Skip to main content
Disciplines
Automotive Business IT + Informatics Construction + Real Estate Electrical Engineering + Electronics Energy + Sustainability Insurance + Risk Finance + Banking Management + Leadership Marketing + Sales Mechanical Engineering + Materials
Events
DE
EN
Books
Journals
Topic Page
Marketing
Start single access now
Access for companies
EXTENDED SEARCH
Log in
Top

2021 | Book

Chatbots for CRM and Dialogue Management Read chapter Read first chapter

Artificial Intelligence for Customer Relationship Management

Solving Customer Problems

Author: Dr. Boris Galitsky

Publisher: Springer International Publishing

Book Series : Human-Computer Interaction Series

Part of: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Login to get access
insite
SEARCH

About this book

The second volume of this research monograph describes a number of applications of Artificial Intelligence in the field of Customer Relationship Management with the focus of solving customer problems. We design a system that tries to understand the customer complaint, his mood, and what can be done to resolve an issue with the product or service.

To solve a customer problem efficiently, we maintain a dialogue with the customer so that the problem can be clarified and multiple ways to fix it can be sought. We introduce dialogue management based on discourse analysis: a systematic linguistic way to handle the thought process of the author of the content to be delivered. We analyze user sentiments and personal traits to tailor dialogue management to individual customers. We also design a number of dialogue scenarios for CRM with replies following certain patterns and propose virtual and social dialogues for various modalities of communication with a customer.

After we learn to detect fake content, deception and hypocrisy, we examine the domain of customer complaints. We simulate mental states, attitudes and emotions of a complainant and try to predict his behavior. Having suggested graph-based formal representations of complaint scenarios, we machine-learn them to identify the best action the customer support organization can chose to retain the complainant as a customer.

