Bayes rule and KBS Flashcards
What is the method for using bayes rule?
1) Find all the P(H) (A prioris)
2) Find all the P(E|H) (evidence given hyp is true)
3) Find P(E) using formula P(E) = P(H1)P(E|H1) + P(H2)P(E|H2) + … + P(Hn)*P(E|Hn)
4) Work out P(H|E) for each H using formula, highest probability is answer
What is a knowledge based system?
A collection of knowledge taken from a human and stored in a such a way that a system can reason with it
What is the difference between machine learning and a knowledge based system?
Inputs
ML -> data, KBS -> facts and knowledge
Approach
ML -> training, KBS -> reasoning
Output
ML -> models, KBS -> new knowledge
What do knowledge based systems do?
Acquire knowledge
Represent knowledge
Apply the knowledge
What is the problem with building a knowledge based system?
- Human knowledge is complex
- Iterative procedure which is time consuming to maintain
What are the two ways we can represent knowledge in a KBS?
Natural language
-> Little uniformity
Predicate logic
-> uniformity
Explain the steps in resolution rule
1) Convert the knowledge base into CNF (conjunctive normal form so add disjunction)
2) Negate the statement and add it to KB
3) Unification (replace variables by concrete instance)
4) Resolution (iteratively eliminate contradictions until and empty clause is resolved (statement true) or a contradiction exists (negated statement is true)
What is natural language processing?
How machines understand human english
What are the different approaches to natural language processing?
Symbolic based
-> rely on rules and logic
Statistics based
-> analyse language patterns to predict words
ANN based
-> handle sequential data with memory
What is meant by the pipeline of natural language processing?
Series of steps to transform raw text into a formal suitable for a machine to understand
Explain the steps in pipelining
Test preprocessing, Feature engineering and Applications
Explain the test preprocessing steps in pipelining?
Normalisation
-> Convert text to standard form (lowercase, no punctuation)
Tokenisation
-> Break down text into words
Stopword removal
-> Filter out common words like “the”, “is”, “and”
Stemming
-> reduce words to their route like “running” to “run”
Explain feature engineering in the pipeline
Part of speech tagging
-> Assign each word its function (noun, verb, adjective etc.)
Named entity recognition
-> Identify named entities (person, number, organisation, location etc.)
Applications of pipelining
Chatbots
Sentimental analysis
Autocorrection
Machine translation
Plagiarism check
