BADM 7402 - Exam 1
Collaborative approach disadvantages:
- Poor scalability - New user problem - New item problem - Recommendation depends on the history of the item
Clustering approaches:
- Random Forest - K-means Algorithm
What is the Knowledge Representation and Reasoning area?
- Second most important - Must have a way of describing the environment and drawing inferences from that representation
What is the best use for a content-based approach?
- Uses textual information - Can help recommend similar products based on past history
What are the different systems to support the elements of the data-information-knowledge-wisdom (DIKW) pyramid?
- Wisdom (experience) - WBS - Knowledge (synthesis) - KBS - Information (analysis) - DSS, GDSS MIS, OAS - Data (raw facts) - TPS
How does deep learning work?
- Works similar to machine learning - Construct the network - Train the network - Add another layer on top of the network - Retrain the network - Repeat with more layers
Difference between content-based and collaborative approaches?
-Content-based uses textual information - Collaborative uses user ratings
What is the search area?
-Fundamental technique - Looking for possible answers and courses of action - Either "blind" or "uninformed"
Machine learning attributes:
- Based on algorithms - Makes predictions based on data - Directed by user - Moderately compute heavy
What is semi-supervised learning?
- Between supervised and unsupervised - System looks for patterns and then asks the user if the pattern is correct
How does data mining differ from machine learning?
- Data Mining extracts knowledge from data - Machine Learning uses data to predict future outcomes
What is DSS, GDSS, MIS & OAS?
- Decisions Support System - Group Decision Support System - Management Information System - Office Automation System (All systems that help support information)
What are considerations for recommendation agents?
- Difficult to set up - They could be wrong - Maintenance - Context is important in the "user X items" space - Non-uniform concept, is highly contextual and task-oriented - Users sometimes need the motivation to rate items
Content-based approach advantages:
- Easy to understand - Don't need user data - New items are fine
What is supervised learning?
- Give system labeled training data - System learns rules that can map new inputs to correct output - Used for classification and for regression
What is unsupervised learning?
- Give system unstructured data - System asked to discover structures and patterns in the data - Used for clustering based upon similarity
Classification approaches:
- K-nearest neighbor - Decision Tree
What is reinforcement learning?
- Learn by trial-and-error - Program takes action, measures feedback and improves - Used for gaming and robotics
Dimensionality reduction approaches:
- Linear Discriminant Analysis - Partial Least Squares - Principle Components
Ensemble learning approaches:
- Majority vote: each algorithm gets a vote - Weighted vote: each algorithm's vote has a weight - Other complex combination functions
Deep learning attributes:
- Millions of data points - Based on neural networks - Learns to interpret data - Self directed - Output can be anything - Very heavy compute
Collaborative approach advantages:
- No need for professional knowledge - Performance improves with number of users - Can process complex items
Content-based approach disadvantages:
- Poor scalability - New user problem - Limited by the words extracted
Mistakes of deep learning:
1. Business mistakes 2. Data mistakes 3. Model mistakes
The different variations of recommendation agents:
1. Cluster Models 2. Item to Item Collaboration 3. Knowledge-Based RS
Examples of the different types of approaches to AI:
1. Cognitive Modeling (think humanly) 2. Turing Test (act humanly) 3. "Laws of Thought" (think rationally) 4. Rational Agent (act rationally)
What are the strategic alternatives for AI in business?
1. Current Infrastructure, New Business Model 2. New Technology, New Business Model 3. Current Infrastructure, Current Business Model 4. New Products, Current Business Model
What are the steps of the machine learning workflow?
1. Gather Data 2. Prepare Data 3. Test 4. Train Model 5. Test and Validate 6. Utilize Model
What are the 3 elements of a recommendation agent?
1. Items 2. Users 3. Transactions
Different types of machine learning problems:
1. Linear Regression 2. Classification 3. Dimensionality Reduction 4. Clustering
The 5 different elements of AI:
1. Search 2. Knowledge Representation and Reasoning 3. Planning 4. Learning 5. Interacting with the Environment
What are the 4 types of machine learning?
1. Supervised 2. Unsupervised 3. Semi-supervised 4. Reinforcement
What are the reasons why the Turing Test was critiqued?
1. Theological 2. Outcomes 3. Mathematical 4. Consciousness
What type of agent is AI?
A rational agent
What is Broad AI?
A system that can be applied to multiple contexts
Accuracy formula: Precision formula: Recall formula:
A: (TP + TN)/(TP + TN+ FP + FN) P: TP/(TP+FP) R:TP/(TP+FN)
Current Infrastructure, Current Business Model:
Add AI-powered features to your products
What is Narrow AI?
