AWS - Personalization - Good Read

Personalize cross-channel customer experiences 

Key steps

• Collect data from multiple sources to store in Amazon Simple Storage Service (Amazon S3).
• Use AWS Step Functions to orchestrate data onboarding and feature engineering.
• Build segments and predictions using SageMaker.
• Use propensity scores for display targeting.
• Send personalized messaging using Amazon Pinpoint.
• Integrate real-time personalized suggestions using Amazon Personalize.

Code Examples - Link

User Segmentation - Link, Amazon Personalise

Data sources connect to AWS Kinesis Link

Build, tune, and deploy an end-to-end churn prediction model using Amazon SageMaker Pipelines

Keep Exploring!!!

Ideas - create exceptional customer experiences - Retail

You need to have a truckload of ideas. This interesting report has a lot of insights on areas to focus on.

Ref - Link

Some key pointers

Touch point-based analysis

Customer feedback-based Insights

Competitor based comparison

Impact of each level of engagement

Ref - Link

Keep Exploring!!!

How can personalization go wrong

There are many reasons why personalization doesn't happen or goes wrong:

• wrong technology stack
• lack of human and financial resources
• complexity of data
• handling personalization as a one-shot game instead of treating it as a process

major consequences of wrong personalization:

• loss of customer trust because of a breach of privacy
• frustration due to facing irrelevant content and messaging

Ref link

The value of getting personalization right—or wrong—is multiplying

• Thoughtful touchpoints such as checking in post-purchase, sending a how-to video or asking consumers to write a review generate positive brand perceptions.
• When asked to define personalization, consumers associate it with positive experiences of being made to feel special. 
• They respond positively when brands demonstrate their investment in the relationship, not just the transaction

Ref - Link

• 64% of organizations are either not collecting data or are storing it in siloed systems, resulting in disjointed experiences for customers.

Multiple-variable customer segments

multiple-variable customer segments group customers by sets of traits or behaviors they have in common, rather than only one. An example might be a customer who:

• Is from Japan
• Has never transacted before
• Has an interest in the headphones product category
• Is female, age 25-30

it’s more realistic to start with small, incremental improvements. Little wins add up, and they also provide insights you can use to fine-tune your strategy and content.

Keep Exploring!!!

ChatGPT - Food Reviews - Pointers - Examples

List NLP use cases for a food review website

• Sentiment Analysis: Analyzing customer reviews to determine the overall sentiment of the customer towards the food item.
• Text Classification: Categorizing customer reviews into different categories such as positive, negative, neutral, etc.
• Named Entity Recognition: Identifying entities such as ingredients, dishes, restaurants, etc. from customer reviews.
• Topic Modeling: Identifying topics from customer reviews to understand what customers are talking about.
• Summarization: Summarizing customer reviews to provide a concise overview of the customer’s opinion.
• Text Clustering: Grouping customer reviews into different clusters based on their content.
• Automated Response Generation: Generating automated responses to customer reviews based on the sentiment of the review.

List NLP use cases to propose content for users of cooking website

• Recipe Recommendations Based on User Preferences
• Automated Meal Planning
• Automated Grocery Shopping List Generation
• Automated Ingredient Substitution
• Automated Recipe Search
• Automated Recipe Rating
• Automated Recipe Summarization
• Automated Cooking Instructions Generation
• Automated Food Image Recognition
• Automated Food Nutrition Analysis

List different features to collect for Recipe Recommendations of food cooking and review website

• Type of cuisine 
• Ingredients 
• Cooking time 
• Number of servings 
• Dietary restrictions 
• Level of difficulty 
• User ratings 
• Number of reviews 
• User-generated tags 
• Nutritional information 
• Allergen information 
• Price range 
• Cooking methods 
• Region of origin 
• Popularity of recipe 
• User preferences 
• Popularity of ingredients

List different features for Automated Recipe Rating from recipe cooking website

• Taste Rating: Users can rate the taste of the recipe on a scale from 1-5.
• Difficulty Rating: Users can rate the difficulty of the recipe on a scale from 1-5.
• Preparation Time: Automated system can calculate the time required to prepare the recipe.
• Nutrition Rating: Automated system can calculate the nutrition value of the recipe.
• User Reviews: Users can give their reviews and ratings on the recipe.
• Popularity Rating: Automated system can calculate the popularity of the recipe based on the number of views and downloads.
• Ingredient Rating: Users can rate the ingredients used in the recipe on a scale from 1-5.
• Cost Rating: Automated system can calculate the cost of the ingredients used in the recipe.

