View Full Microsoft AI-900 Exam Dumps and Practice Test Dumps.

 

Question 361

A company wants an AI system to identify the names of people, organizations, and locations mentioned in customer emails. Which natural language capability should be used?

1. Entity recognition
2. Sentiment analysis
3. Speech synthesis
4. Image classification

Correct Answer: 1

Explanation

Entity recognition identifies specific types of entities within text, such as people, organizations, locations, dates, or other supported categories. In this scenario, the company wants to extract names of people, organizations, and locations from customer emails, making entity recognition appropriate. Sentiment analysis determines the emotional tone of text, while speech synthesis converts written text into spoken audio. Image classification works with visual content rather than written language. Therefore, entity recognition is the suitable natural language processing capability for extracting these named entities.

Question 362

A customer service application needs to determine whether a message expresses satisfaction, dissatisfaction, or a neutral opinion. Which capability should be used?

1. OCR
2. Sentiment analysis
3. Object detection
4. Regression

Correct Answer: 2

Explanation

Sentiment analysis evaluates text to determine the expressed opinion or emotional tone. A customer service application can use it to identify whether feedback is positive, negative, or neutral and then use that information for reporting, routing, or customer experience analysis. OCR extracts text from images, object detection identifies objects in visual content, and regression predicts numerical values. Since the application needs to determine the emotional or opinion-based nature of written customer messages, sentiment analysis is the appropriate capability.

Question 363

An organization wants to create a chatbot that answers employee questions using a collection of frequently asked questions and internal answers. Which Azure AI capability is designed for this type of workload?

1. Custom question answering
2. Image segmentation
3. Speech recognition
4. Regression

Correct Answer: 1

Explanation

Custom question answering is designed to create applications that provide answers to user questions from a defined knowledge base. An organization can populate the knowledge base with frequently asked questions and their corresponding answers and use it as the foundation for a conversational application. Image segmentation analyzes parts of images, speech recognition converts spoken language into text, and regression predicts numerical values. Because the requirement is to build a question-and-answer experience from organizational knowledge, custom question answering is the appropriate capability.

Question 364

A business wants to automatically create a shorter version of lengthy customer feedback while preserving its main points. Which natural language capability is most appropriate?

1. Language detection
2. Entity recognition
3. Text summarization
4. Object detection

Correct Answer: 3

Explanation

Text summarization creates a shorter representation of longer written content while attempting to preserve important information and main ideas. It can help businesses process large volumes of customer feedback more efficiently by producing concise summaries for employees or downstream systems. Language detection identifies the language of text, entity recognition extracts named entities, and object detection works with images and video. Since the requirement is specifically to shorten lengthy written feedback while retaining its key information, text summarization is the appropriate capability.

Question 365

A document-processing solution receives scanned invoices as image files. Which capability can convert the printed or handwritten text in those images into machine-readable text?

1. OCR
2. Clustering
3. Forecasting
4. Sentiment analysis

Correct Answer: 1

Explanation

Optical character recognition, commonly called OCR, extracts text from images and makes that information available to software for further processing. It can be useful when invoices, forms, receipts, or other documents are provided as scanned images rather than structured digital text. Clustering groups similar data, forecasting predicts future values, and sentiment analysis evaluates opinions in language. Therefore, when the primary requirement is to recognize characters and words appearing in scanned invoice images, OCR is the appropriate AI capability.

Question 366

A company needs to determine the exact coordinates of text appearing inside a photograph so that the text can be highlighted on the screen. Which information from OCR is particularly useful?

1. Sentiment scores
2. Bounding regions
3. Audio transcripts
4. Regression values

Correct Answer: 2

Explanation

OCR can provide recognized text together with information describing where the text appears within an image. Bounding regions or similar positional information can be used by an application to draw boxes around detected text or highlight specific areas on the screen. Sentiment scores describe opinions in language, audio transcripts contain converted speech, and regression values represent numerical predictions. Therefore, when an application needs to locate recognized text visually within a photograph, bounding-region information is particularly useful.

Question 367

A developer wants an AI service to generate a natural-sounding spoken version of a written announcement. Which capability should be selected?

1. Speech-to-text
2. Text-to-speech
3. Language detection
4. Entity recognition

Correct Answer: 2

Explanation

Text-to-speech converts written text into spoken audio using an AI-generated voice. It can be used to create spoken announcements, accessibility features, virtual assistants, or audio versions of written content. Speech-to-text performs the reverse operation by converting spoken audio into text. Language detection identifies the language of written or spoken input, while entity recognition identifies meaningful entities in text. Because the source in this scenario is a written announcement and the desired result is natural-sounding speech, text-to-speech is the correct capability.

