Glossary of Terms
80/20 Rule
A principle stating that approximately 80% of effects come from 20% of causes, commonly used in chatbot analytics to identify the most frequently asked questions that drive the majority of user interactions.
Example: In a customer service chatbot, 80% of user queries might relate to just 20% of the available topics, such as password resets and order status.
A/B Testing
An experimental methodology that compares two versions of a chatbot or feature by randomly assigning users to different variants and measuring which performs better based on predefined metrics.
Example: Testing two different response styles in a chatbot to determine which generates higher user satisfaction scores.
Acceptance Rate
The percentage of chatbot responses that users find satisfactory or accept as helpful, typically measured through explicit feedback mechanisms or implicit behavioral signals.
Example: If 850 out of 1,000 chatbot responses receive positive feedback, the acceptance rate is 85%.
Access Policy
A set of rules that defines which users or roles can access specific resources, data, or functionality within a system.
Example: An access policy might specify that only managers can query salary information through a corporate chatbot.
AI Doubling Rate
The rate at which AI capabilities or computational power dedicated to AI research doubles over time, often observed to occur more rapidly than Moore's Law. months in recent years, far exceeding the doubling rate predicted by Moore's Law.
Example: The length of tasks that AI can complete with 50% probability of correctness has been doubling every 7 months.
Example: AI training compute has doubled approximately every 7 months in recent years, far exceeding the doubling rate predicted by Moore's Law.
Measuring AI Ability to Complete Long Tasks
AI Flywheel
A self-reinforcing cycle where user feedback improves AI systems, which in turn attract more users, generating more feedback and creating continuous improvement momentum.
Example: A chatbot that learns from user corrections becomes more accurate, leading to higher satisfaction and more users providing feedback.
AI Timeline
A chronological sequence of significant events, breakthroughs, and milestones in the development of artificial intelligence from its inception to the present.
Example: Key events include the Dartmouth Conference (1956), Deep Blue defeating Kasparov (1997), and GPT-3's release (2020).
Approximate Nearest Neighbor
An algorithm that efficiently finds vectors close to a query vector in high-dimensional space by trading perfect accuracy for significant speed improvements.
Example: FAISS uses approximate nearest neighbor to search millions of embeddings in milliseconds rather than hours.
Artificial Intelligence
The field of computer science focused on creating systems that can perform tasks typically requiring human intelligence, such as reasoning, learning, perception, and language understanding.
Example: A chatbot that understands customer questions and provides relevant answers demonstrates artificial intelligence capabilities.
Attention Mechanism
A neural network component that allows models to focus on relevant parts of input data when generating outputs, forming the core of transformer architectures.
Example: When translating "The cat sat on the mat," attention helps the model focus on "cat" when generating the translated subject.
Augmentation Step
The phase in RAG where retrieved documents or information are added to the user's original query to create an enriched prompt for the language model.
Example: Appending relevant FAQ entries to a user's question before sending it to the LLM for answer generation.
Authentication
The process of verifying the identity of a user, system, or entity attempting to access a resource or service.
Example: Requiring users to enter a username and password before accessing a chatbot with sensitive company information.
Authorization
The process of determining whether an authenticated user has permission to access a specific resource or perform a particular action.
Example: After logging in, checking whether a user is allowed to query financial data through the chatbot.
Boolean Search
A search technique that uses logical operators (AND, OR, NOT) to combine or exclude keywords when querying a database or search engine.
Example: Searching for "chatbot AND healthcare NOT telemedicine" to find healthcare chatbot articles excluding telemedicine topics.
Botpress
An open-source conversational AI platform that provides tools for building, deploying, and managing chatbots with visual flow design and natural language understanding.
Example: A developer uses Botpress to create a customer service bot with visual dialog flows and integrated NLU.
Byte Pair Encoding
A tokenization method that iteratively merges frequently occurring character pairs to create subword units, balancing vocabulary size with the ability to represent rare words.
Example: The word "understanding" might be tokenized as ["under", "stand", "ing"] using byte pair encoding.
Capstone Project
A culminating academic project that integrates and demonstrates knowledge and skills acquired throughout a course or program, typically producing a substantial deliverable.
Example: Building a complete GraphRAG-based chatbot system that answers domain-specific questions using a curated knowledge graph.
Chat History
A chronological record of messages exchanged between a user and a chatbot within a conversation session, used to maintain context and enable reference to previous interactions.
Example: Storing the last 10 messages so the chatbot can reference what the user asked earlier in the conversation.
Chat Interface
The user-facing component that displays conversations and accepts input, typically featuring message bubbles, input fields, and interactive elements.
Example: A web-based chat window with scrollable message history and a text input box at the bottom.
Chat Log
A persistent record of chatbot conversations, including user queries, bot responses, timestamps, and metadata, used for analysis, debugging, and compliance.
Example: Recording every interaction with timestamps and user IDs to analyze common questions and identify failures.
Chat Widget
An embeddable user interface component that provides chatbot functionality within a webpage or application, typically appearing as a small icon that expands into a chat window.
Example: A floating chat bubble in the bottom-right corner of a website that opens a chatbot when clicked.
Chatbot
A software application that simulates human conversation through text or voice interactions, designed to answer questions, provide assistance, or perform tasks.
