UNIT V: Architectures & Applications

1)Explain the DeepQA architecture with an example.

DeepQA Architecture
DeepQA is the architecture behind IBM Watson, developed to process natural language questions
and deliver accurate, evidence-based answers. It is designed to handle open-domain question
answering, especially in complex, ambiguous, and unstructured data environments.
2. Key Features of DeepQA Architecture:
Component
Function
Question Analysis Parses the question to determine its type, focus, and keywords
Hypothesis Creates multiple possible answers (hypotheses)
Generation
Hypothesis Scoring Gathers evidence from various sources and scores each hypothesis
Evidence Retrieval Searches structured and unstructured sources like articles, databases,
etc.
Answer Ranking Combines scores to rank the hypotheses and selects the best answer
Machine Learning Train on large datasets to weigh evidence and improve over time
Models

3. Step-by-Step Flow of DeepQA:

1. Input: Natural language question (e.g., "Who wrote Hamlet?")
2. Question Analysis: Identifies type → person; key focus → Hamlet.
3. Hypothesis Generation: Possible answers → Shakespeare, Marlowe, etc.
4. Evidence Retrieval: Searches books, Wikipedia, plays, databases.
5. Hypothesis Scoring: Checks contextual match (e.g., “Hamlet written by Shakespeare”).
6. Ranking & Selection: Scores indicate William Shakespeare has the most supporting evidence.
7. Output: Final answer → "William Shakespeare", with confidence score.
4. Example:
Question:"Who was the first president of the United States?"

1
 Stage
Output
Input Question: "Who was the first president of the United States?"
Question Analysis Type: Person; Focus: first president, United States
Candidate Generation George Washington, Abraham Lincoln, Thomas Jefferson, etc.
Evidence Scoring Articles and history books point to George Washington
Ranking George Washington (highest score), others lower
Final Answer George Washington

5. Applications of DeepQA:
• Medical diagnosis support (Watson Health)
• Legal document analysis
• Customer support systems
• Academic and research assistance
• Financial services (fraud detection, investment insights)
2)What is UIMA? Describe its role in managing unstructured information.
Unstructured Information Management Architecture (UIMA)
Definition:
UIMA (Unstructured Information Management Architecture) is a software framework developed
by IBM and later donated to the Apache Foundation, designed for analyzing large volumes of
unstructured content such as text, audio, and video. It provides a standardized architecture for
building applications that extract structured information from unstructured sources.
Key Features:

2
 Feature
Description
Component-based UIMA applications are built by combining modular components called
architecture annotators.
Pipeline processing Data is processed through a pipeline where each component adds new
annotations.
Multi-modal support Works with different data types: text, speech, images, etc.
Scalability Supports large-scale processing of unstructured data.
Interoperability Supports integration with various tools and languages (Java, C++,
Python).

Core Components:
1. Collection Reader
• Reads and inputs raw unstructured data into the system.
2. Analysis Engine (AE)
• Processes data and adds metadata (annotations), such as entities, keywords, and concepts.
3. CAS (Common Analysis Structure)
• A data structure used to hold the text and all annotations generated during processing.
4. CAS Consumer
• Extracts final results from the CAS for storage, visualization, or further processing.
Role in Managing Unstructured Information:

3
 Function
Explanation
Information Identifies entities, relationships, and topics from text documents.
Extraction
Standardization Provides a common platform and design pattern for NLP applications.
Customization & Developers can reuse or modify components without changing the
Reuse entire system.
Interoperability Integrates with tools like Apache OpenNLP, Stanford NLP, or IBM
Watson.
Pipeline Processing Supports sequential processing where different tools contribute insights
in stages.
Big Data Readiness Used in big data environments (e.g., Hadoop, Spark) for large-scale NLP
tasks.

