www.womenindata.co.uk #womenindata JCURV 1 Chatbot for social good
www.womenindata.co.uk #womenindata JCURV 1 Chatbot for social good and NLP in the wild
A little bit about me I am a Data Scientist at AI for Good UK working to improve product intelligence and communicate findings with wider stakeholder groups and the founder of the AI for Good community here in London JCURV
3 In this talk I am going to go through... Survey of chatbot technologies Main applications and some case studies of success and failure The current state-of art and the progression to such What is the future direction of this field
rAInbow our chatbot and the problems that we encountered The story behind rAInbow - why is it an innovative solution NLP work and progress Future directions AI for Social Good
JCURV 4 What is a chatbot? NLP papers classify Conversational AI into three categories: question-answering agents, task-oriented dialogue agents and chatbots.
Wikipedia says: A chatbot is a piece of software that conducts a conversation via auditory or textual methods. Such programs are often designed to convincingly simulate how a human would behave as a conversational partner, although as of 2019, they are far short of being able to pass the Turing test. Source: chatbot magazine JCURV
5 Turing Test Source: BFI JCURV
6 Some prevalent use cases of chatbots By Business Functions: Customer Service Marketing Sales HR
JCURV By Industry: E-commerce Travel Finance
7 Successful Chatbot Applications Social Chatbot Legal Chatbot Source: Nautilus | Science Connected, Crunchbase, Google Play
JCURV Mental Health Chatbot 8 Tech Stack behind Chatbot: What makes them intelligent?
Some Natural Language Processing (NLP) Lingo Intent: intentions of the end-user, these intentions or intents are conveyed by the user to your bot e.g. greeting,book_appointment Entity: person names, organizations, locations, medical codes, time expressions, quantities, monetary values, percentages. Named-entity recognition is the technique to automatically identify these entities in free text. Embedding: vectors, e.g. word embeddings
RNN, LSTM: Source: Towards Data Science Lawful Good chatbot architecture Source: Chatbot Magazine
Generative model in the wild Source: Engadget XiaoIce - a social chatbot Wouldnt be possible without advancement in NLU, Speech Recognition and Synthesis, Computer Vision, Information Retrieval, Multimodal intelligence and empathic conversational system.
Requirement of a social chatbot: Understanding users, Interpersonal response generation, Personality, Integration of both EQ and IQ Source: https://arxiv.org/pdf/1801.01957.pdf XiaoIce - continued Source: https://arxiv.org/pdf/1801.01957.pdf
Neural Approaches to Conversational AI Source: https://arxiv.org/pdf/1809.08267.pdf Neural Approaches to Conversational AI Source: https://arxiv.org/pdf/1809.08267.pdf
Neural Approaches to Conversational AI Source: https://arxiv.org/pdf/1809.08267.pdf Meanwhile standard industry toolbox -True Neutral Source: pickSaaS, WikiPedia
SUPPORTED BY A COMPANION FOR WOMEN AT RISK OF DOMESTIC VIOLENCE rAInbow Awards
I know you do this with no thought of praise or reward, but allow me to offer my own recognition of the innovative solutions you have helped to create with the power of AI. From providing mental health
support to adolescents in low income countries, to confidential chatbots that signpost support services to hundreds of thousands of women at risk of domestic abuse. Research phase - South
Africa Stigma, shame, victim blaming A nonjudgemental companion
Barriers to accessing support Lack of trust in support system
Need 24/7 support User-friendly experience and support
Hi Kriti, Im Rainbow, but my friends call me Bo Im here to talk about relationships that dont feel right. This is how youd interact with me
rAInbow user insights 350,000 conversatio ns in 8 months (Nov 2018)
6-10AM and 7-8PM peak hours on rAInbow 9AM to 5PM helpline hours Saturdays 10%
increase Sundays 15% increase Monday to Friday in conversations helpline days
Better Data Better Lives Accurate data is the lifeblood of good policy
and decision-making. - UN Secretary-General Antnio Guterres 350,000 conversatio
ns with 14,000 users Data Science insights Engaged
conversations average 60 messages average 20 minutes more than half returned What users want
personal atconversations their own pace in their own time My relationship doesnt feel right how most people describe
abuse abuse, violence, rape how services are advertised Users who engage with Bo
through AI-powered personalised stories are 40% more likely to take action than those who dont
NLP Research Challenges English proficiency of users - a variety of grammar and spelling - off-the-shelf tools do not work Specific domain - less data - low transferability
NLP Research Current Progress Unsupervised learning can only carry you so far - no free lunch in NLP Tagging is difficult - what categories, intents, entities, sentiments to tag? NLP Research Future Directions Deep learning approaches for domestic
abuse reporting message identification to enable human handover Graph-based NLP approach to cluster sentences for new topics and themes mining A few words about AI for Social Good Meetup communities: AI for Good Community, AI for Good: Element
AI, Microsoft: Impactful AI, Tech for Good London... Companies and organisations to follow: AI for Good, Element AI, DataKind UK, Alan Turing Institute... Conferences and workshops to watch: AI for social good, AI for Climate Change JCURV
Autonomy and International Involvement. Canada has been shaped by significant historical events, both within the country and at a global level. By exploring Canada's role in world events and examining Canada's development as an autonomous nation, you should come to...
All information is kept confidential and the bonus points are not revealed to the school or students. The application is assessed by VTAC and you are given a "score" for each category. This information is passed on to each university....
The Eighth Pouch / The Eighth Bolgia: The Fraudulent Counselors "I was a man of arms, and then I was one who wore the cord" Inferno 27.67 Guido daMontefeltro, is a former Ghibelline-turned-Franciscan friar who is offered "absolution" by Boniface...
Title: Introductions Last modified by: J. Cooper Cutting Document presentation format: On-screen Show Company: Personal Other titles: Times Helvetica Wingdings Arial Times New Roman Bold Stripes Introductions Your syllabus Course Format Why take research methods?
Overcoming Unconscious Bias in Letters of Recommendation. Paula Warrick, Senior Director, Office of Merit Awards, American University ([email protected]) . Kate Mamiseishvili, Associate Dean for Academic and Student Affairs, Professor of Higher Education, University of Arkansas ([email protected])
The account will remain in "delete 2" status in the FRS system until the end of the following fiscal year at which time it will be systematically purged from the system and may no longer be viewed in FRS.
Long lines at the polls. Bad weather "Time-Zone Fallout" - In Eastern/Central time-zones, we close the polls earlier than out West where they are a few hours behind. Results start getting announced and people out West get discouraged to vote....
Understanding cause and effect. Understanding the Four Noble Truths. Right view is the forerunner of the entire path, the guide for all the other factors. It enables us to understand our starting point, our destination, and landmarks to pass as...
Ready to download the document? Go ahead and hit continue!