is becoming a growing concern for firms is online reviews. Since these reviews have become important in influencing purchase decisions (for example, The Case of Yelp.com), there have been cases of their manipulation as well. The current fake review detectors work with the approach of only focusing on what a reviewer writes (that is, the lexical and semantic part). There are inherent flaws in this approach – one of them is that the reviewer may become smarter over time and may use words/expressions which go undetected. One way in which analytics can aid is by finding if the reviewer is more focused on completing more number of fake reviews to earn more, instead of wasting time by engaging in discussions with peers. Another way of handling such reviews is by using supervised learning to classify authentic and fake reviews.
Analytics can also be used for (if not curbing, then at least) preventing ‘web-rooming’. This can be done by continuously monitoring an individual’s surfing habits. If a user’s pattern reflects that of spending significant time on an app/website, doing proper research by using filters/sort options, but not buying – then perhaps personalized offers can do the trick – be it in the form of discounts, shipping charge waivers, freebies or loyalty bonus.
Other Potent Arenas for Leveraging Analytics
Suspicion Algorithms are another domain where analytics can provide much input. A suspicion can be detected corresponding to anything starting from abandoned bags on railway platforms or unnatural patterns in weather, to massive buy/sell of securities in a short span indicative of insider trading. Although the inputs and outputs of these systems may be different, but the intrinsic problem remains the same – how to decide on the criteria of the cut-off line between normal and abnormal/suspicious operation? This is where analytics can help in breaking down patterns in images, weather data or financial securities trading data respectively, as mentioned in the examples of suspicion above.
These days, there is a huge debate around which mobile virtual assistant is better out of Apple’s Siri, Microsoft’s Cortana, Amazon’s Alexa or Google’s Now. These personal assistants are merely a stepping stone towards more comprehensive systems. The evolved form of future systems can be foreseen to be highly responsive to human needs, state of mind and emotions. Just like the show ‘Jetsons’ which used to be broadcasted on Cartoon Network, these systems can be anticipated to be highly evolved in terms of understanding higher functions of human interaction like sarcasm. This can be greatly facilitated with the help of analytics, especially supervised learning – as each language is unique and needs to handle exceptions of its own. Some work has already started in this direction.
A high level application of Analytics using Big Data could also be in precision measurements in scientific research. A live example of this could perhaps be the data processing continuously needed for day-to-day operations of the world’s largest (27 km circumference) and most powerful particle accelerator, the Large Hadron Collider at CERN, Geneva. An enhanced application of analytics can be in the domain of the much fantasized interstellar travel where it can aid in not only course correction of space-crafts used, but also in increasing reliability and cost effectiveness of the space missions.
The concept of driverless cars or self-driven cars is gaining traction these days. Google, Mercedes, BMW and Tesla (among other players) are planning to launch their respective version of these cars soon. Not only do these cars have the advantage of avoiding dependence on human judgement, but also of making smart decisions like re-routing due to traffic congestion. In the future, analytics can aid a lot in training not only cars but also other modes of transport like road construction equipment, ships, trains.
There are several other specific applications through which analytics and data science can leverage machine learning in future.
Implementation of all the ideas suggested above do come with their unique challenges. Machine learning techniques, when applied to a given situation generally face the following five types of problems: how to establish ‘learning’ for large scale of data, different data types, high speed streaming data, data with low value/meaning and uncertain/incomplete data. Each of these problems has specific remedies as mentioned in the source above. These problems come from the five major characteristics of Big Data– Volume, Variety, Velocity, Veracity and Value.
Regardless of the numerous possibilities or challenges that can be associated with machine learning, its true utility will be reflected in its easy integration with our daily routines. Nonetheless, analytics and data science can very well be anticipated to leverage machine learning in future. The first steps of the same can be seen from recent past when Google acquired the artificial intelligence firm DeepMind. The Neural Turing Machine being developed by DeepMind is said to mimic the human brain’s short term working memory.