Built to perform a specific task or set of functions
All of the following are limitations of the collaborative based approach except: a. The new user problem b. The sparsity problem c. Poor scalability d. The new item problem
b. The sparsity problem
Discrete Data, Supervised Learning?
Classification
Discrete Data, Unsupervised Learning?
Clustering
What is the planning area?
Constructs a sequence of actions that achieves a given set of goals
What is the difference between deep learning & machine learning?
DL: uses neural networks to interpret large amounts of data, self-directed, very intense ML: uses algorithms to predict future decisions, directed by user, a score or classification
Creates a set of different decision trees and test each against the data, then combine the most successful trees?
Random Forest
Continuous Data, Supervised Learning?
Regression
What is the Interacting with the Environment area?
Systems must interact with our environment to enable intelligent behavior
New Infrastructure, Current Business Model:
Create AI-powered products in your product mix
New data is run through a decision system that will end at a class, that data is assigned that class?
Decision Tree
Continuous Data, Unsupervised Learning?
Dimensionality Reduction
What is Weak AI?
Executes tasks within a rule based program
What is the learning area?
For a system to act appropriately, it must be able to change its actions in the light of experience
The problem that natural language processing solves for organizations?
Getting computers to perform useful and interesting tasks with human languages
What is the fundamental way a recommendation agent works?
Takes known information about attributes or ratings on an item to recommend another
Why a systems transparency is important for recommendation agents?
Help users understand how the RS works
The problem that recommendation agents serve for organizations:
Information Filtering & Machine Learning
Similar to K-nearest neighbor, but there are no preset clusters?
K-Means Algorithm
New data is compared to training data, and data that is closest to training data is classified with it?
K-Nearest Neighbor
What is KBS?
Knowledge Based System- uses an experts knowledge to make decisions
The objective of machine learning is: a. To get rules from existing data b. To use algorithms from the data and past experience c. To extract knowledge from data d. To use a neural network to uncover hidden patterns in data
b. To use algorithms from the data and past experience
Word meaning, word level?
Lexical
What is the objective of AI systems if the AI technology is to act rationally?
Make logical decisions based on a set of given facts
New Infrastructure, New Business Model:
New AI company in your industry
Current Infrastructure, New Business Model:
Partner with an AI company
What has early research found regarding productivity in relation to AI?
Productivity Paradox - Labor productivity is decreasing with the advancement of AI
What is the fundamental technique of AI?
Search
What is Strong AI?
Self improves and learns through machine learning
For business, the best AI approach is to use... a.Complex algorithms b.A circuit model of the brain c.Predictive analytics d.A rational agent
The correct answer is: A rational agent
Systems that can be applied to many contexts is: a.Weak AI b.Broad AI c.Strong AI d.Narrow AI
The correct answer is: Broad AI
Over time, with experience... a.Humans become more limited in their abilities to process the world around them b.Humans make rational decisions c.Shrink their solution space d.Humans are able to take their knowledge and contextualized data
The correct answer is: Humans are able to take their knowledge and contextualized data
Which of the following explains the AI productivity paradox: a.The spillover effects have been minor b.A clear sense of what AI can do c.No companies have seen gains due to AI d.Implementation lags
The correct answer is: Implementation lags
Knowledge is: a.Data that has been shaped into a form that is meaningful and useful b.Information that has been organized and processed c.Similar to wisdom d.Streams of raw facts representing events
The correct answer is: Information that has been organized and processed
Deep learning: a.Learns to interpet data b.Uses a few thousand data points c.Learns to predict data d.Is directed by a user
The correct answer is: Learns to interpet data
Algorithms that allow computers to learn from examples without being programmed refers to: a.Deep Learning b.Artificial intelligence c.Machine learning d.Neutral networks
The correct answer is: Machine learning
If a firm decides to create AI-powered products in a product mix, which of the following strategic alternatives have they selected? a.New technology, new business model b.New products, current business model c.Current infrastructure, current business model d.Current infrastructure, new business model
The correct answer is: New products, current business model
An AI that self-improves through machine learning, based upon experience is: a.Strong AI b.Weak AI c.Narrow AI d.Broad AI
The correct answer is: Strong AI
Machine learning where a system is given labeled training data reflects: a.Reinforcement learning b.Supervised learning c.Semi-supervised learning d.Unsupervised learning
The correct answer is: Supervised learning
Which of the following are early business applications of AI: a.Tasks that are difficult to process due to excessive or complex data b.Tasks that are similar to the capabilities of humans c.Driverless cars d.Deliberate decisions in a slow moving environment
The correct answer is: Tasks that are difficult to process due to excessive or complex data
Which of the following is true? a.Humans always act rationally b.All intelligent behavior is mediated by logical deliberation c.The right thing for a system to do is that which is expected to maximize goal achievement d.There is one right action for any given set of circumstances
The correct answer is: The right thing for a system to do is that which is expected to maximize goal achievement
The AI approach that systems should think as a humans reflects which of the following approach to AI? a.Think Rationally b.Think Humanly c.Act humanly d.Act rationally
The correct answer is: Think Humanly
The Turing Test stated that: a.AI should think like a human b.Your interaction with a computer should be a unique experience c.Computers should act rationally d.You should be able to interact with a computer not knowing a computer is on the other end
The correct answer is: You should be able to interact with a computer not knowing a computer is on the other end
What is TPS?
