Foundations of Text and Social Media Analytics

Introduction

Organizations increasingly compete through conversations, documents, messages, reviews, searches, and other forms of digitally recorded language. Customer complaints reveal service failures; employee comments indicate organizational climate; social posts signal emerging preferences; analyst reports and news articles influence investment decisions. These sources contain substantial business value, but they are not organized like conventional numerical databases. They are largely unstructured, contextual, multilingual, noisy, dynamic, and often produced by people whose intentions and identities may be uncertain.

Text and social media analytics provides the concepts, methods, and governance practices required to convert such data into reliable insight. It combines natural language processing, statistics, machine learning, information retrieval, network analysis, visualization, and domain expertise. The objective is not merely to count words or identify positive and negative comments. It is to support decisions such as improving products, detecting risks, understanding markets, prioritizing service interventions, and designing digital strategies.

This chapter establishes the foundations for the subject. It explains important terminology, traces the development of the field, distinguishes major analytical tasks, examines the characteristics of text and social data, and positions analytics within business intelligence and digital transformation. It also introduces a practical analytical lifecycle, discusses limitations and ethical obligations, and concludes with an applied business case.

Learning Objectives

After studying this chapter, you should be able to:

1. Define text analytics, social media analytics, natural language processing, and related concepts.
2. Explain the historical development of text and social media analytics.
3. Distinguish descriptive, diagnostic, predictive, and prescriptive analytical tasks.
4. Evaluate the distinctive characteristics and quality problems of text and social data.
5. Explain how text and social analytics contribute to business intelligence, strategy, and digital transformation.
6. Select suitable analytical approaches for common business problems.
7. Recognize methodological, ethical, legal, and managerial risks.

1. The Domain and Its Terminology

1.1 Data, information, insight, and action

Data are recorded observations, such as a customer review, a timestamp, a hashtag, or a support-chat message. Information is data organized and interpreted in relation to a question. Insight is a defensible understanding of a pattern, cause, opportunity, or risk. Action is a decision or intervention based on that insight.

For example, 4,000 product reviews are data. The finding that battery complaints increased sharply after a software update is information. The conclusion that the update may be causing avoidable returns is insight. Releasing a corrective patch and contacting affected customers is action. Analytics is valuable only when the chain from data to action is credible.

1.2 Text analytics

Text analytics is the systematic extraction of useful information, patterns, relationships, and evidence from written or transcribed language. It may involve structured outputs such as topics, entities, sentiment scores, categories, keywords, or relationships. Text analytics includes both automated methods and human interpretation.

Text data may be:

• Internal, such as emails, call transcripts, incident reports, survey comments, and knowledge-base articles.
• External, such as news, competitor websites, public reviews, and regulatory documents.
• User-generated, such as posts, comments, ratings, and discussion threads.

1.3 Social media analytics

Social media analytics is the collection, processing, analysis, and interpretation of data generated through social platforms and online communities. It includes textual content, user reactions, timestamps, hyperlinks, images, videos, location indicators, follower relationships, and engagement events.

Social media analytics is broader than sentiment analysis. It may investigate:

• what people are discussing;
• who influences whom;
• how information spreads;
• which audiences engage with a campaign;
• how a brand is positioned relative to competitors;
• whether a conversation represents a genuine customer issue or coordinated manipulation.

1.4 Related concepts

Natural language processing (NLP) is the field concerned with computational representation, processing, and generation of human language. Text analytics is an application-oriented use of NLP and related analytical methods.

Text mining generally refers to discovering useful patterns from large collections of text. The terms text mining and text analytics are frequently used interchangeably, although text analytics often emphasizes business interpretation and decision support.

Unstructured data lacks a fixed relational schema in its original form. Text is not truly structureless: it contains grammar, sequence, metadata, authorship, and interaction patterns. It is better described as weakly structured or variably structured data.

Semi-structured data combines content with explicit fields or tags, such as a social-media post with an author identifier, timestamp, language field, and engagement count.

Corpus means a collection of documents or text items assembled for analysis. A document may be a report, review, post, sentence, message, or transcript, depending on the analytical design.

Tokenization divides text into units such as words, subwords, sentences, or symbols. Lemmatization maps related word forms to a dictionary base form, while stemming applies a simpler rule-based reduction that may produce non-words.

Feature means a measurable representation used by an analytical model. A feature may be a word count, an embedding coordinate, a sentiment score, an entity occurrence, or a network measure.

Metadata describes data about a content item, including its source, creation time, language, author category, location, platform, and interaction history.

Taxonomy is a predefined classification structure. Ontology represents concepts and relationships in a domain, often with richer semantic meaning.

2. Historical Development

The field developed through several overlapping stages rather than a single linear progression.

2.1 Early computational language processing

Early language-processing research focused on machine translation, information retrieval, dictionaries, grammar rules, and symbolic representations. Systems depended heavily on manually specified linguistic rules. These methods were useful in constrained domains but difficult to scale across languages, writing styles, and ambiguous contexts.

Information retrieval systems later enabled users to search document collections using keywords and indexing. Boolean retrieval, in which documents were matched using logical operators such as AND and OR, established important foundations for enterprise search and digital libraries.