List key features for Recipe Recommendations of food cooking review website based on seasality

• Automated recipe recommendations based on user preferences and dietary restrictions.
• Ability to search for recipes by ingredients, cuisine, and dietary restrictions.
• Ability to filter recipes based on user ratings and reviews.
• Ability to save favorite recipes and create custom recipe collections.
• Ability to share recipes with friends and family.
• Ability to access recipes from multiple sources, including user-generated content.
• Ability to access recipes in multiple languages.
• Ability to access seasonal recipes and special holiday recipes.
• Ability to access recipes with step-by-step instructions and photos.
• Ability to access nutritional information for each recipe.

List features to engage active users, occasional users of food review website

• Gamification: Offer rewards such as badges, points, and leaderboards to encourage users to post more reviews.
• User Profiles: Allow users to create profiles and follow other users to see their reviews and ratings.
• Social Sharing: Allow users to share their reviews and ratings on social media platforms.
• Recommendation Engine: Use algorithms to recommend restaurants and dishes based on user preferences.
• Personalized Feeds: Create personalized feeds for each user based on their interests and preferences.
• Incentives: Offer incentives such as discounts or freebies for users who post reviews.
• Contests: Hold contests and giveaways to encourage users to post more reviews.
• User Reviews: Allow users to rate and review other users’ reviews.
• Commenting: Allow users to comment on reviews and ratings.
• User Groups: Create user groups to discuss restaurants and dishes.

How do we plug seasonality and recent activity in food recipe recommendation website with feature list and algorithms

Feature List:

• User Profile: Collect user preferences and dietary restrictions to tailor recipe recommendations.
• Seasonal Ingredients: Use seasonal ingredients to suggest recipes that are more relevant to the current season.
• Recent Activity: Track user activity and use it to suggest recipes that are similar to what the user has recently viewed or cooked.
• Popularity: Suggest recipes that are popular among other users.
• Location: Use user location to suggest popular recipes in the region.

• Algorithms:
• Collaborative Filtering: Use collaborative filtering to recommend recipes based on user similarities.
• Content-Based Filtering: Use content-based filtering to recommend recipes based on user preferences and dietary restrictions.
• Hybrid Algorithms: Combine collaborative filtering and content-based filtering to create a more personalized recommendation system.
• Natural Language Processing: Use natural language processing to identify user intent and suggest recipes accordingly.

How do we plug seasonality, age, food interest, recent purchase and recent activity in food recipe recommendation website with feature list 

• Seasonality: Utilize seasonality to recommend recipes that are popular during the current season. For example, in the summer, recommend recipes that include seasonal fruits and vegetables.
• Age: Utilize age to recommend recipes that are appropriate for the user's age group. For example, if the user is a teenager, recommend recipes that are easy to make and require minimal ingredients.
• Food Interest: Utilize food interest to recommend recipes that the user is interested in. For example, if the user is interested in Italian cuisine, recommend recipes that are Italian-inspired.
• Recent Purchase: Utilize recent purchase to recommend recipes that use ingredients the user has recently purchased. For example, if the user has recently purchased a certain type of cheese, recommend recipes that use that cheese.
• Recent Activity: Utilize recent activity to recommend recipes that the user has recently viewed or interacted with. For example, if the user has recently viewed a certain type of recipe, recommend similar recipes.

What coupons, discounts features we can promote on food review site to promote more user engagement

• Loyalty Programs: Offer loyalty programs that reward customers for their repeat business. This could include discounts on future orders, free items, or other incentives.
• Referral Programs: Offer referral programs that reward customers for referring their friends and family to your food review site.
• Coupons: Offer coupons for discounts on orders or free items.
• Discounts: Offer discounts for orders over a certain amount or for certain types of customers (e.g. students, seniors, etc.).
• Special Deals: Offer special deals or promotions on certain days or times of the week.
• Contests: Hold contests or giveaways that reward customers for engaging with your food review site.
• Rewards Programs: Offer rewards programs that reward customers for their loyalty and engagement.
• Social Media Promotions: Promote your food review site through social media channels such as Facebook, Twitter, and Instagram.
• User Reviews: Encourage customers to leave reviews on your food review site. This will help to build trust and credibility with potential customers.
• Newsletter Subscriptions: Offer newsletter subscriptions that provide customers with updates on new products, discounts, and other promotions.