Example: A customer service chatbot that helps users track orders and answer product questions on an e-commerce website.
Chatbot Career
Professional opportunities and career paths related to designing, developing, deploying, and maintaining conversational AI systems.
Example: Roles include Conversational AI Engineer, Chatbot Developer, and NLP Specialist.
Chatbot Dashboard
A visual interface that displays key performance indicators, metrics, and analytics about chatbot usage, performance, and user interactions.
Example: A dashboard showing daily active users, acceptance rate, average response time, and most frequently asked questions.
Chatbot Evaluation
The systematic assessment of a chatbot's performance using quantitative metrics and qualitative analysis to measure effectiveness, accuracy, and user satisfaction.
Example: Evaluating a chatbot by measuring F1 score for intent classification and conducting user satisfaction surveys.
Chatbot Framework
A software toolkit or platform that provides pre-built components, tools, and infrastructure for developing conversational AI applications more efficiently.
Example: Using LangChain to build a RAG chatbot without implementing vector search and prompt management from scratch.
Chatbot Metrics
Quantifiable measurements used to assess chatbot performance, user engagement, and system health, such as response time, accuracy, and user satisfaction.
Example: Tracking metrics like acceptance rate (85%), average response time (1.2s), and daily active users (5,000).
Chatbot Response
The output generated by a chatbot in reply to a user's query or message, which may include text, links, buttons, or other interactive elements.
Example: When asked "What's your return policy?", the chatbot responds with "We accept returns within 30 days of purchase."
Confusion Matrix
A table that visualizes the performance of a classification model by showing true positives, false positives, true negatives, and false negatives.
Example: Evaluating an intent classifier by counting correctly identified intents (true positives) versus misclassifications (false positives/negatives).
Context Length Limit
The maximum number of tokens that a language model can process in a single input, constraining how much conversation history and retrieved information can be included.
Example: GPT-3.5 has a context length limit of 4,096 tokens, limiting how much document content can be passed in a RAG system.
Context Window
The amount of text or number of tokens that a language model can process at once, including both input and output, determining how much context the model can consider.
Example: A model with an 8,000-token context window can process about 6,000 words of conversation history and documents.
Contextual Embedding
A vector representation of text where the same word produces different embeddings based on surrounding context, enabling more accurate semantic understanding.
Example: The word "bank" has different contextual embeddings in "river bank" versus "savings bank."
Continuous Improvement
An ongoing process of making incremental enhancements to a system based on feedback, data analysis, and evolving requirements.
Example: Regularly updating a chatbot's knowledge base and fine-tuning responses based on user feedback and query patterns.
Controlled Vocabulary
A standardized list of terms used consistently for indexing, tagging, and searching information, reducing ambiguity and improving search effectiveness.
Example: Using a controlled vocabulary where "automobile," "car," and "vehicle" all map to the standardized term "vehicle."
Conversation Context
The accumulated information from previous messages in a dialog session that helps the chatbot understand user intent and maintain coherent interactions.
Example: Remembering that a user asked about "laptops" so when they ask "What about gaming?" the bot knows to discuss gaming laptops.
Conversational Agent
An AI system designed to engage in dialog with users, understanding natural language input and generating appropriate responses to accomplish tasks or provide information.
Example: A virtual assistant that can book appointments, answer questions, and help navigate complex processes through natural conversation.
Coreference Resolution
The natural language processing task of identifying when different expressions in text refer to the same entity, enabling systems to track references across sentences.
Example: In "John went to the store. He bought milk," resolving that "He" refers to "John."
Corporate Nervous System
An organizational infrastructure of interconnected knowledge graphs, data systems, and AI agents that enables rapid information flow and decision-making throughout an enterprise.
Example: A unified system connecting customer data, product information, and employee knowledge to power intelligent chatbots and decision support tools.
Cosine Similarity
A metric measuring the similarity between two vectors by calculating the cosine of the angle between them, commonly used for comparing embeddings.
Example: Comparing a user query embedding with document embeddings to find the most semantically similar content, where 1.0 indicates identical direction.
Cypher Query Language
A declarative graph query language designed for querying and manipulating property graphs in Neo4j and compatible graph databases.
Example: MATCH (u:User)-[:ASKED]->(q:Question) RETURN u.name, q.text retrieves users and their questions.
Data Privacy
The protection of personal information from unauthorized access, use, or disclosure, ensuring individuals retain control over their data.
Example: Encrypting chat logs containing user queries and removing personally identifiable information before analysis.
Data Retention
Policies and practices governing how long data is stored before deletion, balancing operational needs, legal requirements, and privacy concerns.
Example: Retaining chat logs for 90 days for analysis, then automatically deleting them to protect user privacy.
Database Query
A request for data from a database, typically written in a structured query language like SQL, that specifies what information to retrieve or manipulate.
Example: SELECT * FROM orders WHERE user_id = 123 retrieves all orders for a specific user.
Dependency Parsing
A natural language processing technique that analyzes the grammatical structure of sentences by identifying relationships between words.
Example: In "The quick brown fox jumps," identifying that "quick" and "brown" both modify "fox."
Dialog System
A computer system designed to conduct conversations with users, managing dialog flow, maintaining context, and generating appropriate responses across multiple turns.
