z

 What is a chatbot?

 How do chatbots work?

 z

 A chatbot is a software application designed to simulate

human-like conversations through text or voice interactions.
 How do Chatbots Work?
z
 The working of chatbots can be broken down into a few steps:

1. User Input: The user interacts with the chatbot by typing a question or speaking a command.

2. Natural Language Processing (NLP): The chatbot uses Natural Language Processing (NLP), a branch of artificial
intelligence (AI), to analyze and understand the user’s input. It breaks down the language into components to identify the user’s
intent and extract key information like keywords, emotions, and context.

3. Intent Recognition: The chatbot matches the user's input to predefined intents (i.e., goals or actions the user wants to
achieve). This is done using machine learning algorithms or rule-based systems that compare the input with existing training
data or response patterns.

4. Response Generation: Based on the recognized intent, the chatbot generates an appropriate response. This can be done in
two ways:
1. Rule-based Chatbots: These rely on scripted responses and predefined rules to provide answers. They follow a fixed conversational
flow.

2. AI-powered Chatbots: These use machine learning models to generate dynamic responses, sometimes even learning and improving
from past interactions.

5. Output to User: The chatbot delivers its response back to the user in the form of text, voice, or action (e.g., booking an
appointment, retrieving information).

6. Learning and Improving (for AI-powered chatbots): AI-based chatbots can use user feedback and new input data to improve
over time, becoming more accurate and personalized in their responses.
z
 z

 Why did RAKT's chatbot take longer to respond

after implementing the complex model?
Why did RAKT's chatbot take longer to respond after implementing the
z complex model?

 RAKT's chatbot took longer to respond after implementing the complex model
because the newly developed natural language processing (NLP) model was
designed to better understand the nuances of human interactions.

 While this made the chatbot more capable of handling complex and ambiguous
queries, it also required more processing power and time to analyze each
input.

 As a result, the increased computational demand slowed down the chatbot's

response time, particularly during periods of high query volume.
Conversational Artificial Intelligence (AI)

refers to technologies that enable machines to engage in human-like interactions through

natural language, either via text or voice.

These systems use a combination of advanced techniques, such as Natural Language

Processing (NLP), Natural Language Understanding (NLU), and machine learning,
to understand, process, and respond to human inputs in a way that mimics real conversation.
 •Natural Language Processing (NLP):Helps the AI understand and interpret the structure of
human language.

Natural Language Understanding (NLU) is a subfield of Natural Language Processing (NLP)

that focuses on enabling machines to understand and interpret human language in a meaningful
way.

Unlike basic NLP, which deals with the processing and structuring of language,

NLU aims to comprehend the intent, context, and meaning behind the text or speech, much like
how humans understand language.
Why has the chatbot's response time increased after implementing a complex NLP
model?

What is the role of multiple machine learning models in determining the chatbot’s
response?

How does the "critical path" decision algorithm affect the chatbot's latency?

How can transforming unstructured text into machine-actionable information reduce

latency?

What is the function of the Natural Language Understanding (NLU) pipeline in improving
the chatbot’s performance?

What characteristics should the training dataset have to improve the chatbot’s response
time and understanding?
The "critical path" decision algorithm affects the chatbot's latency by determining the
shortest and most efficient sequence of machine learning models needed to process a user's
query and generate a response.

Each step along this path involves different models that solve specific tasks (such as
understanding context, identifying entities, or analyzing intent).

However, even though this path is optimized, the more models involved in the process, the
longer it takes to generate a response.

This creates dependencies between models, which can increase the overall latency or delay
in the chatbot’s response, especially when the system needs to handle complex inputs or high
volumes of queries.
•include
Why is it important for natural language processing
systems to incorporate all five types of analysis—
lexical, syntactic, semantic, discourse, and
pragmatic ?
1. What is the role of Recurrent Neural Networks (RNNs) in processing variable-length sequential data?
2. How do RNNs maintain a memory of previous inputs during sequence processing?
3. Why is it important to split the dataset into training, validation, and test sets when training an RNN?
4. What are the potential risks of not splitting the dataset appropriately?
5. How do hyperparameters like learning rate and the number of hidden layers affect the performance of an RNN?
6. What challenges might arise when tuning hyperparameters for optimal performance?
7. What is Backpropagation Through Time (BPTT) and how does it differ from regular backpropagation?
8. What specific challenges does BPTT address in RNNs?
9. What is the vanishing gradient problem and how does it impact the training of RNNs using BPTT?
[Link] does the vanishing gradient problem affect the learning of long-term dependencies in RNNs?
[Link] are some methods to overcome this issue in training deep RNNs?
[Link] is Long Short-Term Memory (LSTM) and how is it designed to solve the vanishing gradient problem?
13. How does the three-gate mechanism (input gate, forget gate, output gate) in LSTMs control the flow of
information?
[Link] role does the memory cell state play in LSTM networks?
[Link] do LSTMs retain or forget information over time, and how is this helpful for processing sequential data?
Deep learning is a subset of machine learning that uses
neural networks with many layers (hence "deep") to analyze
and learn from large amounts of data.

Research work: Key Concepts of Deep Learning

Application of Deep Learning

An Artificial Neural Network (ANN) is a computational
model inspired by the way biological neural networks in the
human brain work.

ANNs are at the heart of many modern AI and machine

learning applications, especially deep learning.

They are used to model complex patterns, recognize

features in data, and make predictions or decisions based on
that data.
[Link]

[Link]

[Link]
How do the memory cell state, input gate, forget gate, and output gate work together in

LSTM networks to selectively retain or forget information over time, and how do these

mechanisms improve the handling of long-term dependencies in sequence data

compared to traditional RNNs?

[12 marks]
Briefly explain all Key Factors that Make Transformers More Powerful Than LSTMs. [4]

How do Transformer Neural Networks serve as a powerful alternative to LSTMs in handling

long-range dependencies in sequence data, and what are the key architectural differences
that enable Transformers to outperform LSTMs in tasks such as natural language processing
and time series forecasting?
[12 marks]