Digital Technologies and Marketing Flashcards
What is web scraping
Using application programming interfaces (APIs) to collect data from the internet
Benefits of web data (3)
- Enormous size
- Publicly available
- Cheap to access
Addressing validity struggles of web data (2)
Addressing both technical and legal/ ethical concerns
Matrix of for types of data (4)
Data format: structured/ unstructured
Data Source: external/ internal
Big data (3)
- Volume
- Velocity
- Variety
The internet of things (IoT)
Physical objects that connect and exchange data with other devices through a network.
Benefits of web scraping (4)
- Study new phenomena: new areas of research and faster turnaround
- Ecological value: more controlled real-life data without external involvement
- Methodological advancement: new types of data, new ways to process them.
- Improving measurement: new or more detailed variables.
4 Dimensions of online customer experience
- Informativeness
- Entertainment
- Social presence
- Sensory appeal (stimulates sight and sound)
-> Ultimately impacting purchase decision
Moderators of online customer experience (2)
Product (search vs. experience)
Brand (trustworthiness)
Web design A/B testing
Comparing two versions to see which one works
Letting multiple ads run and see which one performs best
Multivariate testing
Tests multiple features at the same time
Useful for relative/ interaction effects
Full factorial design
Tests all possible combinations, useful to test for interactions
Multi-touch attribution (MTA)
- Consumers can have many (MTA) before making a purchase
- Touch points are very different and their efficiency is difficult to determine
First touch/ click
The purchase happens with the first contact.
Last touch/ click
The purchase happens with the last time they are in touch
Linear attribution
The probability of purchase is the same with every interaction.
Time decay attribution
The more touchpoints happened the higher the probability.
Macro-influencers: and high-arousal
-> Arousal for macro-influencers decreases engagement (monetary interests)
-> Informative goal for macro-influencers leads to more engagement; even more so when arousal is high (monetary interests are less apparent)
Use of high-arousal language by macro-influencers can lead to a decrease in engagement, it is perceived as being overly commercial and decreases trustworthiness. It is preferred to provide informative content rather than solely promoting.
Micro-influencers:
-> Arousal for micro-influencers increases engagement (genuine excitement)
Benefit from the use of high-arousal language (“its totally AMAZING”). IT contributes to a higher level of engagement from the audience. It provokes trustworthiness and authenticity. It is perceived as being more genuine, which in turn fosters a stronger connection and encourages more interaction with the content. This effect highlights the importance of perceived sincerity and enthusiasm in micro-influencer marketing strategies.
Bass model: (3)
- Differentiations between innovation (p) and imitation (q).
- Innovation signals consumers willing to test new things
- Imitation signals social contagion
- Driving imitation can greatly increase diffusion
- Social media can great increase social contagion
Lift (based on occurrence & co-occurrence) Text analysis models
- Lift between words A&B: lift (A,B) = P(A,B)/ P(A)*P(B)
how often they appear vs. how often they appear together.
TF-IDF (term frequency – inverse doc frequency) Text analysis models
- For word i in document j: TFIDF ij = TFij * IDFj
- Useful for identifying which words are meaningful
- Often used for search engines (google)
- Can be used as weighting for other methods
Cosine similarity; Text analysis models
- Can be used to check similarity of reviews
- Can be sentences, reviewers, or word pairings over reviews
Sentiment analyses (dictionary, basis); Text analysis models
- Uses other methods and sees how often the brand name is used with the word “good”
- it can also be used directly on your won specific reviews
- Positive words get +1 and negative -1
- Control the length of the review
–> Can also use AI applications
Explicit recommendations
approval for OTHERS
- using words like: recommend/suggest
- Seem more knowledgeable (expertise and authority)
- Perceived to like product more
- More persuasive and increase purchase
Implicit recommendations
declaration for your SELF
- using words like: I like/ I enjoy/ my favorite novel
What influences the effect of word of mouth(4)
- the language chosen (implicit/ explicit)
- liking/credibility of the sender
- the mediumof WOM
- WOM has similar effects on experts/ novices
Social proof (copying actions from others) (3)
- Psychological phenomenon where people copy actions from others
- Social proof is powerful and omnipresent (widespread)
- Many examples in both offline and online setting
Key findings (Machines vs. Humans: The Impact of Artificial Intelligence Chatbot Disclosure on Customer Purchases)
- Chatbot Disclosure’s Impact: sales 80% when the customer knows they are talking to a chatbot
- Human Perception Plays a Role: Sales drop because people think that bots are less knowledgeable and emphatic
Strategies to Mitigate Negative Impact (Machines vs. Humans: The Impact of Artificial Intelligence Chatbot Disclosure on Customer Purchases)
- delaying disclosure of Bot as long as possible
- Enhancing AI experience before, people with AI experience like chat bots more.