Example: A travel booking system that asks clarifying questions about dates, destinations, and preferences through multi-turn conversation.
Dialogflow
A Google-owned conversational AI platform that provides natural language understanding, intent recognition, and chatbot development tools with integration to Google services.
Example: Using Dialogflow to build a voice-activated assistant for Google Home devices.
Document Corpus
A collection of documents used as a knowledge source for training models, building search indexes, or providing context for retrieval systems.
Example: A corpus of 10,000 company policy documents used to train a corporate chatbot.
Document Frequency
The number of documents in a corpus that contain a specific term, used in TF-IDF calculations to weight term importance.
Example: If the word "chatbot" appears in 50 out of 1,000 documents, its document frequency is 50.
Dublin Core
A standardized metadata schema with 15 core elements for describing digital resources, widely used for cataloging and discovering information.
Example: Tagging documents with Dublin Core elements like Title, Creator, Subject, and Date for better searchability.
Edge
A connection between two nodes in a graph that represents a relationship, often labeled with a type and containing properties.
Example: In a knowledge graph, an edge labeled "AUTHORED" connects a Person node to a Book node.
Embedding Model
A neural network trained to convert text into dense vector representations that capture semantic meaning, enabling similarity comparisons and search.
Example: Sentence-BERT is an embedding model that converts sentences into 768-dimensional vectors for semantic search.
Embedding Vector
A numerical representation of text as a dense, fixed-length array of numbers that encodes semantic meaning in a high-dimensional vector space.
Example: The sentence "artificial intelligence" might be represented as a 512-dimensional vector like [0.23, -0.45, 0.12, ...].
Entity Extraction
The process of identifying and classifying specific entities (people, places, organizations, dates) from unstructured text.
Example: From "John ordered a laptop from Amazon on Monday," extracting Person:John, Product:laptop, Organization:Amazon, Date:Monday.
Entity Linking
The task of connecting extracted entity mentions to specific entries in a knowledge base, disambiguating references to the same real-world entity.
Example: Linking "Apple" in "Apple released new products" to the company Apple Inc. rather than the fruit.
Entity Type
A category or classification for entities recognized in text, such as Person, Organization, Location, Date, or Product.
Example: In named entity recognition, identifying "Microsoft" as an Organization type and "Seattle" as a Location type.
Euclidean Distance
A mathematical measure of the straight-line distance between two points in multi-dimensional space, used to compare vector similarity.
Example: Calculating the distance between two embedding vectors to determine how semantically different two sentences are.
External Knowledge
Information sources publicly available outside an organization, such as websites, encyclopedias, and open datasets, used to augment chatbot capabilities.
Example: Using Wikipedia articles as external knowledge to help a chatbot answer general knowledge questions.
F-Measure
A harmonic mean combining precision and recall that provides a single score balancing both metrics, with variants like F1 giving equal weight to each.
Example: A search system with 80% precision and 70% recall has an F-measure (F1) of approximately 74.7%.
F1 Score
A specific F-measure that equally weights precision and recall, calculated as 2 × (precision × recall) / (precision + recall), commonly used for evaluating classification and search systems.
Example: An intent classifier with 85% precision and 80% recall has an F1 score of 82.4%.
Factual Accuracy
The degree to which information generated or retrieved by a system matches verifiable, objective truth.
Example: Measuring what percentage of a chatbot's factual claims can be verified against trusted sources.
FAISS
Facebook AI Similarity Search, an open-source library for efficient similarity search and clustering of dense vectors, optimized for billion-scale datasets.
Example: Using FAISS to index 10 million document embeddings and retrieve the top 10 most similar results in milliseconds.
False Positive
An instance where a system incorrectly classifies something as belonging to a category when it does not, representing a type of error.
Example: A chatbot incorrectly identifying a general question as a "complaint" intent would be a false positive for the complaint category.
FAQ
Frequently Asked Questions—a collection of common questions and their answers, often used to train chatbots and provide self-service support.
Example: A company website's FAQ section containing 50 common customer questions about shipping, returns, and warranties.
FAQ Analysis
The systematic examination of frequently asked questions to identify patterns, gaps in knowledge coverage, and opportunities for improving chatbot responses.
Example: Analyzing 1,000 customer questions to discover that 30% ask about return policies, indicating a need for better documentation.
FastText
A library developed by Facebook for efficient text classification and learning word embeddings, particularly effective for handling rare words and morphologically rich languages.
Example: Using FastText to generate embeddings that understand that "unhappy" and "happiness" are related through shared subword components.
Feedback Button
A user interface element that allows users to rate or comment on chatbot responses, typically implemented as thumbs up/down or star ratings.
Example: A thumbs up/down button appearing below each chatbot response to collect user satisfaction data.
Feedback Loop
A cyclical process where system outputs are monitored, evaluated, and used to improve future performance, creating continuous learning.
Example: User thumbs-down ratings trigger review of poor responses, leading to knowledge base updates that improve future answers.
Frequency Analysis
The examination of how often specific events, queries, or terms occur, used to identify patterns and prioritize improvements.
Example: Analyzing which questions users ask most frequently to prioritize which answers to improve first.
Full-Text Search
A search technique that examines all words in stored documents to find matches, supporting features like phrase matching, stemming, and relevance ranking.
