Zendesk: Sentiment and Intent With AL

Powerful direction,  How well?

September 15, 2022

Zendesk Launches New Customer Sentiment and Intent Functionality Powered by ML

SAN FRANCISCO, Sept. 15, 2022 — Zendesk, Inc. today announced Intelligent Triage and Smart Assist, new AI solutions empowering businesses to triage customer support requests automatically and access valuable data at scale. By democratizing access to these solutions, companies can see value in minutes by understanding intent and sentiment through account-specific, data-driven models that are customized for individual use cases and drive faster resolutions.

Industry analysts predict that in the near term, AI will touch the majority of customer service interactions, but Zendesk research shows less than a third of companies are currently using AI to help their service teams become more efficient. That’s in part because even as AI technology has rapidly improved in the last five years, software vendors haven’t yet passed those improvements on to their customers and are still selling expensive AI solutions that are extremely time-consuming to set up.

Intelligent Triage and Smart Assist are the next step in Zendesk’s vision to create accessible CX AI for companies of all sizes. The technology uses proprietary industry expertise and insights from trillions of customer data points and applies a vertical lens. This creates models custom to each business capable of identifying the intent, language and sentiment of each customer interaction.

This unique approach to applying machine learning creates more personalized and informed interactions to better serve customers. For example, specific inquiries, such as “I’m having problems with payment”, can be automatically sent to an agent who is equipped to handle billing for a quicker resolution, while inquiries that include language written in all capital letters or in a sarcastic way will indicate a highly negative sentiment and be routed to the top of the queue. ... ' 

Comparision of Sentiment Analyses Methods

An approach we coded long before neural methods.    Here an excellent look at current methods.  Pointers to code.

Sentiment Analysis — Comparing 3 Common Approaches: Naive Bayes, LSTM, and VADER

A Study on Strengths and Drawbacks for the Different Approaches (With Sample Code)

By Kevin C Lee

 Sentiment Analysis, or Opinion Mining, is a subfield of NLP (Natural Language Processing) that aims to extract attitudes, appraisals, opinions, and emotions from text. Inspired by the rapid migration of customer interactions to digital formats e.g. emails, chat rooms, social media posts, comments, reviews, and surveys, Sentiment Analysis has become an integral part of analytics organizations must perform to understand how they are positioned in the market. To be clear, Sentiment Analysis isn’t a novel concept. In fact, it has always been an important part of CRM (Customer Relationship Management) and Market Research — companies rely on knowing their customers better to evolve and innovate. The more recent rise is driven largely by the availability/accessibility of customer interaction records and well as improved computing capabilities to process these data. This advancement has really benefited consumers in meaningful ways. More than ever, organizations are listening to their constituents to improve. There are numerous approaches for Sentiment Analysis. In this article, we’ll explore three such approaches: 1) Naive Bayes, 2) Deep Learning LSTM, and 3) Pre-Trained Rule-Based VADER Models. We will focus on comparing simple out-of-the-box version of the models with the recognition that each approach can be tuned to improve performance. The intention is not to go into great details about how each methodology works but rather a conceptual study on how they compare to help determine when one should be preferred over another. .. "

Machine Learning and Intent

Deriving intent,  goals is still hard.  Especially in multiple component conversations.

Machine Learning, Natural Language Meet to Understand Intent     By Mark Jewett, VP, product marketing, Tableau Software in InformationWeek

Machine learning and natural language capabilities will bring the power of analytics to more people through semantics.

Machine learning and natural language processing promise to better translate human curiosity into pertinent answers. If true, these smart capabilities will broaden the use of analytics and reach people who are less comfortable dealing with data. It will all start with helping machines learn to interpret human intent. The key is semantics.

Sometimes intent is simple and explicit, like asking Siri or Alexa if a flight is delayed. This question has clear intention and a simple response -- returning the flight status answers the question. Such simplicity is seldom the case when it comes to data analysis. Questions are usually more nuanced, making it hard to correctly assume what the user is really looking for. Natural language is even more tricky where ambiguous terms are common.

It’s also difficult for a machine to understand our intent within a limited context. The machine has the data itself but doesn’t grasp the bigger picture in the same way a person with domain expertise can. Asking “How are my sales doing in the Northeast?” is a lot more ambiguous than the flight status example above..... "

Automated Sentiment Analysis Does not Work?

A classic analysis that points to indicators of behavior.   

Research findings send automated sentiment analysis to the trash bin.

The age of social media has opened up exciting opportunities for researchers to investigate people’s emotional states on a massive scale. For example, one study found that tweets contain more positive emotional words in the morning, which was interpreted as showing that most people are in a better mood at that time of day. ... 