Transaction Processing System- supports raw facts
What is WBS?
Wisdom Based System- can learn from experience
If a data scientist tries different methods to find the best approach to the underlying data, what would this be termed: a. Neural Network b. Ensemble Learning c. Machine Learning d. Deep Learning
b. Ensemble Learning
The algorithm for calculating the content based approach is known as: a. Vector-TF b. TF-IDF c. W-TF d. TF-DC
b. TF-IDF
Given the following facts, what is the accuracy level of the machine learning approach- True Positive: 10 cases; False Positive: 8 cases; False Negative: 4 cases; True Negative: 10 cases: a. 62.5% b. 50% c. 60% d. 37.5%
a. 62.5%
Recommendation agents can be improved by: a. Building up a relationship of trust between the user and the agent b. Neglecting to understand context and just focusing upon the results c. Assuming that everyone using the agent is honest d. Obscuring the algorithm
a. Building up a relationship of trust between the user and the agent
If you decide that you want to use the ratings that your users provide to create a recommendation agent, you are using: a. Collaborative based approach b. Content based approach c. Keyword based approach d. Hybrid approach
a. Collaborative based approach
The approach for a recommendation agent that is based upon the content of the documents is known as the: a. Content based approach b. Collaborative based approach c. Hybrid approach d. Keyword based approach
a. Content based approach
If a manager wanted to predict a FICO score using a set of numeric data, the approach would be: a. Regression b. Classification c. Clustering d. Dimensionality Reduction
a. Regression
All of the following are limitations of the content based approach except: a. The new item problem b. The new user problem c. Training of the classifier network d. Poor scalability
a. The new item problem
The no free lunch rule: a. The training and testing data are coming from the same source b. Not having a strong business case c. The deli cannot process your credit card during your lunch break d. The model is too simple to capture the underlying data
a. The training and testing data are coming from the same source
All the following factors influence a model in machine learning except: a. Missing feature values b. Uneven availability of instances c. A small number of features d. Few instances for a complex classification task
c. A small number of features
A recommendation agent can be relevant to information architecture for all of the following reasons except: a. It increases the findability of information b. It reduces search effort c. It allows organizational systems to remain the same d. It enhances browsing
c. It allows organizational systems to remain the same
The method where a new data point is mapped against existing data to find the closest location on an n-dimensional space, this would be: a. K-Means b. Decision Tree c. K-Nearest Neighbor d. Random Forest
c. K-Nearest Neighbor
Content based approaches use what space to represent the relationship between the user and the item: a. Euclidean distance b. User-item space c. Vector space d. User-object space
c. Vector space
A neural network: a. Matches an existing model against new data b. Uses a model defined by the user c. Should only be used in limited circumstances d. Alters the states, weights, and paths in a network to determine the best fit for the data
d. Alters the states, weights, and paths in a network to determine the best fit for the data
If a manager wanted to discover the different types of customers that the business had (and did not know the number before), the approach would be: a. Dimensionality Reduction b. Classification c. Regression d. Clustering
d. Clustering
A variation of a recommendation agent that combines similar items into a recommendation list is known as: a. Knowledge based recommendation agent b. Cluster model c. Item to user similarity d. Item to item similarity
d. Item to item similarity
When setting up a recommendation agent, one should consider: a. The difficulty of setting them up b. Ensuring motivation for users to rate will need to be thought through c. Maintenance of data needs to be part of ongoing processes d. That a recommendation agent will provide truth that is never wrong, thus altering decision making
d. That a recommendation agent will provide truth that is never wrong, thus altering decision making
Classification is: a. Using unsupervised learning on continuous data b. Using unsupervised learning on discrete data c. Using supervised learning on continuous data d. Using supervised learning on discrete data
d. Using supervised learning on discrete data