Browse Prior Art Database

Robust engagement analysis classification in social Customer Relationship Management (CRM) services

IP.com Disclosure Number: IPCOM000239679D
Publication Date: 2014-Nov-24
Document File: 5 page(s) / 96K

Publishing Venue

The IP.com Prior Art Database

Abstract

This idea proposes a method to build a robust engagement analysis classifier that can be applied to various domains by incrementally sampling a domain specific dataset to augment a baseline training set using the same classification feature sets. The method leverages the assumption that there are some commonalities in conversation patterns between agents and customers regardless of domains. The classifier is first trained for domain independent features in conversation. Then domain specific contents and topics are added by incrementally sampling engagement contents in different domains.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 29% of the total text.

Robust engagement analysis classification in social

Customer Relationship Management (CRM) services

This idea proposes a method to build a robust engagement analysis classifier that can be applied to various domains by incrementally sampling a domain specific dataset to augment a baseline training set using the same classification feature sets. The method leverages the assumption that there are some commonalities in conversation patterns between agents and customers regardless of domains. The classifier is first trained for domain independent features in conversation. Then domain specific contents and topics are added by incrementally sampling engagement contents in different domains.

Background

This idea describes a system that monitors the exchanges over social media between customers and a customer relationship management (CRM) team. This system aggregates social media textual messages (over Twitter, Facebook or any other social media) between customers and CRM agents in order to gather the material used for machine learning. The aggregation process relies on a linguistic framework call Dialog Act/Conversation Analysis that acts as a guide. Once data is aggregated, annotators label it according to the linguistic framework and the following 2 tasks: conversation labeling and engagement labeling. Conversation labeling describes how the conversation occurs between the customer and the agent. Engagement labeling is more specific and focuses only on how well the agent solves the customer problem. The labeled data is used to train a sequence classifier offline. During the online step, the classifier is used to classify new incoming data.

Problem

The automatic engagement analysis between agents and customers enables the social Customer Relationship Management (CRM) service to track the engagement conversations and to measure the effectiveness of the engagement in real time. In order to analyze the automatic engagement in real time, we need to have a robust conversation classifier that will work for different contents or formats of engagements. The classifier is usually built off-line using the training datasets in the domain of interests. Even though the basic format of conversations between CRM agents and customers is similar, the contents and the topics of texts and language used could be very domain specific. It is often necessary to have different conversation classifiers for different domains to achieve the desired accuracy.

Detailed Description

The idea proposes the method to improve the robustness of the engagement analysis in a social CRM system that can be applied to different domains by incrementally sampling the training dataset. Figure 1 shows the schematic overview of the proposed system.

1) The Baseline Engagement Analysis Model

First, the content of social media texts from the conversations between CRM agents and customers were collected and stored in a SQL database.(Step 1) The conversation threads between a CRM agent and a custom...