Ideas

• Find insights based on past seasons - Recipe, Ingredients
• Find insights based on holiday seasons - Recipe, Ingredients
• Promote a mix of patterns from past seasons
• The hybrid mix of Events in the timeline (Start Season, Peak Season, End Season)
• User Related groups based on ingredients, age, location, views, follows, likes, shares

Link

• Using a data-driven approach, they segment demand rather than consumers to identify not only what consumers want, but also where, when, why, and how they want it
• Tapping into social media, Chobani obtained real-time information on what kinds of yogurt their consumers want and when they want to eat it. 
• Facebook ads, as a cost-effective way to target consumers demographically, geographically, and psychologically. 
• Top companies charge a premium price that is commensurate with the value of the innovations they make to their product

Incumbents to disruptors: Adopting the start-up culture for innovation

• Free samples Natural
• Sustainable Healthier
• More Effective
• Value-for-money
• Global name
• Discounts

What are Metrics to measure for relevance of personalization

1. Conversion Rate: Measuring the rate at which personalized content leads to conversions can help determine the effectiveness of personalization.

2. Engagement Rate: Measuring the rate at which users interact with personalized content can help determine the relevance of the content.

3. Click-Through Rate: Measuring the rate at which users click on personalized content can help determine the effectiveness of personalization.

4. Time on Site: Measuring the amount of time users spend on a site after being exposed to personalized content can help determine the relevance of the content.

5. Bounce Rate: Measuring the rate at which users leave a site after being exposed to personalized content can help determine the relevance of the content.

Ref - Link

How to we improve engagement on personalized content

1. Make sure the content is relevant to the user.

2. Use data to create personalized content.

3. Use visuals to create a more engaging experience.

4. Ask questions and invite users to interact with the content.

5. Offer incentives for engaging with the content.

6. Utilize social media to reach a larger audience.

7. Use A/B testing to determine what content works best.

8. Monitor user feedback and adjust content accordingly.

9. Encourage user-generated content.

10. Track user engagement and use the data to improve content.

How to increase engagement without personalization in food review site

1. Create a rewards program: Offer customers rewards for leaving reviews, such as discounts or free products. This will encourage customers to leave more reviews and increase engagement.

2. Offer incentives for sharing reviews: Offer customers incentives for sharing their reviews on social media or other platforms. This will help spread the word about your food review site and increase engagement.

3. Host contests: Host contests or giveaways on your food review site. This will encourage customers to leave reviews and increase engagement.

4. Feature customer reviews: Feature customer reviews prominently on your food review site. This will help customers feel like their opinions are valued and increase engagement.

5. Ask customers for feedback: Ask customers for feedback on their experience

How to propose content user has never seen and target engagement

1. Create Unique Content: Create content that is unique and stands out from the rest. This could be in the form of an infographic, a video, an interactive quiz, or a blog post.

2. Utilize Social Media: Leverage social media platforms to reach a wider audience and increase engagement. Share your content on multiple platforms and engage with users who comment or like your posts.

3. Use Targeted Ads: Utilize targeted ads to reach users who are likely to be interested in your content. This could be done through Google Ads, Facebook Ads, or other platforms.

4. Offer Incentives: Offer incentives to users who engage with your content. This could be in the form

What are in organic ways to increase personalization engagement

1. Use personalized emails: Use personalized emails to send targeted messages to customers based on their interests, purchase history, and other data.

2. Personalize content: Create content that is tailored to the individual user’s interests, needs, and preferences.

3. Use segmentation: Segment customers into different groups based on their interests, purchase history, and other data.

4. Leverage user-generated content: Encourage customers to share their experiences with your product or service.

5. Offer personalized discounts: Offer discounts or special offers based on customer data.

6. Utilize customer feedback: Use customer feedback to improve your product or service and create a more personalized experience.

