How Sabre is using Artificial Intelligence to change Travel Technology
The concept of Artificial Intelligence came into existence in 1950 when Alan Turing, wrote his landmark paper “Computing Machinery and Intelligence”.
This paper seeks to answer “Do machines think?” This led to the Turing test, which is a measure of the ability of a machine to demonstrate intelligent behavior equal to that of a human being.
Several researchers are now proposing the Winograd Schema Test as an alternative, and more recently Francois Chollet has introduced a framework called the Abstraction and Reasoning Corpus based on the theory of algorithmic knowledge, where he argues that measuring ability in any given task is not enough to measure intelligence.
Five years later, John McCarthy, a professor who spent most of his career at Stanford University, coined the term “Artificial Intelligence” (AI) and created LISP, the second oldest high-level programming language (after Fortran), in 1955.
Artificial Intelligence as a discipline is quite broad and supports the idea of technology that is capable of performing tasks in a way that we humans would find intelligent.
AI systems have influenced Natural Language Processing (such as machine translation), Machine Vision (such as autonomous self-driving cars), and Machine Learning (such as computer game-play).
Machine Learning (ML) is an AI sub-set where computer programs use data to learn, adapt and improve predictive capabilities over time, rather than performing tasks that they were specifically programmed to do.
The adoption of AI has been slow in the early days and it is only in the last decade, especially with the advent of Deep Neural Networks, that various industry verticals, including travel, have adopted AI-based techniques.
Following the airline industry’s deregulation in 1979, Sabre has played a pivotal role in facilitating all facets of passenger travel by leveraging our core competencies in operations research, computer science, and advanced data analytics to solve complex issues that generate revenue or contribute to cost-effective operational efficiencies.
Flight schedules, booking, air shopping, retailing, airline pricing, revenue control, crew planning, flight operations, personnel planning, cargo and customer service are the main focal areas.
Today we live in a world that is enabled for AI and ML. With such a Machine Learning algorithm, many of Sabre’s key developments in travel can now be improved.
AI and ML are at the crossroads of technology that think, connect and know.
At Sabre, we use these end-to-end approaches through marketing, digital distribution, delivery, and customer care. We also exploit these algorithms to achieve operational efficiencies within our complex, 24/7 running internal systems.
We can incorporate capabilities for continuous learning in an AI-enabled environment without human intervention.
The following are some of the value propositions we have implemented:
Dynamic Thresholds for Service Health Portal
The Sabre Health Portal (SHP) developed internal monitoring of the health of different applications in real-time and generating alerts based on thresholds that are dynamically adjusted.
Since traditional forecasting techniques do not manage volatility well, we used a regression model combined with machine learning to create dynamic bands of confidence.
Specifically, we found that a method called STL (Seasonal and Trend Decomposition using Loess) that uses the smoother Loess (Locally Estimated Scatterplot Smoothing) that uses seasonality and trend accounting for local regression was particularly effective.
Such an approach also allows the detection of anomalies computationally fast.
The advantages of dynamic thresholds in SHP are less false positive warnings, dynamically modified depending on the minute-monitored adjustment to the metrics.
Hotel Identity Deduplication
A key feature of a marketplace like Sabre is its ability to offer rich content from multiple sources (such as hotel rates). Yet different hotel aggregators (or suppliers) might not all reflect the same way a hotel property is.
Hotel names and addresses often contain errors, incorrect spellings or word translations. Integrating these contents from different hotel aggregators requires the ability to ensure continuity of a property through aggregators (i.e. we need to be able to identify the “Marriott Marquis in Time Square New York” as the same property even though different aggregators that depict it differently).
To solve this problem, a combination of Logistic Regression, a rule-based approach for hotel attributes and Fuzzy String Matching is used. TensorFlow Lattices are recent techniques that allow for correlations of such identity.
Time series models and Consumer Choice Modeling (CCM) techniques such as Multinomial Logit (MNL) models control market forecasting.
Machine learning techniques such as Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) can, however, model non-linear relationships, allow collinearity and provide flexibility to automatically implicitly learn customer behavior.
Our results show that, in some but not all cases, machine learning models can outperform traditional methods.
The detection of first and last names on passports from countries which are not compliant with ICAO (International Civil Aviation Organization) causes two problems:
- Inability to use the MRZ (Machine Readable Zone) for PNR (Passenger Name Record) of the Departure Control customer check-up and inability to pass APIS (Advanced Passenger Information System) data correctly at close flight.
- The problem of name recognition was solved through the evaluation of a range of supervised learning techniques. Supporting vector machines and recurring neural networks performed best with ~80% accuracy.
With the era of smart retailing, airlines want customer segmentation that goes beyond conventional booking groups.
Unmonitored learning strategies such as hierarchical clustering, k-means, sequential k-means, k-median, etc. can be used to create people across a range of dimensions such as advance purchase, length of stay, mid-week, weekend, number in a group, length of the journey, etc.
Sabre has recently demonstrated such customer and trip-purpose segmentation in a way that increases feedback query transparently augments input queries and output results, including New Distribution Capability (NDC) requests and responses.
Test and Learn Experimentation with the Multi-Armed Bandit
A fundamental building block in smart retailing is a component that leverages a multi-armed bandit (MAB) reinforcement learning technique to handle online-controlled experiments to learn customer behavior.
The experiment’s goal is to find the best or most efficient behavior and the distribution of randomization can be modified in real-time as the experiment progresses.
In air shopping tests, we also use this method to track shopping costs with cache updates, calculate shopping variety that maximizes conversion rates, determine the best fare with schedule/price/ancillary attributes to view on multi-month calendars, hotel retail with hotel websites offers, and recommend air packages (base fare + air ancillaries) to customers.
Many fields in which we consider such MAB learning useful include customized offer recommendation engines, bid configuration, screen optimization, identification of pattern recognition robotic shopping requests, hotel price anomaly detection, chatbots, hotel product standardization, and fraud detection.
At Sabre, our path to the use of artificial intelligence and machine learning algorithms is growing as we raise awareness of the importance of Artificial Intelligence and Machine Learning across the enterprise, communicate where it can be used, empower teams to learn and optimize our extensive data sources and put forward new travel-related value propositions that add value to our customers.
This article was originally published on sabre.com on January 16, 2020.
You can also subscribe to our YouTube Channel for interesting videos.