Question 368

A multilingual application must convert spoken French into spoken English while a user is talking. Which capability is most relevant?

1. Speech translation
2. Image classification
3. Key phrase extraction
4. Document classification

Correct Answer: 1

Explanation

Speech translation is designed to translate spoken language from one language into another. A multilingual application can use it to process spoken French and produce an English translation, potentially supporting real-time or near-real-time communication scenarios. Image classification analyzes visual content, key phrase extraction identifies important phrases in written text, and document classification assigns categories to documents. Since the application needs to translate spoken language between two languages, speech translation is the most relevant capability.

Question 369

A company has thousands of business documents and wants to create a searchable solution that can discover useful information across those documents. Which Azure service is designed for AI-powered search scenarios?

1. Azure AI Search
2. Azure Machine Learning
3. Azure AI Speech
4. Azure AI Vision

Correct Answer: 1

Explanation

Azure AI Search provides capabilities for building search experiences over organizational data and can support AI-powered scenarios such as indexing, semantic search, vector search, and knowledge discovery. It can help organizations make information within large collections of documents easier to locate and use. Azure Machine Learning focuses on developing and managing machine learning solutions, Azure AI Speech handles speech-related workloads, and Azure AI Vision processes images and visual information. Therefore, Azure AI Search is designed for searchable information and knowledge-discovery scenarios.

Question 370

An application needs to find documents based on the meaning of a user’s query rather than relying only on exact keyword matches. Which search approach is most suitable?

1. Vector search
2. File compression
3. OCR
4. Regression

Correct Answer: 1

Explanation

Vector search represents content and queries as numerical vectors and compares their semantic similarity. This allows a search application to find information that is conceptually related even when the exact words used in the query do not appear in the document. Traditional keyword matching may depend more heavily on matching terms. File compression reduces storage requirements, OCR extracts text from images, and regression predicts numerical values. Therefore, when the goal is to search based on meaning and semantic similarity, vector search is an appropriate approach.

Question 371

A search application combines traditional keyword matching with semantic vector-based retrieval to improve results. What type of search strategy is being used?

1. Batch inference
2. Hybrid search
3. Image classification
4. Speech synthesis

Correct Answer: 2

Explanation

Hybrid search combines different retrieval approaches, commonly keyword-based search and vector search, to improve the ability to find relevant information. Keyword matching can be effective when exact terms are important, while vector search can identify conceptually similar content even when wording differs. Combining both approaches can provide broader and more useful search results. Batch inference processes groups of predictions, image classification categorizes images, and speech synthesis generates spoken audio. Therefore, combining keyword and vector retrieval represents a hybrid search strategy.

Question 372

A machine learning team wants to divide its dataset into training, validation, and test portions. What is the primary purpose of the test dataset?

1. Train the model’s parameters
2. Tune hyperparameters repeatedly
3. Provide an independent final performance evaluation
4. Replace the training dataset

Correct Answer: 3

Explanation

A test dataset is used to provide an independent evaluation of a trained model after development decisions have been made. It helps estimate how well the final model may perform on previously unseen data. The training dataset is used to learn model parameters, while validation data can support model selection and hyperparameter tuning. Replacing the training dataset would prevent the model from learning from the intended training examples. Therefore, the test dataset is primarily used for an independent final assessment of model performance.

Question 373

Which statement best describes a machine learning label?

1. The target value the model learns to predict
2. The hardware used to train the model
3. The number of records in a dataset
4. The storage location of the model

Correct Answer: 1

Explanation

A label represents the target outcome that a supervised machine learning model is trained to predict. For example, in a dataset used to predict whether an email is spam, the label could indicate spam or not spam. Features provide the input information used by the model to make that prediction. Hardware, record counts, and storage locations are not definitions of labels. Therefore, the target value associated with training examples is the label that a supervised model attempts to learn and predict.

Question 374

A dataset contains customer membership types such as Basic, Standard, and Premium. What type of data is the membership type?

