Bye, Bye Research. Hello Data Mining!

Have a question you’d like to ask regarding today’s presentation?We welcome you to typeyour questions in the ‘Question & Answer’ window at any time during today’s Webinar. We will answer as many questions as time allows during the Q & A session following this presentation.Got Tweet? #PLData

Bye, Bye Research.Hello Data Mining!Hosted by Sean Case, SVP, Peanut LabsWednesday, March 10, 2010Peanut Labs, Inc. · 114 Sansome Street, Suite 920 · San Francisco, CA 94104www.peanutlabs.com

Presentations by:Jean Davis, Co-founder, Conversition

Catherine van Zuylen, VP, Product Marketing, Attensity

Jim Schwab, VP, Business Development – Social Media, AlterianToday’s AgendaSocial network mining and analysis

Predictive modeling and analytics

Emerging technologies in data mining

Plus more!Pecha Kucha Defined Usually pronounced in three syllables like “pe-chak-cha”

A presentation format in which one presenter shows 20 slides for 20 seconds each, for a total of six minutes and 40 seconds

Devised in Tokyo in February 2003 by Astrid Klein and Mark Dytham of Tokyo’s Klein-Dytham Architecture

Has since turned into a massive celebration, with events happening in hundreds of cities around the worldJean Marie Davis, Co-founder, Conversition Co-founded Conversition in February 2009

Formerly the President of Ipsos Online, North America

25+ years of experience in global marketing research

Known for her story telling, Jean authored The Little Church that Could, a fun and inspirational review of the signs posted outside one church for an entire year

Follow Jean on Twitter @JeanMarie50

Not Bye, Bye Research.It’s welcome Social Media Research.In the Social Network arena there is the opportunity to add social media data to the Marketing Research field.

Evolution of Research Science Marketing research techniques that assure data quality and create valuable data are very similar for each type of research – mail, face-to-face, phone, online.

Process and methods need to be developed to make social media data be another source for Marketing Research.New Data SetNew Data Collection MethodologyInstead of asking survey participants to answer questions, we listen to what social media contributors want to talk aboutApplying Research ScienceMarket research using a different data sourceResearch means:- Strict data quality processes- Norms and competitive brands- Standardized measures, both box scores and average scores- Key research measures- Category specific measures- Customized client measures- Sampling and weighting

Creating the ProcessCreate SearchCleanCrawlCleanSampleWeightScoreContent AnalysisSpecifywhatclient wantsto measureIdentifyrelevantconversationsIdentify conversations that do notmeet basicquality control requirementsTieredsystem reflectinguniqueneeds ofdifferentdata sourcesContent analysis is applied to every conversationSampling is used to identifywhich sources are appropriate for a clientWeighting is applied to the sampling matrix to ensure that the included sources are reflected in a consistentproportion over time

Sample SizesSocial media presence of Client Brand A and C, and Competitive A, B, and C are very good and well suited to social media research.

Social media presence of Client Brand B is extremely low and may not be suited for quantitative research at this time. DemographicsSocial media contributors do not share their demographic information when they contribute online but we do know the demographic make-up of many popular social media websites including twitter, flickr, and blogger.

People talking about this brand are more likely tobe womenbe aged 35 to 64have a college degreeearn $25k to $75k

Content AnalysisA method of grouping similar Conversations together so that they can be evaluated as a whole.

Retailers: Parking, check-out lines, categories (electronics, apparel)

CPG: taste, feel, product, price

Determine which sets of conversations are similar to each other based on tone of voice and content of the conversation.

Sentiment: positive/negativeSampling & WeightingSources can be sampled and weighted according to the distribution of internet categories

Can be weighted to redistribute sample so that overrepresented categories are less likely to skew the dataReportingData can bring results in familiar data reports.

Data can bring results in new data reports.

PsychographicsMultiple Brand ComparisonSentiment and volume of chatter were tracked beginning from September 1, 2009

Brands with the most positive sentiment include Brand A, Brand G, Brand H, and Brand N.

Brands with the most chatter include Brand B, Brand J, and Brand LPast 30 daysn = 378 to 92,000

Retailer Attribute ComparisonEmployeesCrowdingParking LotAverage Scores5.0 = Positive3.0 = Neutral1.0 = NegativeNorms4.0 = High3.3 = Normal3.0 = LowWebsiteHoursWashroomsRadar maps allow one to evaluate multiple brands on multiple constructs in one single chart. Brands with the largest web, or circle, are viewed the most positively by consumers. In this case, constructs relevant to retailers have been selected to compare Brand Green retailer with Brand Grey retailer.