- M. Frydenberg, p. 2, page 7 
Title: Introducing Big Data Concepts in an Introductory Technology Course
- SAS Website: http://www.sas.com/en_us/insights/analytics/machine-learning.html
Title: Analytics – What it is and why it matters?
- Carlos A. Gomez-Uribe et al., Article Number 13, Volume 6, Issue 4, Journal on Title: ACM Transactions on Management Information Systems (TMIS) [Jan 2016]
Title: The Netflix Recommender System: Algorithms, Business Value, and Innovation
- Stanford University, US 
Title: Recommendation Systems (http://infolab.stanford.edu/~ullman/mmds/ch9.pdf)
- Jian Wang et al., University of California [July 2011]
Title: Utilizing Marginal Net Utility for Recommendation in E-commerce
- Fatima EL Jamiy et al., Morocco 
Title: The Potential and Challenges of Big data – Recommendation Systems next level application
- Larry Drebes, Harvard Business Review [Oct 2011]
Title: Social Login Offers New ROI from Social Media
- Michael Luca, Harvard Business School 
Title: Reviews, Reputation, and Revenue: The Case of Yelp.com
- Daily Mail, UK [Oct 2013]
Title: Samsung ordered to pay $340,000 after it paid people to write negative online reviews about HTC phones
- Dongsong Zhang et al., Journal of Management Information Systems, Vol. 33, No. 2, pp. 456–481 
Title: What Online Reviewer Behaviors Really Matter? Effects of Verbal and Nonverbal Behaviors on Detection of Fake Online Reviews.
- Snehasish Banerjee et al., ACM Digital Library [Jan 2015]
Title: Using Supervised Learning to Classify Authentic and Fake Online Reviews
- MEC (Media Agency), UK [May 2016]
Title: The New Consumer Pathway: From Digital Searching to In-Store Purchasing
- Michael L. Rich, University of Pennsylvania Law Review [Apr 2016]
Title: Analytics, Automated Suspicion Algorithms, and the Fourth Amendment
- Gianluca Demartini, Elsevier ScienceDirect, Computer Networks, Vol. 90, p5-13 [Oct 2015]
Title: Hybrid Human–Machine Information Systems: Challenges and Opportunities
- Lisa S. Pearl et al., Interaction Studies (University of California), Vol. 15 Issue 3, p359-387
Title: Can you read my Mindprint?
- Surekha Sharad Muzumdar et al., India [Aug 2015]
Title: Big Data Analytics Framework using Analytics on Multiple Datasets
- Amy McGovern et al., Springer Link, Volume 84, Issue 3, pp 335–340 [Sept 2011]
Title: Analytics in Space: Extending Our Reach
- Samuel Gibbs, The Guardian [May 2016]
Title: Self-Driving Cars: Who’s Building Them and How Do They Work?
- John Greenough, Business Insider [Jun 2016]
Title: 10 million self-driving cars will be on the road by 2020
- Jemima Kiss, The Guardian (UK) [Oct 2015]
Title: Self-driving Cars: Safe, Reliable – but a challenging sell for Google
- H. James Wilson et al., Harvard Business Review [May 2016]
Title: How Companies Are Using Machine Learning to Get Faster and More Efficient
- Junfei Qui et al., Springer Open, EURASIP Journal on Advances in Signal Processing [May 2016]
Title: A Survey of Machine Learning for Big Data Processing
- Dr. Ervin Ramollari et al., European Journal of Economics and Business Studies, Vol. 4, Nr. 1 [Jun 2016]
Title: Big Data using Cloud Computing – Opportunities for Small and Medium Sized Enterprises
- Mark Prigg, Daily Mail, UK [Oct 2014]
Title: Google reveals it is developing a computer so smart it can program ITSELF
About the Author:
Aayush is a final year student of VGSoM, IIT Kharagpur. He has been nominated as Bloomberg Champion of his college. He has done his summer internship at Wipro Limited in analytics domain. He is an avid reader and has conducted multiple knowledge sharing sessions on the statistical tool ‘R’ in his college.