AI in Marketing: Recommender Systems PDF
Document Details
Uploaded by CheerfulUvarovite
University Canada West
Tags
Summary
This document provides an overview of AI-based recommender systems, exploring different types and their application in achieving higher customer satisfaction in marketing. It showcases how businesses can predict customer preferences to improve product and service recommendations.
Full Transcript
AI in Marketing Marketing Personalization Recommender Systems: AI-Based Personalization Every business is keen on improving its performance by achieving higher customer satisfaction. It is possible with personalization. Just as we pick gifts for our friends by knowing their likes and dislikes, comp...
AI in Marketing Marketing Personalization Recommender Systems: AI-Based Personalization Every business is keen on improving its performance by achieving higher customer satisfaction. It is possible with personalization. Just as we pick gifts for our friends by knowing their likes and dislikes, companies can also predict their customers' choices regarding products and services. Types of Recommender Systems 1. Popularity Based Recommends items based on their overall popularity among all users. Advantages: Simple and easy to implement. Works well for new users or when there is little to no user data available. Disadvantages: Doesn't consider personal preferences of users (recommends same item to all users). May recommend popular items that are not relevant to specific users' interests. Types of Recommender Systems 2. Content-Based Recommendation System: Recommends items similar to those the user has liked or interacted with before, based on the features or characteristics of the items. Advantages: Provides personalized recommendations based on user preferences. Doesn't require user data from other users. Disadvantages: Limited by the quality of item features used for similarity calculations. May suffer from the "filter bubble" problem, where users are only exposed to similar items and may miss out on diverse recommendations. Types of Recommender Systems 3. Collaborative Filtering Recommendation System: Recommends items to a user based on preferences or behavior of similar users. Advantages: Can provide accurate recommendations even for new or niche items. Doesn't rely on item features, which can be subjective or hard to define. Disadvantages: Cold start problem for new users or items with limited data. Privacy concerns as it requires user data to find similar users. Types of Recommender Systems 4. Hybrid Recommendation System: Combines multiple recommendation approaches, such as content-based and collaborative filtering, to provide more accurate and diverse recommendations. Advantages: Mitigates limitations of individual recommendation approaches. Can provide better recommendations by leveraging strengths of different methods. Disadvantages: Complex to implement and maintain. Requires significant computational resources and data integration. Example: Content-Based Recommendation Systems Image result for further reading icon Training watched movies Movie Genres Comedy 0 1 1 Comedy 0 10 8 Comedy 0.3 Mia Adventure 1 1 0 Adventure 2 10 0 Adventure 0.2 2 10 8 Sci-Fi 1 1 1 Sci-Fi 2 10 8 normalize Sci-Fi 0.33 User Rating Dram 0 1 0 Dram 0 10 0 Dram 0.16 Movie Matrix Weighted Movie Matrix Example: Content-Based Recommendation Systems Prediction new movies Mia Comedy 1 0 0 Comedy 0.3 Comedy 0.3 0 0 Adventure 1 0 0 Adventure 0.2 Adventure 0.2 0 0 Sci-Fi 1 1 1 Sci-Fi 0.33 Sci-Fi 0.33 0.33 0.33 Dram 0 1 0 Dram 0.16 Dram 0 0.16 0 Movie Matrix Weighted Movie Matrix 0.83 0.49 0.33 Recommendations Example: Collaborative Filtering Training Comedy Movie matrix Mia - 4 - 10 - 6 not known Adventure Tim 3 1 - - 2 - Sci-Fi Zac - - 4 - 2 2 Dram User rating Comedy not known Adventure Sci-Fi Dram Example: Collaborative Filtering Training Comedy 0 0 1 1 0 1 Guess movie Mia - 4 - 10 - 6 matrix Adventure 0 1 1 1 1 0 Tim 3 1 - - 2 - Sci-Fi 1 1 0 1 0 1 Zac - - 4 - 2 2 Dram 1 0 1 1 1 0 Guess user rating Comedy 3 0 1 Adventure 1 0 0 Sci-Fi 3 1 1 Dram 3 2 2 Example: Collaborative Filtering Prediction Comedy 0 0 1 1 0 1 Mia 6 4 7 10 4 6 Adventure 0 1 1 1 1 0 Tim 3 1 2 3 2 1 Sci-Fi 1 1 0 1 0 1 Zac 3 1 4 4 2 2 Dram 1 0 1 1 1 0 Comedy 3 0 1 Adventure 1 0 0 Sci-Fi 3 1 1 Dram 3 2 2 Graph-based recommendation system Image result for further reading icon Graph Neural Networks (GNNs) have been one of the hottest topics in the AI world in recent years, with many potential business applications. They are representatives of one of the most powerful groups of machine learning algorithms, which are Artificial Neural Networks. Because of their many desirable properties, GNNs have gained popularity for application in the solving of a wide variety of business problems. They are used in fraud detection, in the field of drug discovery or in social networks analysis, amongst others. GNNs take advantage of the fact that in many of these cases, the data can be very easily represented as graphs, such as the relationships between groups of people in the case of social networks. However, arguably one of the most promising applications of GNNs is in recommendation systems. By analyzing the relationships between products and users, GNNs can make personalized