Example: Searching an entire document collection for pages containing "conversational AI" rather than just matching titles or tags.
GDPR
General Data Protection Regulation—a European Union law governing data protection and privacy, imposing strict requirements on how personal data is collected, stored, and processed.
Example: Ensuring chatbot logs are deleted within mandated timeframes and users can request deletion of their conversation history.
Generation Step
The final phase in RAG where a language model generates a response based on the user's query and retrieved contextual information.
Example: After retrieving relevant documents, the LLM generates a natural language answer synthesizing information from those sources.
GloVe
Global Vectors for Word Representation—a word embedding technique that learns vectors by factorizing word co-occurrence statistics from large corpora.
Example: GloVe embeddings trained on Wikipedia can represent semantic relationships like "king - man + woman ≈ queen."
Graph Database
A database that uses graph structures with nodes, edges, and properties to represent and store data, optimized for querying connected information.
Example: Neo4j stores customer relationships, product connections, and interaction histories as an interconnected graph.
Graph Query
A request to retrieve or manipulate data from a graph database, typically using specialized languages like Cypher that leverage graph structure.
Example: Finding all products a customer has viewed within 3 clicks of products they purchased.
GraphRAG Pattern
An architecture combining knowledge graphs with retrieval augmented generation, where structured graph data provides curated context for language model responses.
Example: A chatbot that traverses a corporate knowledge graph to find connected information before generating comprehensive answers.
Grep Command
A command-line utility for searching text using patterns, widely used for finding specific strings or regular expressions in files.
Example: Running grep -i "error" server.log to find all lines containing "error" (case-insensitive) in a log file.
Hallucination
When a language model generates plausible-sounding but factually incorrect or nonsensical information not grounded in its training data or provided context.
Example: A chatbot confidently stating a product has a feature it doesn't actually have, based on plausible inference rather than facts.
Index Performance
The efficiency and speed of search index operations, including query response time, update speed, and resource utilization.
Example: Optimizing an inverted index so that keyword searches return results in under 100 milliseconds even with millions of documents.
Intent Classification
The task of categorizing user inputs into predefined intent categories to understand what action or information the user seeks.
Example: Classifying "I want to return my order" as a "return_request" intent.
Intent Modeling
The process of designing and implementing a system to recognize and categorize user intentions from natural language input.
Example: Creating 20 intent categories like "check_balance," "transfer_money," and "report_fraud" for a banking chatbot.
Intent Recognition
The process of identifying the underlying goal or purpose behind a user's message in a conversational system.
Example: Recognizing that "Can you help me reset my password?" expresses a "password_reset" intent.
Internal Knowledge
Proprietary or confidential information specific to an organization, such as company policies, internal documentation, and customer data.
Example: Using internal product specifications and troubleshooting guides as a knowledge source for an employee support chatbot.
Inverted Index
A data structure mapping terms to the documents or locations where they appear, enabling fast full-text search by looking up terms rather than scanning documents.
Example: An index showing that "chatbot" appears in documents 5, 23, 47, and 102, allowing instant retrieval.
JavaScript Library
A collection of pre-written JavaScript code providing reusable functionality for specific tasks, such as UI components, API integrations, or data manipulation.
Example: Using the React Chatbot library to quickly implement a chat interface without building it from scratch.
Key Performance Indicator
A measurable value that demonstrates how effectively an organization or system is achieving key objectives.
Example: Using acceptance rate, average response time, and resolution rate as KPIs for chatbot performance.
Keyword Search
A search technique that matches exact words or phrases in queries against indexed content, without understanding semantic meaning.
Example: Searching for "chatbot tutorial" returns only documents containing those exact words, missing semantically similar content like "conversational AI guide."
Knowledge Graph
A structured representation of information as entities (nodes) and relationships (edges) that captures semantic connections between concepts, enabling advanced reasoning and retrieval.
Example: A corporate knowledge graph connecting employees, projects, skills, and documents to enable sophisticated question answering.
Knowledge Management
The systematic process of creating, sharing, using, and managing an organization's knowledge and information assets.
Example: Implementing a knowledge graph and chatbot system to make internal expertise easily discoverable across departments.
KPI
See Key Performance Indicator.
LangChain
An open-source framework for developing applications powered by language models, providing tools for prompt management, chains of reasoning, and integration with external data sources.
Example: Using LangChain to build a RAG chatbot that retrieves documents, formats prompts, and generates responses with minimal code.
Large Language Model
A neural network with billions of parameters trained on vast text corpora that can understand and generate human-like text for various natural language tasks.
Example: GPT-4 and Claude are large language models capable of answering questions, writing code, and engaging in complex conversations.
Lemmatization
The linguistic process of reducing words to their base or dictionary form (lemma), considering grammatical context to make more accurate reductions than stemming.
Example: Converting "running," "ran," and "runs" all to the lemma "run" while keeping "better" as "good."
LlamaIndex
A framework designed specifically for building retrieval augmented generation applications, providing tools for indexing, querying, and integrating external data with language models.
Example: Using LlamaIndex to create searchable indexes of company documents that a chatbot can query to answer questions.
Log Analysis
The examination of system logs to identify patterns, diagnose issues, track usage, and extract insights about system behavior and user interactions.
Example: Analyzing chatbot logs to identify the 20% of questions that account for 80% of user queries.