".....But now a new study has thrown a spanner in the works, finding that – for spoken language at least – this assumption might not hold up. In their preprint posted recently on PsyArxiv, Jessie Sun and colleagues found that emotion-related words do not in fact provide a good indication of a person’s mood, although there may be other sets of words that do. .... "

Preprint of paper:    https://psyarxiv.com/8fpjn/

Sentiment Analysis and Merchandising Calls

Quite a range of information is described. 

Can sentiment analysis improve merchandising calls?   by Brian Kilcourse in Retailwire

Through a special arrangement, what follows is a summary of an article from Retail Paradox, RSR Research’s weekly analysis on emerging issues facing retailers, presented here for discussion.

Marketers have long used sentiment analysis to target offers in the digital domain, but a new generation of demand forecasting capabilities is extending its use into merchandise planning.

Sentiment analysis examines “non-transactional” data that consumers leave along their digital paths-to-purchase that can offer an early indicator of demand. A simple example is the number of positive mentions on social or e-mail streams about a new offering. Online clicks, geo-location info and other digital data can also offer clues as to whether underlying sentiment is positive, negative or neutral. ... " 

Conversations in Assistants with Mood

Interesting points are made.  Though the analysis of mood from text is fairly straightforward, wondering if there are liabilities once one classifies mood in real time?  Good interview:

Google Home's Assistant could one day know your mood. Take that, Alexa in CNet  By Richard Nieva

Google Home's Assistant could one day know your mood.  
In an exclusive interview, a top Google exec says the AI could eventually recognize if you're frustrated or pick up conversations where you left off.  ... " 

AI Lie Detectors at some EU Borders

Could indicate increasing uses of AI to determine sentiment style interaction and detection.

An AI Lie Detector Will Interrogate Travelers at Some EU Borders 
 in New Scientist (full article requires sign up)
By Douglas Heaven

An artificial intelligence (AI)-based lie detector system will be used on international travelers after they have passed through border control, during a six-month pilot program at four border crossings in Hungary, Latvia, and Greece. The Web-based tool is designed to make crossing into the European Union quicker and safer, while identifying likely lawbreakers when entering a country. During the interview stage of iBorderCtrl, the system asks questions via a virtual border guard on a laptop or mobile device; as the traveler provides answers, the device's camera films the user's face. AI software then analyzes the video and examines 38 micro-gestures to identify patterns that could indicate lying, such as slight eyelid movements. In a test of the system on 30 people (half were told to lie, the other half to tell the truth), the software was able to identify liars with around 76% accuracy. One of the system’s developers, Keeley Crockett of Manchester Metropolitan University, U.K., said, “We’re quite confident of bringing it up to the 85% level.” .... '

Talking about and Measuring Emotions

Way back when we worked with MIT on measuring emotions when people interacted with our products, we discovered the difficulty of measuring this consistently.   Later when we looked at 'neuromarketing' methods, things were not much better.   A number of machine learning methods now claim to make this easy.  Consideration of some derived sentiment based on emotion.  A neuroscientist looks at the issue of naming emotions:

A neuroscientist explains why we need better ways to talk about emotions
We still don’t know what an emotion is
By Angela Chen   @chengela in the Verge  ... " 

See also, Sentiment Analysis.

Chat Apps with Sentiment

Nice example, technical.  Most interesting is the introduction of sentiment.   We did this for 'Mr Clean', before there were easily integrated sentiment analysis.   Recently have been reexamining sentiment considerations in chat dialog.

Build A Chat App With Sentiment Analysis Using Next.js  in Pusher
Build a realtime chat application with Pusher

This tutorial was written by Chris Nwamba and originally appeared on the Pusher Blog.

Realtime applications have been around for quite a long time as we can see in contexts such as multi-player games, realtime collaboration services, instant messaging services, realtime data analytics tools, to mention a few. As a result, several technologies have been developed over the years to tackle and simplify some of the most challenging aspects of building apps that are sensitive to changes in realtime.

In this tutorial, we’ll build a very simple realtime chat application with sentiments. With sentiment analysis, we will be able to detect the mood of a person based on the words they use in their chat messages.  .... "

Sentiment Analysis Uses and Comparisons

A look at various sentiment applications and services.    Have used a couple of these.

Machine Learning as a Service: Part 1  from Towards Data Science. By Sebastian Kwiatkowski

Sentiment analysis: 10 applications and 4 services
What is sentiment analysis?

The explosive growth in user-generated content and the digitization of archive material have created massive data sets containing opinions expressed by large numbers of people on just about every single topic.

In some cases, the generation of this data is structured through the user interface. It is, for example, relatively easy to process customer reviews on e-commerce sites, because users are required to post a rating alongside the text of the product review.