Table of Contents

Frontmatter
Chapter 1. Chatbots for CRM and Dialogue Management
Abstract
In this chapter, we learn how to manage a dialogue relying on the discourse of its utterances. We show how a dialogue structure can be built from an initial utterance. After that, we introduce an imaginary discourse tree to address the problem of involving background knowledge on demand, answering questions. An approach to dialogue management based on a lattice walk is described. We also propose Doc2Dialogue algorithm of converting a paragraph of text into a hypothetical dialogue based on an analysis of a discourse tree for this paragraph. This technique allows for a substantial extension of chatbot training datasets in an arbitrary domain. We evaluate constructed dialogues and conclude that deploying the proposed algorithm is a key in successful chatbot development in a broad range of domains where manual coding for dialogue management and providing relevant content is not practical.
Boris Galitsky
Chapter 2. Recommendation by Joining a Human Conversation
Abstract
We propose a novel way of the conversational recommendation where instead of asking user questions to acquire her preferences, the recommender tracks her conversations with other people, including customer support agents (CSA) and joins the conversation only when there is something important to recommend and the time is correct to do so. Building a recommender that joins a human conversation (RJC), we propose information extraction, discourse and argumentation analyses, as well as dialogue management techniques to compute a recommendation for a product and service that is badly needed by the customer, as inferred from the conversation. A special case of such conversations is considered where the customer raises his problem with a CSA in an attempt to resolve it, along with receiving a recommendation for a product with features addressing this problem. The performance of RJC is evaluated in a number of human–human and human-chatbot dialogues and demonstrates that RJC is an efficient and less intrusive way to provide high relevance and persuasive recommendations.
Boris Galitsky
Chapter 3. Adjusting Chatbot Conversation to User Personality and Mood
Abstract
As conversational CRM systems communicate with human customers and not other computer systems, they need to tackle human emotions in a way to optimize the outcome of a chatbot session. A chatbot needs to understand the emotions of its peers and produce utterances which not only match their emotional states but also attempt to improve it towards solving a problem. We construct a model and a training dataset of emotions and personality from various sources to properly react to the customer in the emotional space and to navigate him through it. We evaluated an overall contribution of a chatbot enabled with affective computing and observed up to 18% boost in the relevance of responses, as perceived by customers.
Boris Galitsky
Chapter 4. A Virtual Social Promotion Chatbot with Persuasion and Rhetorical Coordination
Abstract
We build a chatbot that delivers content in the form of virtual dialogues automatically produced from documents. Given an initial query, this chatbot finds documents, extracts topics from them, organizes these topics in clusters, receives from the user clarification on which cluster is most relevant, and provides the content for this cluster. This content can be provided in the form of a virtual dialogue so that the answers are derived from the found documents by splitting it, and questions are automatically generated for these answers. Virtual dialogues as search results turn out to be more effective means of information access in comparison with original document chunks provided by a conventional chatbot. To support the natural flow of a conversation in a chatbot, the rhetorical structure of each message had to be analyzed. We classify a pair of paragraphs of text as appropriate for one to follow another, or inappropriate, based on communicative discourse considerations. We then describe a chatbot performing advertising and social promotion (CASP) to assist in the automation of managing friends and other social network contacts. This agent employs a domain-independent natural language relevance technique that filters web-mining results to support a conversation with friends. This technique relies on learning parse trees and parses thickets (sets of parse trees) of paragraphs of text such as Facebook postings. We evaluate CASP in a number of domains, acting on behalf of its human host. Although some Facebook friends did not like CASP postings and even unfriended the host, overall social promotion results are positive as long as relevance, style and rhetorical appropriateness are properly maintained. Finally, we propose a way to improve a discourse parsing by a refinement of default rhetorical relations, based on an analysis of Abstract Meaning Representation (AMR). A number of AMR semantic relations such as Contrast can be used to detect a specific rhetorical relation.
Boris Galitsky
Chapter 5. Concluding a CRM Session
Abstract
We address the issue of how to conclude a CRM session in a comprehensive manner, to satisfy a user with the detailed extended answer with exhaustive information. For a question-answering session, the goal is to enable a user with thorough knowledge related to her initial question, from a simple fact to a comprehensive explanation. In many cases, a lengthy answer text, including multimedia content compiled from multiple sources, is the best. Whereas comprehensive, detailed answer is useful most of the times, in some cases, such an answer needs to defeat a customer claim or demand when it is unreasonable, unfair or is originated from a bad mood. We formulate a problem of finding a defeating reply for a chatbot to force completion of a chatbot session. Defeating a reply is expected to attack the user claims concerning product usability and interaction with customer support and provide an authoritative conclusive answer in an attempt to satisfy this user. We develop a technique to build a representation of a logical argument from discourse structure and to reason about it to confirm or reject this argument. Our evaluation also involves a machine learning approach and confirms that a hybrid system assures the best performance finding a defeating answer from a set of search result candidates.
Boris Galitsky
Chapter 6. Truth, Lie and Hypocrisy
Abstract