Example Application:
In a medical domain, UIMA can process patient records to:
• Extract diseases, treatments, and symptoms from clinical notes.
• Identify patterns across patient populations.
• Support decision-making by structuring critical information.
3)How is structured knowledge used in cognitive systems?
Use of Structured Knowledge in Cognitive Systems
What is Structured Knowledge?
Structured knowledge refers to information that is organized in a defined format—such as
databases, knowledge graphs, ontologies, or rule-based systems—making it easily searchable,
analyzable, and interpretable by machines.
Role of Structured Knowledge in Cognitive Systems:

4
 Aspect
Explanation
Knowledge Structured knowledge is used to represent facts, rules, and relationships
Representation between concepts.
Reasoning & Cognitive systems use it to draw logical conclusions and infer new
Inference information.
Contextual It helps systems understand the context of user queries and data inputs.
Understanding
Improved Decision- Enables evidence-based answers and recommendations by linking
Making structured facts.
Natural Language Enhances language models by connecting words and phrases to
Understanding structured concepts and ontologies.
Answer Generation & Supports generation of reliable answers with traceable sources and
Justification reasoning paths.
Disambiguation Helps resolve ambiguity in user inputs using predefined relationships
and categories.

Examples of Structured Knowledge Formats:

Format
Description
Relational Databases Tables with rows and columns (e.g., SQL databases)
Knowledge Graphs Networks of entities and relationships (e.g., Google Knowledge Graph)
Ontologies Formal definitions of domain-specific terms and their interrelations (e.g.,
OWL)
Taxonomies Hierarchical classification systems (e.g., product categories in e-
commerce)
Rule-Based Systems IF-THEN rules used for logic-based reasoning

Applications in Cognitive Computing:

5
 Domain
How Structured
Knowledge is Used
Healthcare Medical ontologies (like SNOMED CT) help interpret symptoms and
suggest treatments.
Finance Structured financial reports and market data aid in forecasting and risk
analysis.
Customer Support FAQs and knowledge bases are structured for fast and accurate query
resolution.
Education Knowledge graphs help organize curricula and link concepts across
subjects.

4)Discuss Applications of cognitive computing.

Applications of Cognitive Computing
Cognitive computing mimics human thought processes to solve complex problems using artificial
intelligence (AI), machine learning (ML), natural language processing (NLP), and data mining. Its
applications span a wide range of industries.
1. Healthcare
Application
Description
Disease diagnosis Cognitive systems like IBM Watson assist in diagnosing complex
diseases (e.g., cancer).
Personalized treatment Analyze patient data to suggest tailored treatment plans.
Medical imaging Recognize patterns in X-rays, MRIs, etc., for accurate diagnostics.
analysis

2. Finance

6
 Application
Description
Fraud detection Detects anomalies and suspicious transactions in real-time.
Risk assessment Evaluates creditworthiness and market risks using cognitive analysis.
Investment advisory Provides portfolio suggestions based on user goals and market trends.

3. Education
Application
Description
Intelligent tutoring Adapts content based on learner’s style, pace, and progress.
systems
Grading and feedback Automates evaluation of assignments and provides detailed feedback.
Learning analytics Identifies struggling students and predicts academic outcomes.

4. Customer Service
Application
Description
Virtual Answer customer queries using NLP and real-time learning.
assistants/chatbots
Sentiment analysis Understands customer emotions and adjusts responses accordingly.
Support ticket routing Routes issues to the appropriate department based on context.

5. Retail and E-Commerce

7
 Application
Description
Personalized Suggests products based on past behavior and preferences.
recommendations
Inventory management Predicts demand and automates stock control.
Visual search and Enhances customer experience using image recognition and AR.
virtual try-ons

6. Legal and Compliance

Application
Description
Contract analysis Extracts key terms and checks compliance in legal documents.
Case law research Helps lawyers find relevant precedents and rulings quickly.
Regulatory compliance Monitors activities against rules and standards.

7. Manufacturing and Industry 4.0

Application
Description
Predictive maintenance Identifies when machines will likely fail and schedules maintenance.
Quality control Uses vision and data to detect product defects.
Supply chain Makes logistics decisions using real-time and predictive analytics.
optimization

8. Human Resource Management

8
 Application
Description
Talent acquisition Analyzes resumes and matches candidates to job profiles.
Employee engagement Tracks sentiment and productivity through behavioral analytics.
Training and Provides adaptive learning modules for skill development.
development

5)Describe the process of building a cognitive application.