Most data, however, is available in an unstructured form. It does not contain a standardized summary saying “This content expresses a positive, negative, mixed or neutral view.”

WordPress.com, for example, reports that bloggers using their platform have published more than 87 million posts just in May of 2018.[1] According to YouTube CEO Susan Wojcicki, more than 400 hours of content are uploaded to the video-sharing site every minute.[2] Meanwhile, the Google Books project has digitized at least 25 million volumes in 400 languages.[3]

Whenever a user types into a free text field or speaks into a microphone, an inference is required to categorize the sentiment.

Sentiment analysis is the field that focuses on exactly this task. It is a branch of natural language processing that studies functions designed to map a text document to a representation of a sentiment.

With the advent of accurate speech and text recognition, the reach of sentiment analysis extends beyond readily accessible digital text data and covers an increasing number of media. ... "

AVA Does Customer Service

As part of a look and conversation regarding conversational agents for both engagement, loyalty and enhanced customer service for Autodesk.  Does such a solution have to look like a human?   Note below Watson services are being tested to model customer sentiment.   Videos at the link.  Thanks to Walter Riker. 

This Chatbot Is Trying Hard To Look And Feel Like Us    In FastCompany

Modeled on a real person and equipped with a virtual “nervous system,” Autodesk’s AVA is built to be a font of empathy, no matter how mean a customer gets.without the nicotine/content.
Among the attributes credited for Apple’s famous customer loyalty is a network of stores where curious or frustrated consumers can meet the company face-to-face.

The 3D design software maker Autodesk is trying to achieve something similar online with a help service that allows people to interact with what sure looks like an actual human. The company says that next year it will introduce a new version of its Autodesk Virtual Agent (AVA) avatar, with an exceedingly lifelike face, voice, and set of “emotions” provided by a New Zealand AI and effects startup called Soul Machines. Born in February as a roughly sketched avatar on a chat interface, AVA’s CGI makeover will turn her into a hyper-detailed, 3D-rendered character–what Soul Machines calls a digital human. ... " 

Emotion Detection using Machine Learning

Emotion detection regarding product impressions was a long time goal of our research.

Emotion Detection Using Machine Learning
Pulling out context from the text is one of the most remarkable procurements obtained using NLP. A few years back, context extraction was to detect the sentiment from the text and then the definition took a step forward towards emotion detection. These two are very different terms. The sentiment can be positive, negative, neutral while emotions are more refined categories among these three. A positive sentiment could be attributed to happy, excited and even a funny emotion. Similarly, anger, disgust, and sad emotions make the sentiment negative. ... " 

Via http://www.paralleldots.com/

Sentiment Analytics Master Class

Quite a lengthy, complete and interesting view I am looking at.  In Linkedin.

Sentiment Analytics Triology - A master class on extracting sentiments or aspects from data

Published on September 28, 2016 Featured in: Customer Experience, India, Technology ... A master class on extracting sentiments or aspects from data by Snehamoy Mukherjee, Data Science & Analytics Leader

Credit goes to my colleagues Sumit Pratap Singh, Prateek Sharma, Kuntal Basu and Subhamoy Ganguly for delving deep into this field and creating an outstanding capability in this much talked about field. The article below is the first of a series of articles, which we intend to publish over the course of the next few days, and is an attempt to articulate the wisdom and share it with a wider audience, so that others can benefit from it. Hence, I have called it a master class as there is a touch of pedagogy in this attempt to carve out the essence of a complex topic.

Introduction

One of the areas of analytics/data science that has generated a huge amount of hype and interest of late, in the industry has been “sentiment analytics”. And yet, interestingly, though a lot of research and investment has gone into it, there haven’t been many takers for it from the point of view of business users. One of the major reasons for it has been that sentiment analytics has to deal with a lot of noise in the data, which the conventional, automated, “plug and play” solutions can’t handle. A lot of folks in the industry tried to come up with tools and products and trivialized “sentiment analytics” into a product play and hence, nipped an emerging industry in the bud. 

A well-executed sentiment analysis needs a lot of human intelligence to interpret the results as well as to coach the self-learning algorithms to do the right things. In this series of three papers, which we can call the “Sentiment Analytics Trilogy”, we provide a successful approach to performing sentiment analysis that gives outperforms the droves of so called “sentiment analytics products and tools” in the market, which are mostly unstable and error prone.  ... " 

Mood Science and the Internet of Things

Anyone who works with advertising and marketing ideas wants to measure and influence aspects of mood in  a number of context.   Much has been made of neuromarketing ideas in recent years.   'Things'  that can  influence mood could be packaging, product, context, promotion, shelf and fixture.