What algorithms quora uses to Amplify high engaging content

1. Natural Language Processing (NLP): Quora uses NLP to analyze user-generated content and identify topics of interest. This helps them to identify content that is likely to be engaging to their users.
2. Machine Learning: Quora uses machine learning algorithms to identify patterns in user-generated content. This helps them to identify content that is likely to be engaging to their users.
3. Recommendation Algorithms: Quora uses recommendation algorithms to suggest content to users based on their interests and past behavior. This helps them to amplify content that is likely to be engaging to their users.
4. Social Network Analysis: Quora uses social network analysis to identify influential users and content that is likely to be engaging to their users.
5. Sentiment Analysis: Quora uses sentiment analysis to identify content that is likely to be engaging to their users based on the sentiment of the content.

What algorithms quora uses to reduce low engaging content

• Content Quality Score: Quora uses a content quality score to identify low-engaging content. This score is based on factors such as the number of views, upvotes, and comments a post has received. Posts with a low content quality score are more likely to be filtered out or marked as low-engaging content.
• Natural Language Processing: Quora uses natural language processing to identify posts that are not relevant to the topic or contain offensive language. Posts that are flagged as low-engaging content are removed from the platform.
• Machine Learning: Quora uses machine learning algorithms to identify low-engaging content. The algorithms analyze user behavior and engagement levels to determine which posts are not engaging enough to be featured on the platform.
• User Feedback: Quora also uses user feedback to identify low-engaging content. Users can flag posts as low-engaging, and Quora will use this feedback to determine which posts should be removed from the platform.

What algorithms quora uses to clean up contents not engaging

• Natural Language Processing (NLP): Quora uses NLP to detect and remove spam, offensive language, and other inappropriate content.
• Machine Learning: Quora uses machine learning algorithms to identify content that is not engaging and remove it from the platform.
• Text Analysis: Quora uses text analysis to identify and remove content that is not relevant to the topic or discussion.
• Content Moderation: Quora uses a team of moderators to review content and remove any content that is not appropriate for the platform.
• Automated Filtering: Quora uses automated filters to detect and remove content that violates its terms of service.

Key Analysis

• Top Writes
• High-Quality Writers
• Low-Quality Writers
• High content features
• Engagement across ages
• Cusines liked across ages
• Regions with engagement
• Time to visit
• Time spent on weekday/weekend
• Time spent over holidays / before holidays
• Response for new launches
• Rank articles by views, upvotes, and comments
• Identify posts that are not relevant
• Determine which posts are not engaging
• Receipt - Novel, Known, Experiment
• Categorize as new recipe, existing but improvements, cook in 5 mins
• Remove poor content in every category
• Re-rank articles to suit to date / season / weekday / weekend
• Custom word embedding created for different food persona
• A stop list or custom choices to avoid any strong dislikes

Why Are Online Recipes Always So Damn Long-Winded?

• Some people love the stories
• SEO (Search Engine Optimization)
• It’s a business
• Recipes aren’t copyrighted

How to ARCHITECT a search engine like Google Search

• Feature creation
• Entity Extraction
• Keyword Matching
• Custom word to vec creation of receipe, food, entity
• A/B Experiments

Possible metrics include

• 7-day new post rates
• 7-day new Upvotes / Comments
• 30-day Sample Conversion rate
• 90-day retention rate/user churn
• 90-day annual revenue per user

Expected Revenue (XR), XR is meant to measure this.

• Variety of factors
• User activity such as Posting recipe, Upvotes, Replies  
• Utilizing the samples
• Inviting friends to join the group
• User location activity Engagement

Yandex Lessons

• Yandex has a RankBrain analogue called MatrixNet
• Yandex also uses PageRank (almost the same as in Google);
• A lot of text algorithms are the same
• 1,922 ranking factors

Key Lessons

• Link age is a ranking factor
• Traffic and % of organic traffic affect rankings
• Things like CTR, last click, time on site, and bounce rate impact rankings
• They take the average position for all your keywords into account. Does that imply that focusing on fewer, high-ranking KWs is better than going for more KWs (with lower positions)?
• Newer pages and recently updated pages have better ranking (no big surprise there)
• More people search for your brand, the higher your (other) pages rank.
• If people bookmark your page, it sends a good signal (probably trust or authority of thoroughness, etc.) that impacts ranking
• Direct traffic is good, If all your site is getting is organic traffic, it may
• look suspicious.
• If you publish low quality content, it can negatively impact your entire site ranking.
• Wikipedia has a special ranking factor
• Traffic from Wikipedia is a ranking factor.
• Backlinks from main pages are more important than from internal pages.
• Special ranking factors for short videos (TikTok, Shorts, Reels)
• Embedded videos are good for rankings, but broken embed videos are bad.
• JS from Google Analytics is a ranking factor (if you use GA, it's good)
• Host reliability is a ranking factor (40x/50x errors)