Log Storage
The infrastructure and practices for persisting system logs, including storage location, format, retention policies, and access controls.
Example: Storing chat logs in a secure database with encryption and automatic deletion after 90 days.
Logging System
Infrastructure for capturing, storing, and managing records of system events, user interactions, and operational data.
Example: Implementing a logging system that records every chatbot query, response, latency, and user feedback.
Message Bubble
A visual element in chat interfaces that displays individual messages, typically styled as rounded rectangles alternating between user and bot sides.
Example: User messages appear in gray bubbles on the left, while chatbot responses appear in blue bubbles on the right.
Metadata
Structured information that describes, explains, or provides context about other data, facilitating organization, discovery, and management.
Example: Document metadata including title, author, creation date, and tags used to improve search and organization.
Metadata Tagging
The process of assigning descriptive labels or attributes to content to enable better organization, searchability, and categorization.
Example: Tagging documents with subject categories, departments, and relevance scores to improve chatbot retrieval accuracy.
Moore's Law
The observation that the number of transistors on integrated circuits doubles approximately every two years, historically driving exponential growth in computing power.
Example: This doubling effect has enabled the massive computational resources required to train modern large language models.
Named Entity Recognition
The natural language processing task of locating and classifying named entities in text into predefined categories such as persons, organizations, and locations.
Example: From "Apple was founded by Steve Jobs in Cupertino," extracting Apple (Organization), Steve Jobs (Person), and Cupertino (Location).
Natural Language Processing
The field of artificial intelligence focused on enabling computers to understand, interpret, and generate human language in meaningful ways.
Example: NLP powers chatbots' ability to understand questions like "What's the weather?" and generate appropriate responses.
Natural Language to SQL
The task of converting questions posed in natural language into executable SQL database queries.
Example: Translating "How many orders did we have last month?" into SELECT COUNT(*) FROM orders WHERE month = LAST_MONTH.
Neo4j
A popular open-source graph database management system that uses Cypher query language and is optimized for storing and querying highly connected data.
Example: Using Neo4j to store a corporate knowledge graph connecting employees, projects, documents, and expertise.
NLP Pipeline
A sequence of processing steps that transform raw text into structured representations suitable for analysis or downstream tasks.
Example: A pipeline consisting of tokenization → stemming → part-of-speech tagging → named entity recognition.
Node
A fundamental unit in a graph database representing an entity, containing properties and connected to other nodes via edges.
Example: A Person node with properties like name: "Alice" and role: "Engineer" connected to Project nodes via WORKS_ON edges.
Node.js
A JavaScript runtime environment that executes JavaScript code outside browsers, commonly used for building server-side applications and chatbot backends.
Example: Building a chatbot API server using Node.js and Express to handle user queries and integrate with language models.
Ontology
A formal specification of concepts, categories, properties, and relationships within a domain, providing structured knowledge representation.
Example: A medical ontology defining relationships between diseases, symptoms, treatments, and body systems for a healthcare chatbot.
OpenCypher
An open-source specification for the Cypher graph query language, enabling portability of graph queries across compatible database systems.
Example: Writing OpenCypher queries that work with both Neo4j and Amazon Neptune graph databases.
Optimization
The process of improving system performance, efficiency, or effectiveness by refining algorithms, configurations, or resource allocation.
Example: Optimizing a chatbot's response time by caching frequent queries and tuning vector search parameters.
Organizational Knowledge
Collective information, expertise, best practices, and institutional memory accumulated within an organization.
Example: Internal wikis, process documents, and expert knowledge captured in a corporate knowledge graph.
Page Rank Algorithm
A link analysis algorithm that assigns importance scores to web pages based on the quantity and quality of links pointing to them, originally developed for Google Search.
Example: In a knowledge graph, using PageRank to identify the most central and important concepts based on their connections.
Parameter Extraction
The process of identifying and extracting specific values from natural language input to fill slots in structured queries or commands.
Example: From "Show me sales for Q3 2023," extracting quarter=Q3 and year=2023.
Parameterized Query
A database query template with placeholder values that are supplied separately, improving security and enabling query reuse.
Example: SELECT * FROM users WHERE id = ? where the ID value is provided separately to prevent SQL injection.
Pareto Analysis
An analytical technique based on the Pareto Principle that identifies the vital few factors accounting for the majority of effects.
Example: Analyzing chatbot queries to find that 20% of question types generate 80% of user traffic.
Part-of-Speech Tagging
The process of marking words in text with their grammatical categories (noun, verb, adjective, etc.) based on context and definition.
Example: In "The quick brown fox jumps," tagging "The" as determiner, "quick" and "brown" as adjectives, "fox" as noun, "jumps" as verb.
Performance Tuning
The systematic process of adjusting system parameters, configurations, and implementations to optimize speed, efficiency, and resource utilization.
Example: Tuning vector search parameters to balance retrieval accuracy against response latency in a RAG chatbot.
Personally Identifiable Info
See PII.
Personalization
The customization of user experiences based on individual preferences, history, and context to provide more relevant and tailored interactions.
Example: A chatbot that remembers a user's product preferences and suggests relevant items based on past interactions.
PII
Personally Identifiable Information—data that can identify a specific individual, such as names, addresses, phone numbers, or email addresses, requiring special privacy protection.
