2. N
i
c Sophie Van Neck
e Research Manager @ InSites
Consulting
t sophie@insites.eu
o @sophievanneck
http://be.linkedin.com/in/sophievanneck
m
e
e
t
y
o
3. Fact sheet
Spin-off of top-ranked business school
15 years of experience and know-how
Pioneer and innovator in online methods
Covering any marketing domain
Fully independent
Ghent, Rotterdam, London, Timisoara, New York
125 passionate employees
Proprietary research panel in +25 countries
Most awarded agency by ESOMAR
4. The power of client data
Information!
Both structured as well as
unstructured
-website contact forms
-hotel evaluation forms
A lot of textual information that
goes to waste
6. Culture of interviewing
versus culture of caching
Research: asking for content
Content that might already be
available
Embrace and use what is out
there
11. STEP 1: extraction
SPSS Software wants to make software COOL again: well-known, popular and unique
Software detects terms based on
several dictionaries
Count based Linguistische analysis
wants
to
make Part of speech analysis
marketing
research software
COOL COOL
again Well-known
Well-known popular
popular unique
unique
Also add your own terms
20. Sentimeter
Decision Friends
Safety
Religion
Active
Eating
Diagnosis
Transpor professional
t Clothes
Dog Housing
Personal Payment
care Law
Identity
Treatment
Guilt Telephone
call
Buzz volume
21. Eating = enjoying life
Health problems Appetite decreases
Chewing Altered taste
Swallowing
24. APPLICATIONS
versus
DISCOVER MEASURING
BRAND PERFORMANCE
25. Share brands in total amount of conversations
Q: To what extent do people mention brands in the online conversations about ice cream?
Total sample = 1327 About brands = 327
26. Sentiment & performance analysis
Average rating
Emotionality % Positive % Negative
review sites
0,73 68% 28% 2,69
N = 101
0,90 81% 52% 1,75
N = 48
27. Sentiment & performance analysis
Average rating
Emotionality % Positive % Negative
review sites
0,90 81% 52% 1,75
N = 48
• Strengths
– Breyers has a strong tradition. Their vanilla & chocolate flavours are legends and they
always had a strong fan base. People have a lot of trust in the brand
• Weaknesses
– Recently, consumers report a decrease in quality
• Taste
• Texture
• Natural
– Too expensive
28. APPLICATIONS
versus
DISCOVER MEASURING
PRODUCT PERFORMANCE
29. Evolution YourBrand & Competitor conversations – Launch period
Evolution YourBrand & Competitor conversations – Launch period Sentiment (positive MINUS negative)
Sentiment (positive MINUS negative)
600 Your
The launch of PLATFORM created a PLATFORM was discussed with low Brand Y Platform
brand
peak in YourBrand conversations on Launch emotionality, but on average in a
500 17/12 and 20/12. However, the positive way. Overall, YourBrand’s
conversations faded out quickly online conversations became more
afterwards. In January, there was a positive, but this was a result of other 1
,
4
1
1
,
4
1
1
,
4
1 % Positive
400
new small peak when announced conversation topics becoming more ,
2
1
,
,
2
1
,
,
2
1
,
something new, and users hoped it positive (CSR, jobs & campaigns), not
0 0 0
0 0 0
32%
, , ,
8 8 8
0 0 0
would be a new premium membership. so much because of PLATFORM. ,
6
0
,
,
6
0
,
,
6
0
,
4 4 4
0 0 0
, , ,
300 2
-
0
2
-
0
2
-
0
0 0 0
, , ,
- - -
2 2 2
0 0 0
, , ,
- - -
4 4 4
0 0 0
, , ,
-
6 -
6 -
6
0 0 0
200 ,
-
8
,
-
8
,
-
8
% Negative
1 1 1
, , ,
-
0 -
0 -
0
1 1 1
, , ,
-
2 -
2 -
2
1 1 1
, , ,
4 4 4
100 10%
0.43 0.53 0.33
N = 13033 N = 19123 N = 2432
0 + 0.04 + 0.01
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Telenet 82 108 62 80 129 168 56 172 106 84 47 53 107 40 27 60 551 427 215 334 96 69 94 44 25 50 88 66 64 62 48
Belgacom 75 102 74 24 18 59 41 34 79 71 2 55 75 87 93 82 105 35 29 22 51 37 62 44 22 27 14 57 66 35 18
Positive topics
Positive topics Negative topics
Negative topics
2% Engagement Limited access
(Budget)
PLATFORM conversation topics
PLATFORM conversation topics Online game Bugs & errors
Sentimeter
Alternatives
5% Music videos Slow loading time
Overall, PLATFORM 1. Easier subscription to
conversations are about become a member. Updates from celebs Initial log-in issues
Beta test
Other Products
8%
the current functions
and the devices on 2. Opening up access-
which Yelo can, or ibility: non-clients Social media sources
Quality
cannot be used. Social media sources
Occasions
cannot be come a
Current member. 2%
Advertisement Content Suggestions to improve 5% 2%
Most of the online
functionality
Launch the product were often 8% 5%
Questions
3. More music content 8%
buzz about
Competitors mentioned as well. PLATFORM
availability with more Twitter
genres. happened on Twitter.