recommendations based on past behavior and interactions. Graph-based recommendation system Image result for further reading icon Graph Neural Networks (GNNs) have been one of the hottest topics in the AI world in recent years, with many potential business applications. They are representatives of one of the most powerful groups of machine learning algorithms, which are Artificial Neural Networks. Because of their many desirable properties, GNNs have gained popularity for application in the solving of a wide variety of business problems. They are used in fraud detection, in the field of drug discovery or in social networks analysis, amongst others. GNNs take advantage of the fact that in many of these cases, the data can be very easily represented as graphs, such as the relationships between groups of people in the case of social networks. However, arguably one of the most promising applications of GNNs is in recommendation systems. By analyzing the relationships between products and users, GNNs can make personalized recommendations based on past behavior and interactions. Graph-based recommendation system Image result for further reading icon Graph Neural Networks (GNNs) have been one of the hottest topics in the AI world in recent years, with many potential business applications. They are representatives of one of the most powerful groups of machine learning algorithms, which are Artificial Neural Networks. Because of their many desirable properties, GNNs have gained popularity for application in the solving of a wide variety of business problems. They are used in fraud detection, in the field of drug discovery or in social networks analysis, amongst others. GNNs take advantage of the fact that in many of these cases, the data can be very easily represented as graphs, such as the relationships between groups of people in the case of social networks. However, arguably one of the most promising applications of GNNs is in recommendation systems. By analyzing the relationships between products and users, GNNs can make personalized recommendations based on past behavior and interactions. Digital Marketing: Cookies Image result for further reading icon Cookies are short texts that are exchanged between the server and the browser. In this way, websites store relevant information in the researcher’s internet browser, which can then be called up again at a later time. Cookies do not have a good reputation among Internet users: they save the data required for various functions on the hard drive and are able to easily identify visitors to the website. However, it is undeniable that cookies make surfing much more convenient. There are two types of cookies: 1.First-party cookies are created by the website the 2. Third-party cookies are placed on a user’s browser by a website user is visiting. They enable the site to recognize the other than the one they’re currently visiting. They can track visitors user’s device and store information that can improve across websites, which enables you to gather data about users’ their browsing experience, like saving items in a browsing habits, preferences, and interests. This information is then shopping cart or remembering that the user is logged in. used to deliver personalized advertising experiences. Google Phasing Out Third-Party Cookies Image result for further reading icon On January 4, 2024 Google began testing its new privacy features Google’s decision to remove Chrome’s third-party cookie support is part of a and stopped the use of third-party larger Privacy Sandbox launched in August 2019, a series of initiatives “to cookies in the Google Chrome develop a set of open standards to fundamentally enhance privacy on the web.” browser for 1% of users. Why are third- All of the data that third-party cookies collect can be put together to create extensive profiles on party cookies users consisting of thousands upon thousands of data points, such as your Google searches in the under scrutiny? last five years, your credit card transactions, your profile on dating apps, and so on. Image result for further reading icon AI as an alternative to third-party cookies Collecting first-party data is also a highly valuable way to enable your AI marketing tools to perform at higher levels. It also allows your customer base to have more control about how and when they share their data and with which websites, again increasing their trust and improving your personalization efforts. AI in Information Retrieval and Search Engines Image result for further reading icon How Is AI Used in Search? Almost every aspect of your experience with search engines is directly powered by AI. The biggest way search engines use AI is to rank webpages, videos, and other content in search results. The algorithms used by these AI systems have many rules that prioritize different factors, from the types of keywords in your content to your site's user experience. Today, Google has taken understanding search queries a few giant leaps forward. Using its pre-trained language model BERT, the company's search engine now understands complete sentences. That means BERT can understand the context of a search, not just the keywords in it. BERT uses AI in the form of natural language processing (NLP), natural language