Automated detection of text with misrepresentations such as fake reviews is an important task for online reputation management. We form the Ultimate Deception Dataset that consists of customer complaints—emotionally charged texts, which include descriptions of problems customers experienced with certain businesses. Typically, in customer complaints, either customer describes company representative lying, or they lie themselves. The Ultimate Deception Dataset includes almost 3 000 complaints in the personal finance domain and provides clear ground truth based on available factual knowledge about the financial domain. Among them, four hundred texts were manually tagged. Experiments were performed in order to explore the links between implicit cues of the rhetoric structure of texts and the validity of arguments, and also how truthful/deceptive are these texts. We confirmed that communicative discourse trees are essential to detect various forms of misrepresentation in text, achieving 76% F1 on the Ultimate Deception Dataset. We believe that this accuracy is sufficient to assist a manual curation of a CRM environment towards having high-quality, trusted content. Recognizing hypocrisy in customer communication concerning his impression with the company or hypocrisy in customer attitude is fairly important for proper tackling and retaining customers. We collect a dataset of sentences with hypocrisy and learn to detect it relying on syntactic, semantic and discourse-level features and also web mining to correlate contrasting entities. The sources are customer complaints, samples of texts with hypocrisy on the web and tweets tagged as hypocritical. We propose an iterative procedure to grow the training dataset and achieve the detection F1 above 80%, which is expected to be satisfactory for integration into a CRM platform. We conclude this section with the detection of a rumor and misinformation in web document where discourse analysis is also helpful.
Boris Galitsky
Chapter 7. Reasoning for Resolving Customer Complaints
Abstract
We report on a novel approach to modeling a dynamic domain with limited knowledge. A domain may include participating agents where we are uncertain about the motivations and decision-making principles of some of these agents. Our reasoning setting for such domains includes deductive, inductive and abductive components. The deductive component is based on situation calculus and describes the behavior of agents with complete information. The machine learning-based inductive and abductive components involve the previous experience with the agents, whose actions are uncertain to the system. Suggested reasoning machinery is applied to the problem of processing customer complaints in the form of textual messages that contain a multiagent conflict. The task is to predict the future actions of an opponent agent to determine the required course of action to resolve a multiagent conflict. This chapter demonstrates that the hybrid reasoning approach outperforms both stand-alone deductive and inductive components. The suggested methodology reflects the general situation of reasoning in dynamic domains in the conditions of uncertainty, merging analytical (rule-based) and analogy-based reasoning.
Boris Galitsky
Chapter 8. Concept-Based Learning of Complainants’ Behavior
Abstract
In this chapter, we apply concept learning techniques to solve a number of problems in the customer relationship management (CRM) domain. We present a concept learning technique for common scenarios of interaction between conflicting human agents. Customer complaints are classified as valid (requiring some kind of compensation) or invalid (requiring reassuring and calming down) the customer. Scenarios are represented by directed graphs with labeled vertices (for communicative actions) and arcs (for temporal and causal relationships between these actions and their parameters). The classification of a scenario is computed by comparing a partial matching of its graph with graphs of positive and negative examples. We illustrate machine learning of graph structures using the Nearest Neighbor approach and then proceed to JSM-based concept learning, which minimizes the number of false negatives and takes advantage of a more accurate way of matching sequences of communicative actions. Scenario representation and comparative analysis techniques developed herein are applied to the classification of textual customer complaints as a CRM component. In order to estimate complaint validity, we take advantage of the observation (Galitsky and Kuznetsov 2008) that analyzing the structure of communicative actions without context information is frequently sufficient to judge how humans explain their behavior. Therefore, because human attitudes are domain-independent, the proposed concept learning technique is a good compliment to a wide range of CRM technologies where a formal treatment of inter-human interactions such as customer complaints is required in a decision-support mode.
Boris Galitsky
Chapter 9. Reasoning and Simulation of Mental Attitudes of a Customer
Abstract
In this chapter, we employ logic programming to simulate the mental world. A Theory of Mind engine is introduced that takes an initial mental state and produces the consecutive mental states as plausible to a real-world scenario as possible. We simulate a multiagent decision-making environment taking into account intentions, knowledge and beliefs of itself and others. The simulation results are evaluated with respect to precision, completeness and complexity. Metaprogramming techniques of introspection is outlined for putting a CRM component in “customers’ shoes,” better predicting how she would think and act. We conclude that the Theory of Mind engine is adequate to support a broad range of CRM tasks requiring simulation of human mental attitudes.
Boris Galitsky
Chapter 10. CRM Becomes Seriously Ill
Abstract
This is a less technical chapter devoted to a CRM management problem of poor performance of an organization such a call center or a technical support department. We explore a technology that can detect this performance and a root cause for it, in terms of We explore the phenomenon of Distributed Incompetence (DI), which is an opposite to Distributed Knowledge and occurs in various organizations such as customer support. In a DI organization, a team of employees is managed in a way that, being rational, impresses a customer or an external observer with total irrationality and incompetence, an inability to get things done. In most cases, the whole organization or individual team members gain from DI by means of refusing customer compensation while avoiding other obligations. We investigate DI in a variety of organizations to analyze its commonality as well as specific DI features for organizations and communities. A discourse-level analysis to detect DI in textual descriptions of customers and observers is outlined. We report a detected DI rate in financial organizations and propose a solution to handle it, such as a chatbot.
Boris Galitsky
Chapter 11. Conclusions
Abstract
We draw the conclusions for Volume 1 and 2 of this book.
Boris Galitsky
Metadata
Title
Artificial Intelligence for Customer Relationship Management
Author
Dr. Boris Galitsky
Copyright Year
2021
Electronic ISBN
978-3-030-61641-0
Print ISBN
978-3-030-61640-3
DOI
https://doi.org/10.1007/978-3-030-61641-0