Scores are most positive in relation to crowding, the parking lot, and the hours. On the other hands, scores are much lower for opinions of employees and the website.

While Brand Green outperforms Brand Grey on nearly every construct. However, Brand Green and Brand Grey generate very similar opinions related to their websites.CloudsData clouds indicate the specific words and phrases that people use in their conversations

Popular words indicate:- The interests of people talking about the brand, and therefore the contents of marketing materials- Co-branding and co-sponsorship opportunities that are relevant to your consumers- Appropriate language to use in marketing materials, whether slang or formalUse tennis or football metaphorsShow basketball or football in marketing materialsObtain tennis or football celebrity endorsements

PsychographicsDespite the fact that Brand A and Brand B generate similar emotion scores, by reviewing the assortment of constructs and identifying those with greater and lesser frequencies, psychographic differentiators of brands can be discovered

The first three constructs are revealing in that each word relates to the exact same idea. However, the words used among Brand A consumers are more intellectual.

This trend follows through in the discussions of technology where Brand A consumers use more technical words.

Income and schooling also reflect a higher socio-economic status for Brand A consumers

Brand A consumers reflect a higher socio-economic status than Brand B consumersThe EndSay “Hello” to Social Media Research- New data collection methodology-Create a process from data collection to reporting- Apply research techniques to the data to create a valid, valuable, actionable data set- Create new and familiar reports- Continue to validate and improve processes

March 10, 2010Thank you to Peanut Labs for inviting Conversition to share in their webinar!Jean Davis, jean@conversition.comMarch 10, 2010

Any questions for Jean?We welcome you to type your questions in the ‘Question & Answer’ window at any time during today’s Webinar. We will answer as many questions as time allows during the Q & A session following this presentation.

Catherine H van Zuylen, VP, Product Marketing, Attensity A consultant at The Grommet Group

Formerly Vice President of Marketing at Block Shield

20 years of experience in product management, product marketing and marketing communications

A Silicon Valley native who grew up across from an apricot orchard and won several blue ribbons at the country fair for her fruits and vegetables

Follow Catherine on Twitter @catevzLeveraging Customer Conversations Through LARACatherine H van ZuylenVP, Product Marketingcvanzuylen@attensity.comwww.attensity.comTwitter: @attensity

Attensity: Over 20 years experience understanding customer conversations in text; 6 patents in natural language processingSuite of applications for social media monitoring, Voice of the Customer Analysis, and Self-Service/Agent ServiceOver 500 customers worldwideMe: 15 years in marketing; 10+ years in text analytics and internet mediaA Few Words About Me and Attensity

“Customer Information” is changing and growing exponentiallyTwitter hit the 10 billion tweet mark last week : over 20% are about products and servicesOver 247 billion emails are sent every dayMillions of customer interaction records in a typical large company.

To effectively harness these “customer conversations”, you need a program to comprehensivelyListen across customer conversation channelsAnalyze accurately and efficientlyRelate this information to other informationAct on the informationWe call this the LARA methodology

LARA Methodology: Listen, Analyze, Relate, ActAre you listening where your customers are talking? Are your “social media” listening efforts isolated from your “CRM” listening efforts and separate from your “survey” listening? Are you monitoring your internal customer communities?Text Analysis can help bridge these gaps.

Text Analysis is not Search“Search” is for finding relevant or recent documents that contain a term of interest

But it’s hard with search to get the “big picture”What do people think about my company?What problems are they having?What do they like about me vs. the competition?What new ideas do they have?Who is thinking of switching?34

“Search” starts with you feeding a system words to look for. “Text Analysis” starts with the data itself and lets it tell a storyDynamic Text ProfilingDocumentsEntities, sentiments, events and relationships, intent, etc?XML or other “tags”

Text Analysis starts the same way some search engines do…Automatic Language and Character Encoding Identification Identify paragraphs and sentences within textWord Segmentation (Tokenization) and De-CompoundingPart-of-Speech Tagging Stemming Noun-Phrase Identification

Then continues with Entity Extraction…Who: People, Person Position, Social Security NumbersWhat: Companies, Organizations, Financial Indexes, Products (software, weapons, vehicles, etc…)When: Dates, Days, Holidays, Months, Years, Times, Time PeriodsWhere: Addresses, Cities, States, Countries, Facilities (stadiums, plants), Internet Addresses, Phone Numbers How Much: Currencies, MeasuresConcepts (i.e. Global piracy, unstructured data…)Can be pattern-based – tell the system that a “Prop-Noun followed by Smith” is probably a personOr machine learning – feed it a million proper names and let it deduce names from those examples…

Practical Text Analysis in ActionLet’s say that I am a major retailer, and someone posted a review that starts outI bought this Gucciscarffor my mom in your Santana Row store last week. Entities (brands, people, locations, times, products…)

To “connect the dots” in data, you also need to extract noun-verb relationships, sentiment…I bought this Gucci scarf for my mom in your Santana Row store last week. I really like the pattern, but I don’t like how it itches.Entities (brands, people, locations, times, products…)Events and relationships: action and purchasing reasonSentiment (extreme positive, positive, negative, extreme negative)

To “connect the dots” in data, you also need to extract suggestions, intent…I bought this Gucci scarf for my mom in your Santana Row store last week. I really like the pattern, but I don’t like how it itches. I wish this scarf came in cotton. If Gucci made more cotton scarves, I would buy them all.Entities (brands, people, locations, times, products…)Events and relationships (I : buy : this Gucci scarf | I : buy : for mom)Sentiment (extreme positive, positive, negative, extreme negative)Suggestions (I : wish : this scarf came in cotton)Intent (to purchase, to leave) (If Gucci made more cotton scarves, I would buy them.)

How do you do this? You parse sentences like a human…and extract triples…

…and voices (intent, recurrence, etc)Question [?] voice:How can I get free shipping with future orders? Condition [if/then] voice:.I would shop more frequently if you offered free shipping. Intent [intent] voice:I plan to place an order today. Negation [not] negates the meaning of the verb:You did not have the size I was looking for in stock

…and voices (intent, recurrence, etc)Question [?] voice:How can I get free shipping with future orders? Condition [if/then] voice:.I would shop more frequently if you offered free shipping. Intent [intent] voice:I plan to place an order today. Negation [not] negates the meaning of the verb:You did not have the size I was looking for in stock Augment [more] voice:The staff were incredibly professional Recurrence [again] voice:I had to enter my information several times for the order to process Indefinite voice representing suggestions or requests.You should sell wedding dresses, too!

LARA Methodology: Listen, Analyze, Relate, ActOnce you’ve done text analysis, you can relate the text to structured information…01/24/2010By errodd from San Jose, CAI bought this Gucci scarf for my mom in your Santana Row store last week. I really like the pattern, but I don’t like how it itches. I wish this scarf came in cotton. If Gucci made more cotton scarves, I would buy them all.Can help you answer questions likeWhat were the top concerns of people who rated this product a “4”?

LARA Methodology: Listen, Analyze, Relate, Act: What Can You Do with Text Analysis?The output from text analysis can be exported as XML…It can also be used directly in applications thatSeek out and deliver information to those who need itRoute and respond to communicationsMine and report on information

“Seek Out” information for a self-service knowledgebaseProblemSolutionManufacturer: AppleProduct: Macbook, Projector, MonitorComponent: Adapter cord, Mini-DVI, VGAAction: Do a presentation, connect

Route and respond to all customer communicationsResponses can be reviewed by agent before sending“refund policy” email response auto-generatedRead text and extractknowledge about what the document is sayingPeoplePlacesEventsTopicsSentiment …Refund policy? EmailRouted to Customer Service for Follow-up and Resolutionintent to leave tweetAutomatically routed as a mobile alert to legal for reviewThreatening to sue posting

Mine and report on sentiments, complaints, compliments, and “intentional” behavior across all customer conversationsBetter understanding their customersBetter understanding their customers and gain early warning on product issues

Thank You.Leveraging Customer Conversations Through LARACatherine H van ZuylenVP, Product Marketingcvanzuylen@attensity.comwww.attensity.comTwitter: @attensity

Any questions for Catherine?We welcome you to type your questions in the ‘Question & Answer’ window at any time during today’s Webinar. We will answer as many questions as time allows during the Q & A session following this presentation.

Bye, Bye Research. Hello Data Mining!

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Bye, Bye Research. Hello Data Mining! (20)

Recently uploaded (20)

Bye, Bye Research. Hello Data Mining!

Editor's Notes