1. Continuous numerical data
2. Categorical data
3. Audio data
4. Image data

Correct Answer: 2

Explanation

Categorical data represents values that belong to distinct groups or categories. Basic, Standard, and Premium are predefined membership categories rather than continuous numerical measurements. Categorical variables can be useful as machine learning features, although they may require suitable encoding before being provided to certain algorithms. Continuous numerical data represents measurable quantities such as temperature or income, while audio and image data are different data modalities. Therefore, customer membership type is an example of categorical data.

Question 375

A fraud-detection model is evaluated using a confusion matrix. What does the matrix help the team understand?

1. The number of CPU cores available
2. The distribution of classification outcomes
3. The storage capacity of the dataset
4. The model’s source code structure

Correct Answer: 2

Explanation

A confusion matrix summarizes the outcomes of a classification model by comparing predicted classes with actual classes. It can show counts of true positives, true negatives, false positives, and false negatives. These results help teams understand where a classifier is making correct predictions and where different types of errors occur. A confusion matrix does not describe CPU availability, dataset storage capacity, or source-code organization. Therefore, it is primarily useful for understanding the distribution of classification outcomes and evaluating classifier performance.

Question 376

A model evaluation shows that a classifier has many false negatives. In a medical screening scenario, what does a false negative represent?

1. A healthy person correctly identified as healthy
2. A positive case incorrectly predicted as negative
3. A negative case incorrectly predicted as positive
4. A positive case correctly identified as positive

Correct Answer: 2

Explanation

A false negative occurs when the actual condition is positive but the model predicts the negative class. In a medical screening example, this could mean that a person who actually has the condition is incorrectly classified as not having it. A true negative is a negative case correctly identified as negative, while a false positive occurs when a negative case is incorrectly predicted as positive. A true positive is a positive case correctly identified. Therefore, a false negative represents an actual positive incorrectly predicted as negative.

Question 377

A machine learning team wants to automate the process of testing multiple algorithms and configurations to identify a suitable model for its data. Which Azure Machine Learning capability can help?

1. Automated ML
2. OCR
3. Speech translation
4. Sentiment analysis

Correct Answer: 1

Explanation

Automated ML in Azure Machine Learning can automate parts of the model development process by trying different algorithms, configurations, and preprocessing approaches based on the selected machine learning task and evaluation criteria. This can help teams identify promising models without manually testing every possible configuration. OCR handles text extraction from images, speech translation handles spoken-language translation, and sentiment analysis evaluates opinions in text. Therefore, when the goal is to automate experimentation across machine learning model configurations, Automated ML is the appropriate capability.

Question 378

An organization wants to expose a trained machine learning model so that applications can send requests and receive predictions immediately. What type of endpoint is appropriate?

1. Online endpoint
2. Archive endpoint
3. OCR endpoint
4. Batch-only endpoint

Correct Answer: 1

Explanation

An online endpoint is designed for real-time or near-real-time inference. Applications can send individual requests to the deployed model and receive predictions without waiting for a large batch-processing job to complete. This is useful for scenarios such as fraud scoring, recommendations, or interactive applications. Batch inference is more appropriate when predictions can be processed asynchronously over a collection of records. OCR and archive endpoints are not general terms for real-time machine learning model serving. Therefore, an online endpoint is appropriate for immediate prediction requests.

Question 379

A data science team needs to process millions of records overnight and does not require immediate responses for individual requests. Which inference approach is more suitable?

1. Online inference
2. Batch inference
3. Speech recognition
4. Image captioning

Correct Answer: 2

Explanation

Batch inference is designed for processing a large collection of inputs without requiring an immediate response for every individual request. A team can submit millions of records for processing and retrieve the predictions after the job completes. This approach is appropriate for scheduled workloads such as nightly scoring, reporting, or large-scale data processing. Online inference is intended for real-time or near-real-time requests. Speech recognition and image captioning are specific AI capabilities rather than general inference approaches for large record collections. Therefore, batch inference is suitable for this overnight workload.

Question 380

An AI system is being designed to protect sensitive customer information and prevent unauthorized access to personal data. Which responsible AI area is most directly related to this requirement?

1. Creativity
2. Privacy and security
3. Image resolution
4. Model compression

Correct Answer: 2

Explanation

Privacy and security are important responsible AI considerations when systems process sensitive or personal information. Organizations should protect data from unauthorized access, use appropriate security controls, and consider how personal information is collected, stored, processed, and shared. Creativity, image resolution, and model compression may be relevant to other technical or application requirements but do not directly address protection of sensitive customer information. Therefore, privacy and security are the responsible AI areas most directly related to safeguarding personal data and controlling unauthorized access.