Example: Scrubbing user names and email addresses from chat logs before analysis to protect privacy.
Pinecone
A managed vector database service optimized for storing and querying high-dimensional embeddings at scale with low latency.
Example: Using Pinecone to store and search 100 million document embeddings for a large-scale chatbot application.
Private Documents
Confidential or restricted organizational content not intended for public access, often used as internal knowledge sources for chatbots.
Example: Employee handbooks, internal procedures, and confidential reports used to train an internal corporate chatbot.
Prompt Engineering
The practice of designing and refining input prompts to language models to elicit desired outputs, behaviors, or reasoning patterns.
Example: Crafting a system prompt like "You are a helpful technical support agent. Provide concise, accurate answers with step-by-step instructions."
Public Knowledge Base
A collection of information openly available to anyone, such as public documentation, wikis, or educational resources.
Example: Using publicly available product manuals and FAQs as a knowledge source for a customer-facing chatbot.
Query Description
A natural language or structured explanation of what information a query seeks, used to improve query understanding and execution.
Example: Describing "Find all orders placed this week" before converting it to SQL.
Query Execution
The process of running a query against a database or search system and returning results.
Example: Executing SELECT * FROM products WHERE category='electronics' and retrieving 247 matching products.
Query Frequency
The rate at which specific queries or query types occur, used to identify common questions and prioritize improvements.
Example: Discovering that "how do I reset my password" is asked 500 times daily, making it a high-priority question to answer well.
Query Optimization
The process of improving query performance through better indexing, query restructuring, or caching strategies.
Example: Adding an index on the "created_date" column to speed up queries filtering by date range.
Query Parameter
A value extracted from user input that fills a slot in a structured query template.
Example: Extracting "electronics" and "under $100" as parameters from "Show me electronics under $100."
Query Parser
A component that analyzes and interprets search queries, extracting keywords, operators, and structure to construct effective searches.
Example: Parsing "iPhone OR Android -tablet" into separate tokens with operators to execute a Boolean search.
Query Template
A reusable query structure with placeholders for variable values, enabling consistent query construction from natural language.
Example: "SELECT * FROM orders WHERE customer_id = {user_id} AND status = {status}" as a template for order status queries.
Question to Query Mapping
The process of translating natural language questions into structured database queries or search operations.
Example: Mapping "Who ordered the most products last month?" to a SQL query with aggregation and date filtering.
Question-Answer Pair
A structured unit consisting of a question and its corresponding answer, commonly used to train chatbots and build FAQ systems.
Example: Q: "What is your return policy?" A: "We accept returns within 30 days of purchase with original receipt."
RAG Limitations
The constraints and challenges of Retrieval Augmented Generation systems, including context length limits, retrieval quality dependency, and difficulty handling multi-hop reasoning.
Example: A RAG system struggling to answer questions requiring information synthesis from multiple unconnected documents.
RAG Pattern
See Retrieval Augmented Generation.
Rasa
An open-source conversational AI framework that provides tools for natural language understanding, dialog management, and custom chatbot development with full control.
Example: Using Rasa to build a customer service bot with custom entity extraction and dialog policies.
RBAC
See Role-Based Access Control.
React Chatbot
A chatbot user interface component built using the React JavaScript library, leveraging component-based architecture for reusable UI elements.
Example: Creating a React chatbot component that can be embedded in any React application with minimal configuration.
RDF
Resource Description Framework—a standard model for data interchange on the web using subject-predicate-object triples to represent information.
Example: Representing "Alice knows Bob" as the RDF triple (Alice, knows, Bob).
Regular Expressions
Patterns that describe sets of strings, used for text searching, matching, and manipulation with compact and powerful syntax.
Example: Using the regex \d{3}-\d{2}-\d{4} to match Social Security numbers in the format XXX-XX-XXXX.
Response Accuracy
The degree to which a chatbot's answers are correct, relevant, and truthful, measured against ground truth or expert evaluation.
Example: Evaluating 100 chatbot responses and finding that 87 are factually correct, yielding 87% accuracy.
Response Generation
The process of creating appropriate output messages in response to user inputs, either through template selection, retrieval, or language model generation.
Example: Using an LLM to generate a natural language answer based on retrieved documents and user query.
Response Latency
The time delay between receiving a user query and delivering a response, a critical performance metric affecting user experience.
Example: Measuring that 95% of chatbot responses are delivered within 2 seconds.
Response Quality
The overall effectiveness of a chatbot response considering accuracy, relevance, helpfulness, clarity, and appropriateness.
Example: Rating responses on a 1-5 scale based on whether they correctly and completely address the user's question.
Retrieval Augmented Generation
An architecture that combines information retrieval with language generation, where relevant documents are first retrieved and then used as context for generating responses.
Example: When asked about a company policy, first retrieving relevant policy documents, then using them as context for the LLM to generate an answer.
Retrieval Step
The first phase in RAG where relevant documents or information are identified and retrieved from a knowledge base based on the user's query.
Example: Using semantic search to find the top 5 most relevant FAQ entries for a user's question.
Reverse Index
See Inverted Index.
Role-Based Access Control
A security model that restricts system access based on user roles, with permissions assigned to roles rather than individual users.