Step 1: Item-based recommendation
Demeaning - calculate the mean of the user and then take every value minus the mean
Step 2: Item-based recommendation
v are all value from the top
r are all demanded values
-> How important the features are for the participant
Step 3: Item-based recommendation
- consider all attributes in one calculation
- Be careful and consider the multiplication in the square root
Step 1: user-based recommendations
- demean all the value based on the participant
A. Lift (based on occurrence & co-occurrence)
-> how often they appear vs. how often they appear together.
TF ij (def)
How many times a word appears in this specific comment (indicating prminence)
IDF𝑖 = log(𝑁/𝑛𝑖) (2)
𝑁: the total number of comments
𝑛𝑖: in how many comments the word appears, it does not matter if that happens more than one time, it is just about the total number of comments with that word (indicating rarity).
Interpretation Lift
Lift > 1 -> positive association
Lift < 1 -> no association
TF-IDF Score interpretation
A measure of the importance of a word in a document relative to a collection of documents.
Three stages collecting web data
- Selecting data sources
- Designing collection
- Extracting data
- Linguistic style
- Descriptive product detail
- Bulleted product features
- Return policy information
Visual elements - Webpage Design Elements (4)
- Product feature crop
- Lifestyle photo
- Photo size
- Product video
Verbal/ Visual Elements - Webpage Design Elements (4)
- Customer star ratings
- Expert endorsement
- Comparison matrix
- Recommendation agent
- Content filter
Search products (2)
- attributes can be easily be evaluated before the purchase
- informativeness is more important
Experience products (2)
- Quality or satisfaction level can only be assessed after use
- Socia experience is more important
Brand (trustworthiness) characteristics (2)
- Entertainment and customer customized content is more important if trustworthiness is low.
- Information is more important if trustworthiness is high.
Characteristics of big influencers (4)
- Reaches larger audience
- Generate more sales
- Benefit from promoting a limited number of products
- Trustworthiness decreases the more influencers promote the same product
Characteristics of small influencers (3)
- More effective at increasing the audiences conversation rate
- Small-influencer-only strategy is dominant
- Seem more relatable and approachable to normal people
Using small and big influencers (2)
- Strong negative effect on big influencers due to decreased trust, specially when small influencer promotes first.
- When using both influencers, the advertisement campaign does not seem genuine anymore.
Influencer goal (4)
- Creating awareness
- Encouraging trial/ purchase
- Informative motives are less clearly linked to advertising
- Commercial motives are more clearly linked to advertising
How much does a big influencer increases sales
+ 259.7%
How much does a small influencer increases sales
+18.5%
Content-based filtering
recommends content to users on the basis of topic they already subscribed to
Social filtering algorithm
Recommends content with which users’ social connections are already engaged
Effects of shifing from content-based to social-based algorithm
shifted online activity from content-oriented to social-oriented, resulting in a rich-get-richer effect.
Moved from content-based filtering to social filtering
Main effects (4)
- Socia interactions increased by 15%, more social relationships. People got more inclined by following already popular users “rich-get-richer” effect.
- Decreased content engagement: Reduced question subscription (subscribed to get answer to a topic) by 20% and reduced answer contribution (subscribed to give answers) by 23%. User spend more time following people rather than subscribing to or engaging with new content.
- Diverse content consumption: Content did not get homogenized, social filtering diversified content instead. Popular topics saw fewer new subscribers, while niche topics gained more attention. This is because popular users would engage with more niche topics.
- Shift from content to social focus: User behavior from content oriented to socially-oriented
Def. metaphor
Metaphors manifest in text in a recognized structure known as target source, where something less familiar (target) is something more familiar (source).
Short comings of traditional sentiment analysis (SA): (3)
- Object identification (what is the actual target)
- Content vs. sentiment (is the content positive or negative)
- Sentiment-in-context (puts the qualitative data into numbers which changes its texture)
Benefits of MEMSA: (3)
- Adds contextual depth
- Strong capacity to express sentiments (identify what the sentiment is directed at)
- The depth of emotional associations they carry with (insights into emotional state and sociocultural context)
MEMSA Methodology: A step-by-step approach (4)
Step 1: Metaphor dictionary creation
- Creating custom dictionaries that catalog metaphors commonly used in discussion about a specific topic.
Step 2: Metaphor characterization
- Internpreting the meaning of the before listed metaphors
Step 3: Metaphor enables sentiment analysis
- Applying the before collected and defined metaphors to large scale textual data
Step 4: Triangulation
- To ensure robustness of the data, authors cross-validate their results with other research methods such as qualitative interview and discourse analysis.
