Deep learning in IT Services

In my previous blog i mentioned about four areas of deep learning that is possibly applicable. viz Text Processing , Image Processing , Object Processing and Voice Processing. we are today in a situation where multiple IT services offered are automated and what manufacturing industry is witnessing in 1980s is what we are witnessing in automation today.I could think of the following areas of IT where deep learning can be applied.

1. CSAT(Customer satisfaction) - The biggest issue that we are having today in IT services is to analyze the sentiments of the customers. The customers express their dissatisfaction in the words and if I need process the text then there are three possible outcomes positive , negative and neutral. Assume that I want to analyze the customer comments about a product expressed in Twitter we cannot go thru each and every comment and come to a conclusion. Here the deep learning helps us to analyze the comments and classifies and categorizes the comments into positive , negative and neutral. Assume that we do a service operation and the service desk engineer wants to know CSAT based on how the customer speaks or expresses the text it is easier to know. MetaMind a company that had been taken over by salesforce.com wanted to analyze the twitter comments based on the hashtag and arrived at the three possible sentiments.  Text processing or voice processing in real time the customer feedback specifically the service desk engineers will help them to improve their service. 

2. Root Cause Analysis Model

Currently the biggest issue in IT industry is the P1 problem. A P1 outage can cause a major worry for business. What does the resolver group typically when a problem happens ? Typically the resolver group gets together and analyze all the log files to get to the root cause. In a large enterprise this becomes more complex as more log files needs to be analyzed and complex patterns needs to be figured out typically from a heterogeneous set of people who are working on different specializations. During the P1 every minute counts and in short we need a machine that can recognize complex patterns in no time. Deep learning by the way of multiple levels of learning it does helps us to get to the point quickly than what humans can do. In a recent image recognition context conducted by Google trained machines were able to recognize complex patterns compared to humans. So what is our goal here ? To build a model that can easily search complex patterns across different devices to arrive at an intelligent conclusion. we need to process large amounts of Text more accurately and we need a machine that can be trained to recognize complex patterns in large amount of Texts.

3. Prediction :

The third important stuff in IT industry is our ability to predict the availability of Infra and apps. It is a known thing that more parameters to you supply the more accurate is a prediction. The downside to more parameters on a prediction of availability is the complexity. So it becomes more complex to predict if the no of parameters are more . 

All the above points to one requirement. Ability to train a machine to process large amounts of Text and recognize complex patterns. How to train a machine that recognizes large amount of Text Patterns. we use a neural network structure called Recurrent Net. Using Recurrent Net once can train a machine to recognize complex patterns and decisions can be taken based on the complex patterns. As we have decided to use Text Processing a way to acknowledge complex patterns let us learn about recurrent Net in my next blog. Recurrent Net is the way to solve all the above 3 issues.