IoT and Recommendation Systems
The Internet of Things (IoT) is a system of interrelated computing devices, mechanical and digital machines, objects, animals or people that are provided with unique identifiers and the ability to transfer data over a network without requiring human-to-human or human-to-computer interaction.
Near – field communications, real-time localisation and embedded sensors turn everyday objects into smart objects that can understand and react to their environment.
Hence, running shoes, thermostats, mirrors, food measurement machines, weighing machines, the refrigerator washing machine and dishwasher can all become part of the Internet of Things.
The ability to combine data from several heterogeneous sources (insert what this means) and share that data with other systems is where the true power of IoT lies (Kranz et al. 2010).
One of the fastest-growing segments within IoT is individual health metrics and is an outflow of the interest consumers are directing at personal health monitoring.
The quantified self refers to the ability to capture and track large amounts of data about oneself, typically with the objective of improved health outcomes or productivity.
The algorithmic self is the recognition of the role that algorithms play in the analysis of data, that is then turned into insights and recommendations. A key growth segment for IoT is in individual health metrics, via self-tracking wearables, clinical remote monitoring, wearable sensor patches, Wi-Fi scales and other biosensing applications.
Simple devices and applications can record data on the number of steps taken, or miles run, calories burnt, and the amount and quality of sleep. Other wearables measure blood chemistry and headbands monitor brain activity via EEG to identify seizures and mood changes. This data can be tracked over time and compared to data from the quantified self- population.
Longevity, the Internet of Things (IoT) and the Quantified Self
Changes in demographics, such as falling birth and increasing mortality rates in the western hemisphere, means that individuals are living longer. Hence the increase in the numbers of people presenting cardiovascular lifestyle-related diseases.
As western nations, de-populate due to low birth rates, and the inhabitants live longer the reversal in population growth will present a unique set of challenges for both the developed and developing world.
In the developed world declines in birth rates, the rise in the proportion of individuals entering retirement, relative to those that support them will have a significant impact on the healthcare systems of countries.
Factors Driving the Adoption of Big Data Analytics and IoT in Healthcare
Conversely, the increase in the forecasted numbers of people across all age groups developing (CVD's) again will place significant pressure on healthcare infrastructures within emerging markets.
As purchasing power parity increases within developing nations, consumers incumbent within these, countries increasingly adopt western lifestyles particularly concerning diet and less exercise.
As mentioned in the previous post titled “The Consumerisation of Healthcare” the Wellness Movement and its convergence of the
Quantified Self, and personalised health is enabling young and mature consumers to take greater control of their health outcomes particularly with regards to lifestyle-related diseases.
Over the past 10 years the emergence of personal health tracking apps, and the drive to empower consumers with personalised data to help them make better health-related decisions.
Healthcare Consumers are Increasingly Using Tech to Manage Their Health.
Has resulted in dramatic growth in the use of health care apps. One way in which these applications facilitate the delivery of personalised advice is through the use of recommendation systems.
Recommendation Systems and the Consumerisation of Health
A recommender system is a subclass of information filtering systems that seeks to predict the rating or preference a user would apply to an item. They are predominantly used for commercial applications.
Felferning and Burke 2008, described a recommender system as any system that guides a user to interesting or useful objects for the user in a large space of possible options or that produces such objects as output.
Modern mobile and sensor technologies enable the recording of all kinds of data, related to a persons' daily lifestyle such as exercise, steps taken, body weight, food consumption, blood pressure, cigarettes smoked. This type of self-data tracking is often referred to as the Quantified – Self.
The tracking of measurements such as step counts, spent calories and body weight are very effective in empowering and motivating individuals to make lifestyle changes. When monitored over time, consumers can gain insights into their progress and experience the direct relationship between their efforts on the goals they set for themselves.
There are several types of recommender systems in this blog post we will focus on three: the virtual coach, the virtual nurse and the virtual sleep regulator.
The Virtual Coach: The virtual coach is used to motivate users' applications that quantify their activities. By recommending new activities based on their demographic information and historical behaviour.
The Virtual Nurse: Helps chronic patients to reach their target by recommending activities based upon their medical history.
The Virtual Sleep Regulator: provides walking and sleeping time recommendations for insomnia patients to improve their sleep qualities.
In the USA, by 2013, the Quantified Self movement has already gone mainstream, with 60% of U.S adults currently tracking their weight, diet or exercise routine. 33% are monitoring other factors such as blood sugar, and a further 27% of U.S Internet users track health data online, with 9% signed up for text message heart alerts. There are 40,000 health applications available.
Total mHealth Forecast Worldwide
However, this begs the question of why are cardiovascular diseases, such as obesity and diabetes still prevalent within the USA? And other countries where there is high adoption of health applications?
