HP INCFollow

Abstract

Automating Printer Diagnostics to solve customer issues presents lot of challenges. If we can solve this

problem in a reliable way, it can help to save costs which are incur for every phone call made to call

agents by customers. Instead of customer calling a support agent, a tool should guide customers (either

residing at customer’s computer or over cloud) how to fix those issues. There are rule-based and Machine

learning based solutions developed to solve this issue using call agent record /Telemetry, but to the

best of our knowledge this problem has not been solved reliably. We present a machine learning

based way to solve this problem using model free reinforcement learning algorithms. Reinforcement

Learning Paradigm is a natural fit to this problem as it solves the problem of Sequential decision Making,

which is what a call agent does when he interacts with the customer over call. A Reinforcement

Learning (RL) agent learns from the environment while interacting with it and suggest an optimal action

based upon current state of the environment. The main idea proposed here is how to specify Printer

Diagnostics problem in Reinforcement Learning Framework. To that we propose 1) a way to design

the state/action/reward spaces and 2) a way to generate simulated environment (set of episodes) using

Printer hardware simulator. Using these two ideas we can use standard model free reinforcement learning

algorithms like MonteCarlo/TemporalDifference/QLearning to learn value function and update optimal

policy. The salient feature of this approach, as applied to Printer Diagnostics, is that a fix is predicted

by sensing the customer environment holistically unlike other existing approaches to Printer Diagnostics

where decisions are made based upon certain aspects of the user environment.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 License.

Download

Share

COinS