understanding (NLU), and sentiment analysis to process every word in a search query in relation to all the other words in a sentence. This is different from how Google search used to work. In the past, Google's AI processed Voice Search and Image Search words one-by-one in order. The results were accurate, but literal. Voice search and image search are newer search capabilities made For example, Google shares that a search like "2019 Brazil traveler to USA need a visa" possible by AI. would have been interpreted in the past as a US traveler wanting a visa to Brazil. AI technologies like NLP have gotten so advanced they can actually In that scenario, Google didn't account for prepositions that change the context of understand human voices in real-time, like when you use a Google searches. It wouldn't have taken the word "to" into account, which fundamentally Assistant. These AI-powered systems can understand your words, changes the search intent. then translate those into search results. BERT is different. BERT takes the whole sentence into account and understands the The same goes for images. AI technologies like image recognition can search is a Brazilian looking for a US visa. determine what is being depicted in an image, then deliver relevant search results around that image. Types of Search Engines Binary (Based on Existence) Existence search focuses on whether a term exists in a document rather than its frequency or importance. Simple and efficient, however, ignores term frequency and importance and may not differentiate well between documents with varying levels of relevance. Term Frequency (TF) or Bag of Words (BoW) TF/BoW is used to rank documents based on the frequency of the search query terms within each document. Simple to calculate and understand. Effective for short documents, however, ignores the importance of terms in the overall corpus and is sensitive to document length. Term Frequency - Inverse Document Frequency (TF-IDF) TF-IDF is used to rank documents based on the importance of search query terms within each document relative to the entire document collection. It combines TF with IDF, which penalizes terms that are common across the entire document collection. Considers both local (within-document) and global (across-document) importance of terms. Effective for long documents and large corpora, however, can be computationally expensive for large document collections. Types of Search Engines Embeddings (Like Word2Vec) Embeddings could capture semantic relationships and context. Models like Word2Vec learn to encode word meanings based on word co-occurrence patterns in a large text corpus. Can handle synonyms and word ambiguity, however, requires large amounts of training data and computational resources. Transformer Based Models (Like BERT) Effective for understanding natural language queries, document ranking, and relevance matching. Captures nuanced semantics and context in natural language. Requires significant computational resources for training and inference. HOW TF-IDF Works? Image result for further reading icon Document 1: It is going to rain today. Document 2: Today, I go to school. Document 3: I am busy today. Words to be merged: go & going Step 1 is & am Step 2 Clean data and Tokenize Find TF = (Number of repetitions of word in a document) / (# of words in a document) word today to is I go it busy rain school word today to is I go it busy rain school count 3 2 2 2 2 1 1 1 1 Document 1 0.16 0.16 0.16 0 0.16 0.16 0 0.16 0 Document 2 0.2 0.2 0 0.2 0.2 0 0 0 0.2 Document 3 0.25 0 0.25 0.25 0 0 0.25 0 0 TF Table Step 3 Find IDF = Log[(Number of documents) / (Number of documents containing the word)] word today to is I go it busy rain school A measure of weather a term is rare or common in log(3/3) log(3/2) log(3/2) log(3/2) log(3/2) log(3/1) log(3/1) log(3/1) log(3/1) IDF a collection of documents =0 = 0.41 = 0.41 = 0.41 = 0.41 = 1.09 = 1.09 = 1.09 = 1.09 IDF Table HOW TF-IDF Works? Document 1: It is going to rain today. Document 2: Today, I go to school. Document 3: I am busy today. Step 4 Build the model and use it to answer the Search word today to is I go it busy rain school Document 1 0 0.07 0.07 0 0.07 0.17 0 0.17 0 TF Table Document 2 0 0.08 0 0.08 0.08 0 0 0 0.22 Document 3 0 0 0.1 0.1 0 0 0.27 0 0 IDF Table word today to is I go it busy rain school Search: go search 0 0 0 0 1 0 0 0 0 Customer Service Chatbot Image result for further reading icon A chatbot for customer service is a software program or application that is specifically designed to interact with customers and provide support or assistance. It is deployed on platforms such as websites, messaging apps, or mobile apps, enabling customers to engage in text-based or voice-based conversations. Rule-based chatbots: Rule-based chatbots operate on a predefined set of rules and responses. They follow a decision tree or flowchart structure to give answers based on specific keywords or patterns in customer queries. Rule-based chatbots are typically limited in their capabilities and require manual updating when new rules or responses need to be added. AI-powered chatbots: AI-powered chatbots utilize artificial intelligence and machine learning techniques to understand and respond to customer queries. AI customer