And things that influence, can also measure and interconnect.   Measuring means data you can leverage it.  We mocked up a number of related ideas, starting with the simplest RFID tags.   Consider links with Beacons too.    Lots of interesting thoughts in the article linked to below.

These IoT Sensors Want to Know How You Feel – And Maybe Even Change Your Mood  Posted by William Vorhies  

Summary:  Sensors that know how you feel?  Sensors that want to change the way you feel?  When did that happen and better yet how?  ... 

Mood Science
" ... I’ve been tracking the uses of IoT sensors particularly those with human interaction (think Fitbit) but I didn’t see the big picture until I came across this article “Design for Mood: Twenty Activity-Based Opportunities to Design for Mood Regulation” by Pieter M. A. Desmet, a member of the Faculty of Industrial Design Engineering, Delft University of Technology.  This is one of those articles you know you should trust because it contains a reference bibliography of 169 learned articles.

For the most part it seems that in academic circles the desire to determine how to ‘regulate mood’ is pretty benign and generally couched in terms like improving subjective well-being.  After all who doesn’t want an extra helping of well-being? .... " 

Emotion Analysis API

More companies are looking at text analysis ....

AlchemyLanguage Emotion Analysis API is Generally Available, and It’s Getting Better

Many in the field of Cognitive AI research and development speak of the importance of context. Context could be visualized similar to that of an onion, with multiple levels of nested, related and non-related context. But perhaps one of the most important layers is Emotional context, as it has the power to transform dynamic decision making internal to the intelligence.”

—Brennon Williams, Chief Executive Officer & Founder of Iridium Systems and Robotics Corporation

On July 1st, 2016, the Emotion Analysis capability in AlchemyLanguage became Generally Available for production use. Now, with our latest updates, you can use Sentiment & Emotion Analysis to understand social data at a deeper level than ever before.

AlchemyLanguage users take their Sentiment Analysis one step deeper to detect five distinct emotions in text – joy, fear, sadness, anger, and disgust. Users employ our sentiment and emotion capabilities to discover emotional trends in social media, prioritize inbound social data, and more. ... " 

Apple, Intel and Realsense

Some time ago I looked at the use of Intel's Realsense.  But had not heard anything from that direction for a long time.  The potential application used interactive screens.  Has Apple stopped the work?

" .... Intel has grand plans for computers that will recognize human emotion using its RealSense 3D camera, but Apple appears to have dealt it a setback.

RealSense uses a combination of infrared, laser and optical cameras to measure depth and track motion. It's been used on a drone that can navigate its own way through a forest, for example.

It can also detect changes in facial expressions, and Intel wanted to give RealSense the ability to read human emotions by combining it with an emotion recognition technology developed by Emotient. ... " 

Detecting Funny

The emergence of a New Yorker crowdsourcing system that tries to determine, with interaction and algorithm, what is funny.  In CNet.  Seems this could also be linked to Cyc and perhaps Lucid, to determine the subtlety of our knowledge about our world.    You can join in the crowd sourcing process here 

A Look at Text Classification

From an introduction, to simple methods, to depth  This was an area we used extensively for 'content' analysis to understand consumer responses.  Most recently have looked at it as a method of classifying written comments in compliance documents.  New methods in machine learning have improved it considerably.

Text Classification & Sentiment Analysis tutorial / blog   
Posted by Ahmet Taspinar  

We all know that with Machine Learning you can automatically classify text documents or analyze its subjectivity. We've just released a guide that gives a brief introduction to Text Classification. 
It cover the three most used classifiers; Naive Bayes, Maximum Entropy and Support Vector Machines and will give practical examples in the form of the sentiment analysis of book reviews.  ... "
 

Gathering Television User Sentiment

Gathering Television Viewer Sentiment ... Case study: Learn how Hong Kong–based TVB uses social media to analyze audience sentiment    ... " 

Text Signals as Sentiment

An area that a number of groups are looking at.

" ... In this burgeoning era of big data, a substantial majority of all the data is unstructured. Much of this unstructured data is textual, such as the data in reports, articles, emails, tweets, and even conversations or support calls recorded in textual transcripts. Because some of this information is perishable, the capability to process it quickly—in many cases, in real time or near-real time—is becoming quite important to enterprises. This processing requires text analytics capabilities.

What does the capability to perform text analytics mean? One simple example is processing a social media feed, such as Twitter, to extract any tweet that mentions a specific element of data such as a company name or a product. This simple approach to matching keywords can provide a quick glimpse into the presence of a product name in the public’s mind. Further, using Twitter-based metadata, for example, creates the possibility to divide this public perception into regions. ... "