Cocreation Devices The PLATFORM website blogs
Twitter
Blogs were a second
is mentioned in a rather 4. A better TV guide for blogs
forums popular platform.
Yelo.be
negative way, due to it the platform. forums
Facebook
Facebook only
Improvements being down during the 5. Adding updates daily
Facebook accounted for a very
launch. There are no instead of one per week. small share of the
More relevance of the 85%
85%
Problems true winning themes so PLATFORM
far. updates. conversations.
Volume
85%
37. Stay tuned!
Connecting on LinkedIn
We believe in sharing knowledge with
all of our stakeholders. Connecting on Facebook
Become friends with us, visit our blogs,
Free content on Slideshare
download our papers and presentations,
order our books… InSites Consulting blog
The Conversation Manager blog
How cool brands stay hot blog
Notes de l'éditeur
Customer care centra Contact formulier van de website Email Geen meta learning Geen trends evoluties Impact van bepaalde acties Veel cijfermatige info maar vaak gaat de tekstuele date die verloren
Alle initiatieven van brands Short marketing Geen trends geen metalearning Eerder connecteren eerder dan research
In contrast to the vast amount of spontaneous user-generated-content online, the market research keeps on asking for more content We organize focus groups, surveys, and interviews where we ask consumers for their opinion. Yes, we have adopted some of the social media tools by setting up research communities, online focus groups with chat and blogs. The large information of social media content has however been largely ignored. Nowedays, we ask our research participants information that might be already be available on social media We should therefore change our philosophy of interviewing to a culture of caching where we listen to the online conversations online
Reduced interviewing bias. Interviewing bias can originate from the interviewer, the participant or the situation itself. When performing social media nethnography we are not asking questions to participants, we collect answers which we try to relate to questions. The social media activities, which generate the information, occur in a natural context. It is part of people’s daily activities and thus in that sense less biased. The observation is unbiased as there is no interpretation or deliberation from the participant’s end. New consumer and patient insights. Within the context of a commercial business operation, the only answers that are business critical are those to new and relevant questions – all the rest is just 'nice to know'. The “new and relevant” bits are where social media represent breakthroughs for companies. Until now, one never knew if the questions asked in a questionnaire were indeed the most relevant and important in the consumer's (here patient's) mind. We only get answers to those questions that are explicitly asked. As we are sitting on a mountain of data generated by users we need to ask ourselves what to get out of it. When doing so we will discover topics and patterns which we did not think of naturally. Contextual information.The web has a reference frame (like a truth thermometer) built-in which allows identifying relevant and important (new) answers and to qualify and quantify them in context. The input from patients is in their own natural language and they report what they find important from their personal perspective. The information also contains more contextual information, as people report their thoughts and feelings in the “heat of the moment”, not when they are probed to recall it. Emotional insights. As opposed to traditional interview based research, nethnography also gives (more) complete answers as the emotional component of an answer is captured and connected to the rational response. An emotionally charged answer (regardless of the type of emotion) is always more important and actionable. Besides, emotions are not evident to measure with explicit and direct questions in surveys on all potential topics. An additional advantage from a methodological perspective is that one can do research over long periods of time, as the researcher can go back in time as far as the social medium holds relevant data from user posts. We organize focus groups, surveys, and interviews where we ask consumers for their opinion. Yes, we have adopted some of the social media tools by setting up research communities, online focus groups with chat and blogs. The large information of social media content has however been largely ignored. Nowadays, we ask our research participants information that might be already be available on social media We should therefore change our philosophy of interviewing to a culture of caching where we listen to the online conversations online
Probleem opgelost text analyse
Two approaches can be taken: You can look at social media content top-down: in this case you look at the data you have collected with a specific research question in the back of your mind. Examples are fi looking what people say about a certain brand. Looking for buzz about a communication campaign. Etc. A second approach is bottom-up analysis: in this case you take a very open approach towards the content by NOT asking specific research questions but instead looking at the big themes that spontaneously emerge in the data.