service chatbot exploits natural language processing (NLP) algorithms to interpret and extract meaning from text or voice inputs. AI chatbots can learn from interactions, sharpen their responses over time, and handle more complex conversations. Customer Relationship Management (CRM) Image result for further reading icon Customer relationship management (CRM) is one of the most exciting innovations in modern marketing for businesses in a variety of different industries. CRM is the term used to describe the tools and strategies businesses use to track the various stages of their relationships with any client, from the moment they are onboarded and throughout the various project collaborations. CRM systems are the actual technology used for managing a company’s various relationships and communications with existing and potential clients. A CRM system assists companies in staying connected with their client base, responding more fluidly to the changing needs of clients, and improving profitability by maximizing every client interaction and more efficiently generating new client leads. Image result for further reading icon Application of CRM: Targeted Marketing Customer Clustering & Segmentation There are many benefits to using customer segmentation in CRM, including: Improved customer satisfaction Better resource allocation By tailoring your approach to each customer By focusing your resources on the customer segment, you can better meet their unique segments that are most valuable to your needs and preferences, resulting in higher business, you can optimize your use of customer satisfaction and loyalty. resources and achieve better results. Increased customer lifetime value More personalized customer experience By building stronger relationships with your By providing a more personalized customer customers, you can increase their lifetime experience, you can differentiate your value, which is the total value a customer is business from your competitors and create a likely to generate for your business over the more memorable and meaningful interaction course of their lifetime. with your customers. Examples of Application of AI in Targeted Marketing: More effective marketing Increased customer insights By targeting your marketing efforts more By gathering data on your customer Define the target group using clustering techniques effectively, you can achieve better results segments, you can gain deeper insights into with your campaigns, resulting in higher your customers' needs and preferences, Use GenAI tools like ChatGPT to define tactics accordingly conversion rates, increased customer which can inform your business decisions engagement, and improved return on and improve your overall approach to investment. customer relationship management. Website Design Some of the web design platforms: Platform Price Features Ease of Use WordPress Free plan available; Paid Extensive plugin ecosystem, customizable themes, User-friendly interface, plans start at $4/month blogging capabilities, e-commerce functionality (on paid slight learning curve Image result for further reading icon billed annually plans), SEO tools, and more websim ai: Wix Free plan available; Paid Drag-and-drop website builder, hundreds of templates, Very beginner-friendly, plans start at $14/month third-party app integrations, e-commerce capabilities, intuitive interface billed annually SEO tools, and more Weebly Free plan available; Paid Drag-and-drop website builder, customizable templates, Beginner-friendly, easy-to- plans start at $6/month e-commerce tools, SEO tools, and more use interface billed annually Google Sites Completely free with a Simple drag-and-drop interface, integration with Google Extremely easy to use, Google account services like Drive and Docs, basic website building especially for Google users capabilities suitable for small projects or internal use GitHub Free with a GitHub Hosting for static websites, version control with Git, Requires familiarity with account integration with GitHub repositories, support for custom Git and basic web Pages domains development skills Sentiment Analysis Sentiment analysis, also known as opinion mining, is the process of analyzing and categorizing opinions expressed in text to determine the sentiment conveyed, typically as positive, negative, or neutral. It utilizes natural language processing (NLP) techniques to extract and understand subjective information from vast amounts of textual data. Sentiment analysis is crucial in various domains, including business, marketing, social media, and customer feedback analysis. By automatically analyzing sentiments, organizations can gain valuable insights into public opinion, customer satisfaction, brand perception, and emerging trends. For example, sentiment analysis can help businesses Online Tools understand customer feedback on products or services, monitor social media sentiment towards their brand, and identify potential issues or opportunities for improvement. Its importance lies in its ability to provide actionable insights that drive strategic decision-making, enhance Open AI - ChatGPT customer experiences, and improve overall business performance. Real World Examples of AI in Marketing Image result for further reading icon Further reading: