Download [CCD] Certified Chatbot Developer Certification Exam Preparation and more Exams Technology in PDF only on Docsity!
Exam Preparation
Question 1. Which early chatbot is known for using pattern‑matching rules rather than true AI? A) Siri B) ELIZA C) Watson D) GPT‑ 3 Answer: B Explanation: ELIZA (1966) used simple pattern‑matching scripts like “DOCTOR” to simulate conversation, predating modern AI techniques. Question 2. In the evolution of conversational AI, what distinguishes Generative AI from rule‑based systems? A) Use of decision trees B) Ability to produce novel sentences C) Fixed response libraries D) Manual scripting of every utterance Answer: B Explanation: Generative AI (e.g., LLMs) creates new text based on learned patterns, unlike rule‑based bots that only return predefined replies. Question 3. An intelligent agent that perceives its environment and takes actions to maximize a reward is best described as a: A) Reactive system B) Finite state machine C) Reinforcement learning agent D) Heuristic search algorithm Answer: C
Exam Preparation
Explanation: Reinforcement learning agents learn policies that map perceptions to actions to maximize cumulative reward. Question 4. Which chatbot typology combines deterministic flows with natural language understanding? A) Rule‑based only B) AI‑powered only C) Hybrid model D) Stateless bot Answer: C Explanation: Hybrid models merge structured decision trees with flexible NLP, offering both control and adaptability. Question 5. A healthcare triage bot primarily assists patients by: A) Performing surgeries B) Scheduling appointments only C) Collecting symptoms and recommending next steps D) Billing insurance companies Answer: C Explanation: Triage bots gather symptom data, assess urgency, and suggest actions such as seeing a doctor or self‑care. Question 6. In text preprocessing, which operation reduces words to their root form while preserving meaning? A) Tokenization B) Stop‑word removal C) Stemming
Exam Preparation
B) Bag‑of‑words C) Word embeddings D) TF‑IDF Answer: C Explanation: Word embeddings (e.g., Word2Vec, GloVe) map words to continuous vectors where similar words are close in space. Question 10. Part‑of‑speech tagging is useful for: A) Determining user intent B) Identifying nouns, verbs, adjectives, etc. C) Translating text to another language D) Generating speech synthesis Answer: B Explanation: POS tagging labels each token with its grammatical role, aiding downstream tasks like parsing. Question 11. In dialogue mapping, a “node” typically represents: A) A database table B) A single user turn or bot response C) A machine‑learning model D) An API endpoint Answer: B Explanation: Nodes in flow diagrams denote conversational steps (utterances or actions) within a multi‑turn dialog. Question 12. Designing a bot’s “persona” primarily affects:
Exam Preparation
A) API latency B) Memory consumption C) Tone and language style D) Database schema Answer: C Explanation: A persona defines how the bot speaks (friendly, formal, humorous), aligning with brand identity. Question 13. If a user abruptly changes topic, the bot should: A) End the conversation immediately B) Ignore the new input and continue the original flow C) Use a digression handling strategy to acknowledge and possibly pivot D) Crash and restart Answer: C Explanation: Proper digression handling acknowledges the shift, answers if possible, or gracefully returns to the main flow. Question 14. A “fallback” response is used when: A) The bot successfully completes a task B) The user types a command the bot cannot parse C) The bot reaches maximum memory usage D) The API key expires Answer: B Explanation: Fallbacks provide graceful error messages when intent recognition fails or confidence is low.
Exam Preparation
Question 18. Seq2Seq models are especially suited for: A) Image classification B) Time‑series forecasting only C) Generating a response sequence given an input sequence D) Clustering unlabeled data Answer: C Explanation: Sequence‑to‑sequence architectures map an input token sequence (user utterance) to an output token sequence (bot reply). Question 19. Which deep‑learning framework uses “dynamic computation graphs” that are defined at runtime? A) TensorFlow 1.x (static graph) B) PyTorch C) Caffe D) Theano Answer: B Explanation: PyTorch builds graphs on the fly, facilitating debugging and variable‑length inputs. Question 20. Jupyter Notebook is primarily used for: A) Deploying production bots on cloud servers B) Interactive development and experimentation with code and data C) Managing DNS records D) Compiling Java applications Answer: B Explanation: Jupyter provides an interactive environment for writing, testing, and visualizing Python code, ideal for prototyping bots.
Exam Preparation
Question 21. In Dialogflow, the component that handles business logic after intent detection is called: A) Entity B) Fulfillment C) Context D) Slot Answer: B Explanation: Fulfillment can be a webhook that runs backend code to fetch data or perform actions. Question 22. Rasa’s “stories” are used to: A) Define user authentication policies B) Train NLU models only C) Describe example conversational paths for dialogue management D. Store user credentials Answer: C Explanation: Stories are annotated transcripts that teach the Rasa Core policy how to respond in various contexts. Question 23. The Microsoft Bot Framework SDK is available in which programming languages? A) Only C# B) Only JavaScript/Node.js C) Both C# and JavaScript/Node.js (and Python via extensions) D) Only Java
Exam Preparation
C) Lemmatization D) Stemming Answer: B Explanation: Stop‑word removal filters out high‑frequency, low‑information words. Question 27. In sentiment analysis, a “polarity” score of – 0.8 indicates: A) Strong positive sentiment B) Neutral sentiment C) Strong negative sentiment D) No sentiment detected Answer: C Explanation: Polarity ranges from – 1 (very negative) to +1 (very positive); – 0.8 is markedly negative. Question 28. Which of the following best describes “slot filling” in conversational design? A) Compressing audio files B) Collecting required pieces of information across multiple turns C) Encrypting user data D) Scaling server resources automatically Answer: B Explanation: Slot filling asks the user for missing entities (e.g., date, location) needed to complete a task. Question 29. When a bot needs to remember a user’s preferred language throughout a session, this is an example of: A) Intent detection
Exam Preparation
B) Entity extraction C) Contextual state management D) Model pruning Answer: C Explanation: Storing session‑level variables (like language preference) is part of context management. Question 30. The “softmax” function is commonly used in intent classification to: A) Normalize raw scores into probabilities across classes B) Reduce model size C) Increase training speed by skipping backpropagation D) Encode words as binary vectors Answer: A Explanation: Softmax converts logits into a probability distribution over possible intents. Question 31. Which evaluation metric is most appropriate for measuring the accuracy of an intent classifier on a balanced dataset? A) BLEU score B) Mean Squared Error C) Accuracy (percentage of correct predictions) D) Perplexity Answer: C Explanation: Accuracy directly reflects the proportion of correctly classified intents when classes are balanced.
Exam Preparation
Explanation: Rate limiting caps the number of requests per time window, protecting against abuse and denial‑of‑service attacks. Question 35. Which machine‑learning paradigm is most appropriate for continuously improving a chatbot’s responses based on live user feedback? A. Supervised learning with static data B. Unsupervised clustering C. Reinforcement learning with reward signals D. Transfer learning without fine‑tuning Answer: C Explanation: Reinforcement learning can adjust policies using real‑time rewards (e.g., user satisfaction scores). Question 36. When deploying a bot on Azure Bot Service, the primary component that handles message routing is called: A. Bot Connector B. Bot Emulator C. Bot Builder SDK D. Bot Analyzer Answer: A Explanation: The Bot Connector service routes messages between channels (e.g., Teams, web chat) and the bot’s backend. Question 37. Which of the following best illustrates “entity resolution” in a chatbot? A. Converting “NYC” to the canonical city name “New York City” B. Detecting the user’s intent to book a flight C. Generating a random greeting message
Exam Preparation
D. Logging conversation duration Answer: A Explanation: Entity resolution standardizes varied mentions of the same concept to a single, canonical form. Question 38. A “fallback intent” in Dialogflow is typically configured to trigger when: A. The user says “goodbye” B. The confidence score of the matched intent falls below a threshold C. The bot reaches the end of a story D. The user requests a joke Answer: B Explanation: The fallback intent captures low‑confidence matches, allowing the bot to ask for clarification. Question 39. Which of the following is NOT a typical step in the NLU pipeline? A. Tokenization B. Intent classification C. Speech synthesis D. Entity extraction Answer: C Explanation: Speech synthesis (text‑to‑speech) occurs after NLU; it is not part of language understanding. Question 40. In a conversational UI, “quick replies” are used to: A. Provide pre‑defined answer buttons that reduce typing effort B. Increase latency of the bot
Exam Preparation
A. Database schemas B. Intent examples C. Slots, entities, responses, and actions available to the bot D. Network firewall rules Answer: C Explanation: domain.yml lists the bot’s conversational elements such as intents, entities, slots, and response templates. Question 44. Which of the following best describes “few‑shot learning” in the context of chatbot development? A. Training a model with millions of labeled examples B. Fine‑tuning a pre‑trained model with only a handful of examples per intent C. Using zero training data D. Running inference on edge devices only Answer: B Explanation: Few‑shot learning leverages large pre‑trained models and adapts them using very limited new data. Question 45. When a bot asks “May I have your email address?” and the user replies “my email is [email protected]”, the bot must perform which NLP task to capture the address? A. Sentiment analysis B. Entity extraction (email entity) C. Language detection D. Part‑of‑speech tagging Answer: B Explanation: The bot extracts the email address as a specific entity from the user’s utterance.
Exam Preparation
Question 46. In conversation design, a “confirmation prompt” is used to: A. End the conversation abruptly B. Verify that the bot correctly understood user input before proceeding C. Increase the bot’s processing speed D. Randomly change the bot’s tone Answer: B Explanation: Confirmation prompts reduce errors by asking the user to confirm captured information. Question 47. Which of the following is a benefit of using a “state machine” for managing simple bot flows? A. Automatic natural language generation B. Predictable, deterministic transitions between states C. Real‑time sentiment analysis D. Dynamic learning from user feedback Answer: B Explanation: State machines provide clear, rule‑based pathways, making debugging and maintenance easier for simple dialogs. Question 48. The “BLEU” metric is primarily used to evaluate: A. Intent classification accuracy B. The quality of generated text compared to reference sentences C. Entity extraction precision D. API response time Answer: B
Exam Preparation
D. Adding more attention heads Answer: B Explanation: Quantization reduces precision of weights (e.g., from 32‑bit float to 8‑bit integer), shrinking model size and inference latency. Question 52. When a bot needs to fetch the current weather for a user‑provided city, the typical flow includes: A. Directly answering without external data B. Sending a webhook request to a weather API, parsing JSON, and responding C. Training a new intent on the fly D. Modifying the bot’s source code at runtime Answer: B Explanation: The bot invokes a webhook to retrieve real‑time weather data, then formats the response for the user. Question 53. In Dialogflow CX, “pages” are used to: A. Store user credentials B. Define distinct conversational states with associated routes and form fields C. Host static HTML content D. Manage version control of the bot Answer: B Explanation: CX pages represent steps in a flow, each containing prompts, form slots, and transition rules. Question 54. Which of the following best describes “transfer learning” in chatbot NLU? A. Training a model from scratch on a small dataset
Exam Preparation
B. Re‑using a pre‑trained language model and fine‑tuning it on domain‑specific intents C. Randomly initializing weights for each new intent D. Using unsupervised clustering to generate intents Answer: B Explanation: Transfer learning leverages knowledge from large corpora and adapts it to a specific task with limited data. Question 55. A “slot” in a conversational bot is: A. A hardware port for network connectivity B. A variable that stores a piece of extracted information needed to complete an action C. A type of activation function D. An encryption key for API calls Answer: B Explanation: Slots hold values like dates, locations, or amounts that the bot collects during the dialog. Question 56. Which metric is most suitable for evaluating the performance of an entity extraction system? A. F1‑score (harmonic mean of precision and recall) B. Perplexity C. ROC‑AUC D. Mean Absolute Error Answer: A Explanation: F1‑score balances precision (correctness) and recall (completeness) for named entity recognition.