Applications of social media nethnography
Communication needs & input for a campaign Consumer insights ( Unmet needs / decision process / patient journey/ consumption moments /… )
Communication needs & input for a campaign Consumer insights ( Unmet needs / decision process / patient journey/ consumption moments /… )
Applications of social media nethnography
Applications of social media nethnography
Applications of social media nethnography
Through text analysis we bring structure into these conversations to use them for research. Almost 2.500 conversations were analyzed, resulting in theme detection through pattern detection. For each theme, we determine the size of the cluster and the sentiment of the cluster 1: Act: themes in the market that are often mentioned with a negative sentiment 2: Develop: themes in the market that are often mentioned with a positive sentiment 3: Threats: themes with negative sentiment that are currently not often mentioned but that are explicitly negative for certain market niches 4: Potential: themes with mixed sentiment that are often mentioned. In the future, we can try to influence the sentiment of those themes positively
I hope I have been able to show you the applications of socialmedia netnography, but also the limitations
text analytics is still at its evolutionary beginning, where methods are predominantly based on mining the frequency of emotion words (positive emotion words such as “happy”, “satisfied” or negative emotion words such as “angry” “frustrated”). Surely, the intensity with which such emotion words are present in a text gives an indication of the writers intended sentiment, but is it to sufficient to just mine emotion words? current sentiment analysis neglects other important word categories which may aid in better capturing a writer's intend. We consider further speech acts, that boost or attenuate the expressed sentiment and show that by mining online product and service reviews on these additional categories improves the accuracy in predicting a writers sentiment. Importantly, the speech acts we consider pertain solely to the style of “how” something is formulated rather than the content of “what” is stated, making these additional acts universally applicable to all product and service contexts.
Illustratively, consider these two car assessments. “ This is certainly a great car I consider as a top-class buy” is more definite and directive than “ This is a potentially great car one may consider buying”. While the essence (a positive car assessment) is the same. Reviewer A reinforces her commitment to the opinion using a superlative (certainly) as well as quantifier (top-class). She is not merely expressing her sentiment, but underscoring her commitment to the reader (modal meaning). Conversely reviewer B attenuates her commitment, using tentative words (potentially) and impersonal constructions (one), constituting a different modal meaning.
First, the typical online retail sites (like Amazon, Barnes and Nobles) where customers’ reviews are featured as decision aids and official review sites where products and services are evaluated (like e-opinion, trip advisor, rotten tomato). We aimed to capture a broad spectrum of products and services, varying in the level of involvement customers have with a purchase. Particularly we base our study on a total of 110.000 customer reviews for Beers, Books, Cameras, Cars, Doctors, Hotels, Movies and Airlines. publish a quality star-rating, next to the written review, to rank their overall experience
We first used just the traditionally applied expressive speech acts (emotions) to predict the star-rating of a review. Results were rather poor, with an prediction accuracy of 54.2% in the hold out sample, highlighting the limitations of this approach.
In a second step we then introduced word categories ( Quantifiers, tentative words, comparrissons, word frreflecting certainty, superlatives, impersonal pronouns, fillers, personal pronouns, word count and cognitive processes ) which do not directly have an expressive meaning but increase or diminish the strength of an expression. Our prediction accuracy (of the reviews’ star-ratings) increased to 82.3% in the hold out sample. Introducing the additional word categories, which do not carry any meaning themselves but merely attune the expressive words in a text thus significantly increased prediction accuracy. Concluding thoughts: In conclusion then, beyond mining emotion words, sentiment analysis should start including other word categories writers commonly use to boost or attenuate the strength of their emotion words. Doing so will improve prediction accuracy and lead to more robust managerial decision making.