Novices and WOM (3)
- Less aware of preference heterogeneity
- Less sophisticated tastes and assume other people will share their tase
- Use more explicit endorsement
Manager Response to Positive Reviews
NEGATIVE Effect
- Fake/ too promoting and the customer already knows they liked the service
Manager Response to Negative Reviews
POSITIVE Effect
- Feels like confronting the problem
- Should be timely and customized
Why are chatbots less liked compared to talking to a human (2)
- Perceived as less knowledgable
- Less Emphatic
Why are less AI literat people more likely to implement AI
- People with less AI experience think that is capable of much more
- Aren’t as aware of its limitations and of its ethical constraints
AI literacy
knowledge about AI
AI receptivity
interested in using AI
Removal Rate
= Sum of the conversions based on the removed channel / Total Conversion Rate
Weighted Removal Rate
= Removal Rate / Sum of all removal rates
Validity issue with web scraping (4)
(1) Missing contextual information of fast changing environment
(2) No alignment with psychological process (trends are changing fast, but you also don’t want to gather redundant data)
(3) Dismissing personalized algorithms on websites, algorithm bias
(4) Failing to retain raw website
- Selecting data sources (collecting web data) 3
1.1. Potential sources: should consider a broad spectrum of websites and APIs
1.2. Consider alternatives to web scraping: which are more scalable and have less legal issues
1.3. Mapping the data in context: Understanding the context of data
- Designing collection (collecting web data) 4
2.1. Which information to extract
2.2. How to sample: random sample/ sample size etc.
2.3. At which frequence to extract information: disappear/ evolve
2.4. How to process the information during the extraction: deciding on degree of “rawness” of the date
- Extracting data (collecting web data) 3
3.1. Improve performance of the data extraction: scaling the operations
3.2. How to monitor data quality during and after the extraction: live monitoring data extraction
3.3. How to document data during and after the extraction: record relevant information about the data in real-time
Motivation for: Metaphor-Enabled Marketplace Sentiment Analysis
Motivation for study: Traditional sentiment analysis struggles with identifying the exact target of sentiments. Metaphors can provide context by linking emotions to specific marketplace topics like “drowning in debt”.
Motivation for research: “Bad News? Send an AI. Good News? Send a Human
Motivation of the research: When people receive bad news the satisfaction decreases vs. when they receive better than expected new the satisfaction increases. But how is the human reaction influenced by if a human transmits the bad news vs. AI transmits the bad news.
Why considering using chat bots in the customer service communication: (2)
- Good speaking ability, deeply understanding customer needs and responding with compassion and humor.
- Chat Bots are more reliable in their performance since they don’t have bad days or get tired.
Using chat bots in customer purchase (4)
- Undisclosed chatbots are as effective as proficient workers, 4x more effective than inexperienced workers.
- The research includes different categories of experience level of workers and time of disclosure of chatbot.
- People would just hang-up when they realize they are talking to a bot; because they don’t feel comfortable talking to a robot about personal needs. Bots are perceived as less knowledgeable, less empathetic and less trust worthy.
- Prior AI experience reduces the negative disclosure effect
Study 1: Cross-Country AI Literacy and AI Receptivity (2)
- RQ: Does national AI literacy predict AI receptivity?
- Key Finding: Countries with lower AI literacy had higher AI receptivity.
Study 2: AI Literacy and Student Use of Generative AI (2)
- RQ: Does AI literacy predict students’ use of AI for assignments?
- Key Finding: Students with lower AI literacy were more likely to use generative AI like ChatGPT for assignments
Study 3: AI Literacy and Past AI Usage (2)
- RQ: Does AI literacy predict prior AI usage beyond other personal traits?
- Key Finding: People with lower AI literacy reported using AI more frequently in the past six months, even when controlling for tech readiness and general knowledge
Study 4: AI Literacy and Perceptions of AI as Magical
- RQ: Does perceiving AI as “magical” explain why lower AI literacy increases AI receptivity?
- Key Finding: Lower AI literacy was linked to seeing AI as magical, which increased AI receptivity
Study 5: AI Literacy, Awe, and Fear of AI
- RQ: Do magical perceptions and awe explain the AI literacy-receptivity link?
- Key Finding: Lower AI literacy increased AI receptivity through awe (Ehrfurcht), despite also increasing fear of AI.
Study 6: Generalization Across Various Tasks
People are less comfortable using algorithms for subjective tasks.
But if the algorithm is seen as highly effective or human-like, they trust it more.- RQ: Does the link between low AI literacy and higher AI receptivity apply to different types of tasks?
- Key Finding: Yes, but it depends on the type of task. The link was weaker for tasks that were more objective or non-human (like data tasks), because they didn’t make people see AI as magical.
Study 7: Influence of Task Human-ness on AI Receptivity
- RQ: Does the human-ness of a task change how AI literacy affects AI receptivity?
- Key Finding: Yes. For human-like tasks (e.g., creativity), people with low AI literacy were more open to AI. But for objective tasks, it was the opposite – people with high AI literacy were more open to AI.