Process of Building a Cognitive Application
Building a cognitive application involves combining AI, machine learning, natural language processing,
and big data technologies to simulate human thought processes. Here's a step-by-step outline of the
process:
1. Define the Problem
• Clearly identify the business or user problem the cognitive application should solve.
• Understand the domain (e.g., healthcare, finance, education).
• Define the goals and expected outcomes.
2. Data Collection and Integration
• Gather relevant structured and unstructured data from various sources.
◦ Structured: Databases, spreadsheets.
◦ Unstructured: Emails, documents, social media, audio, images.
• Use data integration tools to consolidate the information.
3. Data Preprocessing
• Clean the data by removing duplicates, filling missing values, and correcting errors.
• Annotate and label unstructured data for training purposes.
• Format the data for compatibility with cognitive algorithms.
4. Choose Cognitive Technologies
• Select appropriate technologies and tools:
◦ Natural Language Processing (NLP)
◦ Machine Learning (ML)
◦ Speech and image recognition
◦ Knowledge representation and reasoning
• Use frameworks like IBM Watson, TensorFlow, OpenCog, or UIMA as needed.
5. Model Development
• Train machine learning models using historical data.
9
 • Apply NLP models for understanding language and intent.
• Use cognitive architectures (e.g., ACT-R, SOAR) for reasoning tasks.
6. System Integration
• Integrate cognitive models with application logic and user interfaces.
• Connect the system with APIs, databases, and external services.
• Ensure the system can process real-time or batch data inputs.
7. Testing and Validation
• Validate the cognitive application with test cases and real-world data.
• Evaluate performance metrics like accuracy, precision, recall, and user satisfaction.
• Refine models based on feedback and error analysis.
8. Deployment
• Deploy the application to the target environment (cloud, web, mobile).
• Ensure scalability and availability.
• Set up monitoring tools to track performance and user behavior.
9. Continuous Learning and Improvement
• Use feedback loops to update and retrain models.
• Allow the system to learn from new data and improve over time.
• Periodically review and fine-tune algorithms for accuracy and relevance.
6)Explain about Cognitive Modelling.
Cognitive Modelling:
Cognitive modelling is a field in cognitive science and artificial intelligence that aims to simulate
human thought processes using computational methods. It attempts to replicate how humans
perceive, think, learn, remember, and make decisions.
Definition
Cognitive modelling refers to the creation of mathematical or computational models that represent the
internal processes of human cognition.
Objectives of Cognitive Modelling
• To understand human cognitive processes (like memory, attention, reasoning).
• To predict human behavior in specific tasks.
• To build intelligent systems that behave like humans.
• To evaluate psychological theories through simulations.
Types of Cognitive Models

10
 Type
Description
Symbolic Models Based on logic and rules; use symbols to represent mental states (e.g.,
SOAR).
Connectionist Models Also called neural networks; model cognition through interconnected
nodes.
Bayesian Models Use probability theory to represent uncertainty in human reasoning.
Hybrid Models Combine two or more approaches (e.g., ACT-R mixes symbolic and sub-
symbolic).

Core Areas of Cognitive Modelling

1. Perception – Modeling how we interpret sensory input.
2. Attention – Simulating how humans focus on specific stimuli.
3. Memory – Understanding how information is stored and retrieved.
4. Learning – Creating models of skill acquisition and adaptation.
5. Language – Modeling language processing, understanding, and generation.
6. Decision-making – Representing how choices are made under various conditions.
Popular Cognitive Architectures
Architecture
Description
ACT-R Models human cognition based on production rules and memory
chunks.
SOAR Focuses on general intelligence and problem-solving through symbolic
reasoning.
OpenCog Aimed at Artificial General Intelligence (AGI) using a distributed
framework.
CopyCat Specializes in analogy-making and fluid intelligence modeling.

Applications of Cognitive Modelling

• User Interface Design – Predicting user behavior and improving usability.
• Education – Designing intelligent tutoring systems.
11
• Robotics – Creating autonomous agents that mimic human thought.
• Healthcare – Diagnosing cognitive impairments.
• AI Development – Building explainable and human-like AI systems.

12