Example: Allowing users with the "Manager" role to query financial data while restricting "Employee" role users to general information.
Search Index
A data structure that enables fast information retrieval by pre-processing and organizing content for efficient querying.
Example: Building a search index of 1 million documents that enables keyword searches to return results in milliseconds.
Search Performance
The efficiency of a search system measured by query response time, resource utilization, and scalability.
Example: A search system that handles 10,000 queries per second with average response time under 50ms demonstrates excellent performance.
Search Precision
The fraction of retrieved results that are relevant, calculated as (relevant retrieved) / (total retrieved), measuring result quality.
Example: If a search returns 20 results and 15 are relevant, precision is 75%.
Search Query
A request submitted to a search system specifying what information to find, typically consisting of keywords, phrases, or structured criteria.
Example: A user entering "chatbot frameworks for Python" into a search system.
Search Ranking
The process of ordering search results by relevance, importance, or other criteria to present the most useful results first.
Example: Ranking search results using a combination of keyword matching, semantic similarity, and PageRank scores.
Search Recall
The fraction of all relevant documents that are successfully retrieved, calculated as (relevant retrieved) / (total relevant), measuring completeness.
Example: If 100 relevant documents exist and the search retrieves 70 of them, recall is 70%.
Security
The protection of systems, data, and resources from unauthorized access, use, disclosure, disruption, or destruction.
Example: Implementing authentication, encryption, and access controls to protect sensitive chatbot interactions.
Semantic Search
A search approach that understands the meaning and intent behind queries rather than just matching keywords, using techniques like embeddings and vector similarity.
Example: Searching for "AI chatbot" and also retrieving results about "conversational agents" and "virtual assistants" based on semantic similarity.
Sentence Embedding
A dense vector representation of an entire sentence that captures its semantic meaning, enabling sentence-level similarity comparisons.
Example: Converting "I love programming" into a 384-dimensional vector that can be compared with other sentences for similarity.
Session Management
The process of maintaining state and context across multiple interactions within a conversation or user session.
Example: Tracking conversation history and user preferences throughout a chat session that may span multiple questions.
Slot Filling
The process of extracting specific pieces of information from user input to populate predefined fields or parameters needed for task completion.
Example: From "Book a flight to Paris on Friday," filling slots: destination=Paris, date=Friday.
SQL Query
A structured query written in Structured Query Language to retrieve, insert, update, or delete data from relational databases.
Example: SELECT name, price FROM products WHERE category='electronics' ORDER BY price DESC.
Stemming
The process of reducing words to their root form by removing suffixes, enabling matching of related word forms.
Example: Reducing "running," "runner," and "runs" to the stem "run."
String Matching
The process of finding occurrences of a character sequence (pattern) within another string, fundamental to text search.
Example: Finding all occurrences of "chatbot" in a document, regardless of whether it appears in "chatbots" or "chatbot's."
Subject-Predicate-Object
A three-part structure for representing knowledge, where a subject is related to an object via a predicate, forming the basis of triples in RDF and knowledge graphs.
Example: In the triple (Alice, works_at, Google), "Alice" is the subject, "works_at" is the predicate, and "Google" is the object.
Subword Tokenization
A tokenization approach that splits words into smaller units when necessary, balancing vocabulary size with the ability to represent rare or unknown words.
Example: Splitting "unhappiness" into ["un", "happiness"] to handle a rare word using known components.
Synonym Expansion
A query enhancement technique that adds synonymous terms to improve recall by matching alternative expressions of the same concept.
Example: Expanding "automobile" to include "car," "vehicle," and "auto" in the search query.
System Prompt
An initial instruction or context provided to a language model that defines its role, behavior, and constraints for subsequent interactions.
Example: "You are a knowledgeable customer service agent for TechCorp. Be helpful, concise, and professional in all responses."
Taxonomy
A hierarchical classification system that organizes concepts into categories and subcategories based on relationships and shared characteristics.
Example: A product taxonomy organizing items into categories like Electronics > Computers > Laptops > Gaming Laptops.
Team Project
A collaborative assignment where students work together to design, implement, and deliver a project, developing teamwork and communication skills.
Example: A team of four students building a chatbot for a local business, with members handling frontend, backend, NLP, and testing.
Term Frequency
The number of times a term appears in a document, used in TF-IDF calculations to measure term importance within that document.
Example: If "chatbot" appears 15 times in a document, its term frequency is 15.
Text Normalization
The process of transforming text into a consistent, canonical form by handling case, punctuation, spacing, and character variations.
Example: Converting "ChatBot," "chatbot," and "CHATBOT" all to "chatbot" for consistent matching.
Text Preprocessing
The initial steps of cleaning and transforming raw text into a standardized format suitable for analysis or processing.
Example: Removing special characters, converting to lowercase, and removing extra whitespace before tokenization.
Text Processing
The computational manipulation and analysis of textual data, including operations like parsing, searching, extracting, and transforming text.
Example: Processing customer reviews to extract sentiment, topics, and product mentions.
TF-IDF
Term Frequency-Inverse Document Frequency—a numerical statistic that reflects how important a word is to a document in a corpus by balancing term occurrence with rarity.
Example: Words like "the" have high term frequency but low IDF (common across documents), while domain-specific terms score high on both.
