31
Monitoring social media the “smart” way. SETTING UP A MACHINE LEARNING PLATFORM

Setting up a Machine Learning Platform - Monitoring social media the “smart” way

Embed Size (px)

Citation preview

Monitoring social media the“smart” way.

SETTING UP A MACHINE LEARNING PLATFORM

TODAY’S EXAMPLES

● “Classic” learning — Social media monitor○ 10xnation.com/social-customer-care-amazon-machine-learning

● “Deep” learning — Extending the social monitor○ 10xnation.com/social-customer-care-ibm-watson

● Gathering data — Website traffic○ https://10xnation.com/wordpress-analytics

“CLASSIC” LEARNING

Example #2: A social media monitor using Amazon Machine Learning

THE PROBLEM

Too many social media posts to track and read.

Many of our customers/prospects are feeling neglected because we don’t have the resources to

read and respond to all of them.

Need a way to filter them down to only the ones where the sender is expecting a response.

THE QUESTION

Is this tweet actionable?

DETERMINING THE ANSWER

Is the sender…

● Making a request● Asking a question● Reporting a problem● Angry or Unhappy● None of the above

Actionable?

Yes (1)Yes (1)Yes (1)Yes (1)No (0)

THE SOLUTION

Use Amazon Machine Learning to analyze a Twitter stream in real-time and make a determination about

whether or not a tweet requires a response. (binary classification: yes or no)

Then route the positives to a customer service agent.

Based on: github.com/awslabs/machine-learning-samples/tree/master/social-media

Speechto Text

Sentiment Analysis

Actionable Analysis

Customer Support

PREDICTIVE ENGAGEMENT

Customer support call recordings

Convert audiointo text

Analyze formood keywords

Determine ifresponse is required

Reach out to customer/prospect

Blog & community comments

Social media mentions

Press & blog coverage

Customer support chat

Product reviews

Inbound emails

BREAK IT DOWN

Twitter API

Mechanical Turk

Amazon Kinesis

Amazon Machine Learning

Amazon Lamda

Model

Amazon SNS

Customer Service

Labels training data

Responds to tweets

Forwards “actionable” tweets to support team

Captures Twitter stream

Relays tweets between Kinesis, ML & SNS

Classifies tweets as “actionable” or not

END RESULT

Your staff doesn’t have to read each tweet, andyour customers feel appreciated and happy.

THE FEATUREScreated_at_in_secondsdescriptionfavorite_countfavoritedfavourites_countfollowers_countfriends_countgeo_enabledin_reply_to_screen_namein_reply_to_status_idin_reply_to_user_idlocationr.created_at_in_secondsr.descriptionr.favorite_count

r.favoritedr.favourites_countr.followers_countr.friends_countr.geo_enabledr.in_reply_to_screen_namer.in_reply_to_status_idr.in_reply_to_user_idr.locationr.retweet_countr.screen_namer.sidr.statuses_countr.textr.time_zone

r.uidr.user.namer.utc_offsetr.verifiedretweet_countscreen_namesidstatuses_counttexttime_zoneuiduser.nameutc_offsetverifiedtrainingLabel

Warning: Live social media content.

10xnation.com/social-customer-care-amazon-machine-learning

STEP BY STEP GUIDE

● Step 1: Requirements ● Step 2: Gather training data ● Step 3: Prepare raw tweets for labeling● Step 4: Submit job to Mechanical Turk ● Step 5: Format labeled data ● Step 9: Upload training data to S3

● Step 7: Generate the Model ● Step 8: Configure Machine Learning ● Step 9: Configure Kinesis ● Step 10: Configure IAM ● Step 11: Configure SNS ● Step 12: Configure Lambda ● Step 13: Configure Twitter ● Step 14: Fire it up

“DEEP” LEARNING

Example #2: A social media monitor using IBM Watson

EXTENDING THE SOCIAL MONITOR

Let’s make our new social media monitor even better…

● Wrap a UI around it● Pre-populate a tweet response● Categorizes topic of each tweet● Determine sentiment of each tweet● Provide insight into personality of sender

THE SOLUTION

Use IBM Watson to analyze a Twitter stream in real-time and determine…

● Sentiment● If response required● Type of response required

Based on: github.com/watson-developer-cloud/social-customer-care

Speechto Text

Sentiment Analysis

Actionable Analysis

Customer Support

PREDICTIVE ENGAGEMENT

Customer support call recordings

Convert audiointo text

Analyze formood keywords

Determine ifresponse is required

Reach out to customer/prospect

Blog & community comments

Social media mentions

Press & blog coverage

Customer support chat

Product reviews

Inbound emails

Twitter API

AlchemyAPI

Responds to tweets

Customer Service

User Interface

Model

Personality Insights

Tone Analyzer

Natural Language Classifier

Analyzes sender’s prior tweets to estimate their

personality

Sentiment analysis of tweet stream

Classify topics in tweet stream

Analyzes sender’s prior tweets to determine common topics

END RESULT

A “smart” application that streamlines your customer service processes on Twitter.

Warning: Live social media content.

STEP BY STEP GUIDE10xnation.com/social-customer-care-ibm-watson

● Step 1: Requirements ● Step 2: Configure Natural Language Classifier ● Step 3: Configure Alchemy Language ● Step 4: Configure Personality Insights ● Step 5: Configure Tone Analyzer ● Step 6: Configure Twitter ● Step 7: Train the Natural Language Classifier ● Step 8: Create the application ● Step 9: Fire it up

ENDLESS POSSIBILITIES

● Give customer service agents a way to provide feedback on the system’s accuracy

● Capture the agent’s feedback and tweet responses● Use new data to further refine prediction accuracy● Automate more and more as system gets “smarter”

GATHERING DATA

Example #3: Capturing website traffic data using WordPress

THE PROBLEM

To make accurate predictions and insights, we need data. The more, the better.

But most of us don’t have much data today.

THE QUESTION

How can we gather more data from our website,so we can better understand our visitors

— and predict their behavior.

THE SOLUTION

We’ll use WordPress for this example.

And we’ll capture as much data from it as we can.

THOUGHTS?

Hurdles looking easier to navigate?

UNLEASH YOUR BUSINESSEMBRACE EXPONENTIAL

10xnation.com