Reddit Recommendation Inspirations

• Candidate Post Generation - Submissions from the past 24 hours, and filter it through criteria intended to tell us what each user might enjoy
• Community subscriptions: each community you’ve joined
• Similar communities: communities similar to those you have joined (currently we use semantic similarity)
• Onboarding categories: categories you said they were interested in during onboarding (like “Animals & Awws” or “Travel & Nature”)
• Recent communities: communities that the user visited in recent days
• Popular and geo-popular: Posts that are popular among all redditors, or among redditors in their local area (only if permitted in app settings)

Filtering Criteria for Posts

• Every post we show on Reddit must meet a quality and safety threshold, so on the Best Sort we remove posts from the list that we think might be:
• Spam, deleted, removed, hidden, or promoted
• Posts the user has already seen
• Posts from subreddits or topics that the user asked we show less of
• Posts the user has hidden
• Posts from authors the user has blocked

ML Model - “features” for each candidate post.

• Post votes: The number of votes on the post. The magic of Reddit is that it is primarily curated by redditors via voting. This remains at the core of how Reddit works.
• Post source: How we found this post (subscriptions, onboarding categories, etc.)
• Post type: The type of the post (text, image, video, link, etc.)
• Post text: The text of the post
• Subreddit: Which subreddit the post is from, and the ratings, topics, and activity in that subreddit (for more on Ratings and Topics read this).
• Post age: The age of the post (we value giving you a “fresh” Home feed)
• Comments: Comments and comment voting
• Post URL: The URL the post links to, if the post is a link post
• Post flairs: Flairs and spoiler tags on the post

We could rank posts assigned based on views and use it for new posts ranking.

Machine Learning-Powered Search Ranking of Airbnb Experiences

Building signals based on which we will rank: In Stage 1 of our ML model, we decided to rank solely based on Experience Features. In total we built 25 features, some of which were:

Customizing features for Food Review

• Experience duration (e.g. 1h, 2h, 3h, etc.)
• Product Price and Price-per-hour
• Category (e.g. cooking class, music, surfing, etc.)
• Reviews (rating, number of reviews)
• Number of similar items samples used (last 7 days, last 30 days)
• Conversion of coupons (e.g. 60%)
• Click-through rate

Ref - Link

The most commonly observed gamification elements were:

• Levels or progress feedback
• Points or scoring
• Rewards or prizes
• Narrative or theme
• Personalization
• Customization

The least commonly observed gamification elements were:

• Artificial assistance
• Unlockable content
• Social cooperation
• Exploratory or open-world approach
• Artificial challenge
• Randomness

Moving from Tactical Testing to Strategic Experimentation and Personalization

Understanding the difference between personalization and simple A/B testing

The mechanics of games are often reproduced unimaginatively, making the whole experience not fun.

It's important that people actually care about the rewards they receive! 

You should always be upfront about what you are designing with your users.

Gamification isn’t meant to be about tricking you, it’s meant to help you get you on your way, in an ethical way.

Keep Exploring!!!

Personalized - Redefined - Recommendations - Paid Service

What if we get a customized recommendation based on our needs than based on what we are forced to see. 

Workday recommendations (Articles / Videos based on interests)

• Data Science (20%)
• Startups (20%)
• Stock markets (20%)
• Travel Vlog (10%)
• Paranormal Vlog (5%)
• Food Vlog(5%)
• Emotions / Wellness (10%)

Weekend Recommendations

• Travel Vlog (20%)
• Paranormal Vlog (20%)
• Food Vlog(10%)
• Emotions / Wellness (20%)
• Music (30%)

Tired of seeing the same swiggy ads, zomato ads, irrelevant to the context ads.

Ref - Link

Keep Thinking!!!

A very interesting read - Personalization also can lead to unfair pricing

When you visit several times, Your interest is captured in clicks. More you visit, More you have interest to engage

Tinder - participants aged 30-49 on average paid 65.3% more than those aged 18-29

Age group 30-49 - Earn more than teens, They can afford, Higher conversions/success rates. The fear of missing out to settle have a family / engage / search and end up paying more :)

Reference - Link1, Link2

Keep Exploring!!!