Thesaurus
A reference work listing words grouped by similarity of meaning, used in search systems for synonym expansion and query enhancement.
Example: Using a thesaurus to expand "fast" to include "quick," "rapid," and "speedy" in search queries.
Thumbs Up/Down
A simple binary feedback mechanism allowing users to indicate approval or disapproval of chatbot responses.
Example: Displaying thumbs up and down buttons below each response to collect quick satisfaction feedback.
Token
A basic unit of text resulting from tokenization, which may represent a word, subword, character, or other meaningful segment.
Example: The sentence "I love AI" might be tokenized into three tokens: ["I", "love", "AI"].
Tokenization
The process of breaking text into smaller units (tokens) such as words, subwords, or characters for processing by NLP systems.
Example: Splitting "Hello, world!" into tokens ["Hello", ",", "world", "!"].
Transformer Architecture
A neural network architecture based on self-attention mechanisms that processes entire sequences in parallel, forming the foundation of modern large language models.
Example: BERT, GPT, and Claude all use transformer architectures to understand and generate text.
Triple
A three-part knowledge representation consisting of subject, predicate, and object that expresses a single fact or relationship.
Example: The triple (Neo4j, is_a, GraphDatabase) represents the fact that Neo4j is a type of graph database.
True Positive
A correct prediction where a system correctly identifies something as belonging to a category when it actually does.
Example: A chatbot correctly classifying "I want a refund" as a "refund_request" intent is a true positive.
User Context
Information about a user's current situation, preferences, history, and environment that helps personalize and improve interaction relevance.
Example: Knowing a user previously asked about laptops helps the chatbot understand that "What about warranty?" refers to laptop warranty.
User Feedback
Information provided by users about their experience, satisfaction, or problems with a system, used to drive improvements.
Example: Collecting thumbs up/down ratings, comments, and behavioral signals like query reformulation.
User History
A record of a user's past interactions, queries, preferences, and behaviors used to personalize future experiences.
Example: Storing what products a user has viewed and purchased to provide personalized recommendations.
User Intent
The underlying goal or purpose a user wants to accomplish through their interaction with a system.
Example: Behind the query "Where's my package?", the user intent is to track an order shipment.
User Interface
The visual and interactive components through which users interact with a system, including layouts, controls, and feedback elements.
Example: A chatbot's user interface includes the message display area, input field, send button, and feedback controls.
User Permission
Authorization granted to a user specifying what actions they can perform and what resources they can access.
Example: Granting a user permission to query product information but not modify pricing data.
User Preferences
Settings, choices, and favored options that customize a user's experience based on their individual tastes and needs.
Example: A user's preference for concise answers rather than detailed explanations, stored to customize chatbot response style.
User Profile
A collection of information about a user including demographics, preferences, history, and permissions that enables personalization and access control.
Example: A profile containing user role, department, preferred topics, and interaction history.
User Prompt
The input or question provided by a user to a language model or chatbot, as opposed to system-generated prompts.
Example: A user typing "What is semantic search?" into a chatbot interface.
User Query
A question or request submitted by a user to a search system or chatbot seeking information or assistance.
Example: A user asking "How do I return an item?" in a customer service chatbot.
User Satisfaction
The degree to which users are pleased with their experience, often measured through surveys, ratings, and behavioral metrics.
Example: Measuring user satisfaction through post-interaction surveys showing 4.2 out of 5 stars average rating.
Vector Database
A specialized database optimized for storing, indexing, and querying high-dimensional vector embeddings with efficient similarity search capabilities.
Example: Using Pinecone or Weaviate to store and search millions of document embeddings for a chatbot.
Vector Dimension
The number of numeric values in a vector representation, determining the capacity to encode information and computational requirements.
Example: Sentence-BERT generates 768-dimensional vectors, meaning each sentence is represented by 768 numbers.
Vector Index
A data structure that organizes vector embeddings to enable fast similarity search, using techniques like locality-sensitive hashing or clustering.
Example: Building a FAISS index of document embeddings to enable millisecond similarity searches.
Vector Similarity
A measure of how close two vectors are in multi-dimensional space, indicating semantic similarity between their represented meanings.
Example: Calculating cosine similarity between query and document embeddings to find the most relevant results.
Vector Space Model
A mathematical model representing text as vectors in a high-dimensional space, where semantic similarity corresponds to geometric proximity.
Example: Representing documents and queries as vectors so that semantically similar texts cluster together in vector space.
Vector Store
A system for persisting and retrieving vector embeddings, providing the infrastructure for semantic search in RAG systems.
Example: Storing product description embeddings in a vector store to enable semantic product search.
Weaviate
An open-source vector database that combines vector search with structured filtering and integrates with various AI models for embedding generation.
Example: Using Weaviate to store documents with both embeddings for semantic search and metadata for filtering by date or category.
Word Embedding
A dense vector representation of a word that captures its semantic meaning based on context and usage patterns in text.
Example: The word "king" might be represented as a 300-dimensional vector that's mathematically closer to "queen" than to "car."
Word2Vec
A technique for learning word embeddings by predicting words from their context (CBOW) or context from words (Skip-gram), creating semantically meaningful vector representations.
Example: Word2Vec learns that "doctor" and "nurse" are similar by observing they appear in similar contexts.