FMGC Case Study 7 - Inference Labs

Inference Labs Solution to Text & Voice Analytics

Problem statement

Create a Feedback Management System using Text analytics & Voice Analytics, which will be able to segregate subjective feedback based on sentiment and keywords, leading to action-oriented insights.

Data Insights provided by

Feedback Collected by	Lead Management Tools
	User Facing Applications
	Call Recordings
	Email Communication
Data Contains	Contains Leads
	Best Practices
	Market Intelligence
	Product Feedback
	Voice of Customers
Data Type	Un-Structured

Inference Labs Approach to Problem Statement

1. Data Collection

2. Data Storage

3. Data Tagging

Raw data that Is available in the database is used as training data to develop customized models by using techniques like manual tagging.
In this manual tagging process, the data will be tagged by trusted sources into appropriate sentiments that suits the textual data and later it’ll also be classified using certain keywords.
The manually tagged data will be useful for Building customizable models.

4. Data Cleaning

5. Model Building

Create a customizable AI model based on feedback categories, with corresponding outcomes tailored to meet the business requirements.
The pre-existing Inference AI models will be tested with the Data that is available in the database and then based on the Accuracy checks, the models can be customized with the data which can be used as training data, which results in better models that fits the business requirements.

6. Auto Training

Auto Training can be useful to build bigger and better models, so that the AI engine can be able to learn new keywords and sentiments on its own.
The newly approved keywords will be used to train the keyword extraction model, and the updated model weights will be saved.
The updated inference AI engine will receive the existing data from the database and generate new sentiments and keywords.
The newly generated sentiments and keywords are stored in the database and will be further classified into various categories.

Use case & Inference Labs Approach

Tone Detection

Tone detection refers to the process of identifying the emotion in a speaker voice to extract the tone of the speaker we use certain APIs that extract the necessary information with higher accuracy.

Sentiment Analysis

Inference AI Engine can generate the sentiments for the textual data with a certain degree of accuracy. With more addition to the classes of sentiments, models can be re-trained on the manually tagged data.

Keyword Classification

a. Keyword identification

The Inference AI engine can be able to identify the keywords that are relevant to the text, which are pre-determined by the business using manual tagging. Business can be able to add more and more keywords to the list of initial manual tagging.

b. Auto generate Keywords

Inference AI Engine has a capability to identify Keywords that are not predefined by the business. These Auto identified keywords are called Bigrams and Trigrams which can be further approved, to be involved in the original keywords by the Business.

Now, these recommended keywords will be scrutinized, and only approved keywords will be added back to the database.

Trigger Mechanism

The Database will be directed to UI mechanism and create a nuance of the data and can be visualized using some advanced graphs and charts and future trends.
The analysis and alerts will be generated based on the sentiments and keyword categories. Based on the Thresholds and given sentiment limitations by the Business heads and the assigned stakeholders will be updated with these alerts.