Deep Reinforcement Learning - Learning Individual Preferences

The main principle is to infer user preferences ahead of time to derive a matrix of user preferences based on what other users with similar interests prefer. Multilayer Perceptron for collaborative filtering can be used to accurately infer user preferences ahead of time by having the network learn and adapt as the users interact with a system. Given enough data points, the system becomes remarkably accurate in inferring user preferences as people who have commonalities tend to a cluster.

Overview of the Organisational Challenge

Organisations is required to provide services that are accessible to a wide-diverse demographic. A system that considers individualistic user preferences both programmatically and semantically for everyone is challenging to define. This is especially compounded by the fact that individual preferences can change day-by-day or depending on the individual's stage of life.

Solving this problem is essential because displaying content in one manner may be preferable to specific users while causing detraction from other users, directly affecting the ceiling of users a product can likely achieve and the amount of time a user spends on the platform. Real-world impacts have been observed with the social media app TikTok disrupting established platforms like YouTube and Instagram. While the later-mentioned platforms use social-media graph analysis to suggest content, TikTok relies solely on user-provided information and a combination of computer vision, natural language processing, and meta-data analysis to curate content. It has worked so well that user retention on the platform exceeds competitors.

The use of traditional machine learning to curate content is a well-established idea that later evolved and progressed to using artificial neural networks as artificial intelligence frameworks became more accessible. An early example of using Machine Learning to curate content was the Netflix Prize (https://en.wikipedia.org/wiki/Netflix_Prize), where Netflix called for submissions of machine learning models rewarding $1,000,000 USD to the winner. Later, iterations of this idea came to fruition with the MovieLens dataset (https://movielens.org/).

Current and future platforms will be required to establish this capability that uses artificial neural networks to retrain and attract users.

Organisational Data Available as AI Input

Data sources available for use in AI forecasting are as follows:

The following provides a high-level process for how to provide customer personalisation via artificial intelligence coupled with deep learning methods:

1. Customer meta-data from CRM systems (i.e. Salesforce, Microsoft CRM)
2. Purchase History (i.e. Amazon, Shopify)
3. Transaction timestamps and amounts (i.e. PoS Systems, Stripe, PayPal)

Integration Methodology

1. Capture features about a user that can infer user preferences
2. Train a deep-learning model with the captured features
3. Predict what the user would prefer based on the features
4. Customise content based via predictions for what the user wants to see
5. Continuously correct the model as the user interacts with the online system, improving the system over time.

Given Telemus AI™ takes care of most of the work, the organisation can focus on the business logic rather than the technical implementation.

Organisational Applications

The following lists other potential applications for your organisation:

• Customising content for a user to increase the likelihood of purchases
• Ensuring customer satisfaction with a service that improves user retention
• Ensuring content is fresh and relevant to the user

Potential and Realised Benefits

Telemus AI™ is an Australian-based artificial intelligence company providing advanced solutions to governments and enterprises. Contact us today for a free consultation on how the Telemus AI™ can be integrated into your organisation.