Recommender systems can help to enable the easy identification of relevant artefacts for users and thus improve user experiences. Currently, recommender systems are widely and effectively used in the e-commerce domain e.g., online music services, and online bookstores. Recommendation systems are split into 3 categories:
First Generation Recommendation Systems: Knowledge-based, content-based, collaborative filtering, and Hybrid.
Collaborative Filtering: methods are based on the analysis of large amounts of user behaviours, activities and preferences to predict what uses will like best based upon their similarity to other users.
How a collaborative filtering recommender works is through identifying the k-nearest neighbours and to use the preference of the users as a proxy for the preference of the current user.
Content-based Filtering: Assumes monotonic personal interests and hence uses the content consumed by users, to identify new similar content or items the users will respond to positively. During content filtering the attributes of an item is calculated the higher the degree of similarity the greater the probability that the item will be of interest to the consumer.
Knowledge-based Recommendation: Does not use ratings and textual items descriptions. But instead offered items represented in terms of constraints, rules or similarity metrics. The user states what they want in terms of item specification, the constraints are interpreted ad the resulting items are presented.
Hybrid Recommendation: is based on the idea of combining basic recommendation approaches in such a way as to compensate for the weaknesses of the other. Hence if the user has already consumed some items in a category similar items will be recommended to the user. Combined recommender approaches improve the prediction quality of recommendation systems.
Hybrid recommendation systems may be used to deliver personalised professional medical recommendations based on hybrid matrix factorisation. Through the development of a topic model, used to develop user preferences based upon the topics they searched for.
Including, reviews, to determine user sentiment regarding the quality of care they received and the rating of their encounters with medical professionals.
Group Recommender Systems: Assume that recommendations are not just for an individual, insofar as a whole group or customer segment should be satisfied by a given recommendation.
Hence recommendations are determined by group decision heuristics, “least misery” heuristic recommends items that will minimise the misery of all group members. Conversely, a heuristic concerned with generating the "most pleasure" for a group will attempt to maximise the pleasure of individual group members.
To improve the accuracy of recommendations individual group recommendation heuristics can be combined.
Second Generation Recommendation Systems: Matrix factorisation, web usage mining based, Personality Based recommendation systems.
Matrix Factorisation techniques are prediction algorithms use in data science. When applied to business problems they may be used to make recommendations to consumers with similar preferences. Other algorithms may be used to predict the likely ratings that users will give to a movie, hotel, item of clothing or book, from a similar segment.
When users give feedback on a hotel, movie, a consumer electronic good. The collection of feedback can be represented within a matrix. One of the key strengths of matrix factorisation is that it can give implicit feedback on information not directly given but derived from analysing user behaviour.
Web Usage Mining: Focuses on predicting users’ preferences and behaviour by analysing traditional data mining techniques. Customer clickstream data is a rich source of information. Navigational data is useful in providing recommendations to users based on the pages visited, time spent etc.
Personality Based: Personality theorists claim that a user’s personality traits have a substantial influence on preferences and subsequently on behaviour.
The human personality significantly influences the way people think, feel and, especially, behave. Determining a user’s preferences is an important condition for effectively running these automatic recommendation systems.
Personality traits are defined as “endogenous, stable, hierarchically structured basic dispositions governed by biological factors such as genes and brain structures”.
These traits remain quite stable over the entire lifetime and through varying situations and that is why a user’s personality is a good starting point for predicting user behaviour.
Especially in electronic markets where digitised information for mining a user’s personality is frequently available. In online social networks (OSN) such as Facebook, and LinkedIn.
Third Generation Recommendation Systems: Collaborating Filtering using Deep Learning techniques. Deep content-based, Combine Modelling of Users and Items Using Reviews, CoNN etc.
Collaborative Filtering Using Deep Learning: Neural networks are trained to predict ratings or interactions based on the item and user attributes. They can also be used to predict the next action a user will take based upon historical data and content. The different methods used to apply deep learning recommendation systems include.
- Predicting ratings through the use of feed-forward networks.
- Predict the next user action by recurrent neural networks and or reinforcement learning
- Embed user and or item attributes into low dimensional space using Autoencoders.
Deep Content-Based Recommendation Systems: Recommendation systems based on content filtering approaches exploit the interactions between users and items e.g. clicks or ratings, which are represented in a user-item (rating) matrix R.
The task is then to predict missing rating for pairs of users U and items I and recommend to the target user u the (unseen) items with the highest predictions.
Collaborative Filtering Models: identify similarities between users and or items either in low – dimensional joint representation of users and the items. Or by directly computing similarities from user-item matrix memory-based CF.
The application of recommendation systems, in the software used by consumers to quantify themselves. Enable individuals to self-track, biological, physical, behavioural and or environmental information as individuals and groups.
Female Founders in Virtual Reality and Augmented Reality