2.2 Statistical and machine-learning approaches

From the 1990s onward, statistical methods became increasingly important. Instead of encoding every language rule manually, systems learned relationships from labeled or observed data. Probabilistic models, vector-space retrieval, clustering, classification, and topic models supported spam detection, document categorization, recommendation, and search ranking.

A common document representation is the term-document matrix, in which rows represent documents and columns represent terms. The entry may be a term frequency, a binary occurrence indicator, or a weighted value. This representation is computationally useful but usually ignores word order and much of the semantic context.

2.3 Web and social-media expansion

The growth of the Web generated enormous volumes of searchable text. The emergence of blogs, forums, online reviews, and social networks then added interactive, timestamped, and networked communication. Organizations began monitoring brand mentions, customer complaints, campaign reactions, and competitor conversations.

The analytical problem changed from studying relatively stable document collections to studying fast-moving streams. A post is both content and a social event. Its meaning may depend on replies, quoted material, author communities, platform conventions, and the timing of publication.

2.4 Deep learning and foundation models

Neural networks and distributed representations improved the ability of systems to model similarity and context. Word embeddings represented words as vectors learned from usage patterns. Transformer-based architectures subsequently improved language understanding and generation by modeling relationships among tokens across a sequence. Large language models extended these capabilities across summarization, classification, extraction, question answering, and content generation.

These advances do not eliminate the need for business judgment. Models can reproduce bias, generate unsupported statements, misunderstand domain language, and behave differently when the data distribution changes. In professional settings, model outputs should therefore be treated as evidence requiring validation, not as unquestionable facts.

2.5 A historical perspective on business value

The commercial evolution can be summarized as follows:

3. Analytical Tasks

A sound project begins with a business question, not with a fashionable algorithm. The same corpus can support different tasks, and each task has different data, evaluation, and governance requirements.

3.1 Descriptive and exploratory tasks

Descriptive analytics answers “What happened?” Examples include message volume by day, frequently mentioned products, distribution of complaint categories, and engagement by campaign. Exploratory analysis examines patterns without assuming that the initial hypothesis is correct. It may reveal unusual terms, sudden bursts, emerging communities, or unexpected vocabulary.

Volume alone is not equivalent to importance. A topic with many low-value mentions may matter less than a small number of high-severity regulatory complaints. Counts should therefore be interpreted alongside reach, credibility, business impact, and representativeness.

3.2 Classification and routing

Classification assigns an item to one or more predefined categories. Examples include identifying complaint type, detecting spam, routing a support ticket, or labeling a post as product-related or not product-related.

Classification may be:

• Binary, with two classes;
• Multiclass, with one of several mutually exclusive classes;
• Multilabel, where one item may receive several labels.

Accuracy can be misleading when classes are imbalanced. For a binary classifier, precision and recall are defined as:

Here, TP is true positives, FP is false positives, and FN is false negatives. Precision measures the proportion of predicted positives that are correct. Recall measures the proportion of actual positives that are found. These measures assume a defined positive class and a known decision threshold.

The F1 score is the harmonic mean of precision and recall:

It is useful when both types of error matter and the two measures should be balanced. It should not be interpreted as a complete evaluation when costs, prevalence, or business consequences differ substantially.

3.3 Sentiment, emotion, and opinion analysis

Sentiment analysis estimates polarity or evaluative orientation, often using categories such as positive, negative, and neutral. Emotion analysis attempts to identify affective states such as anger, joy, fear, or disappointment. Opinion mining may identify the holder of an opinion, the target or aspect, and the expressed evaluation.

Aspect-based sentiment is often more useful than an overall score. A hotel review may be positive about staff, negative about cleanliness, and neutral about location. An aggregate sentiment label would hide this managerial distinction.

Sarcasm, irony, negation, mixed opinions, cultural expression, and domain-specific meanings create substantial difficulty. “This update is sick” may be positive in one community and negative in another.

3.4 Topic discovery and thematic analysis

Topic analysis identifies recurring themes or semantic areas in a corpus. Topic models can summarize large collections, while clustering groups similar documents. These are exploratory techniques: a discovered topic is a statistical pattern, not automatically a meaningful managerial category. Analysts must inspect representative documents, name topics carefully, and test their stability over time.

3.5 Information extraction

Information extraction converts text into structured fields. Important tasks include:

• Named-entity recognition: identifying persons, organizations, products, places, dates, and other entities.
• Relation extraction: identifying relations such as “supplier serves customer” or “product has defect.”
• Event extraction: identifying events, participants, dates, and consequences.
• Keyword and keyphrase extraction: identifying terms that summarize a document or collection.

Extraction is especially valuable when organizations need to integrate narrative documents into dashboards, case-management systems, or knowledge graphs.

3.6 Summarization and semantic search

Summarization produces a shorter representation of a document or collection. Extractive summaries select existing passages; abstractive systems generate new wording. The latter may be more fluent but introduce unsupported claims, omissions, or fabricated details.

Semantic search retrieves content according to meaning rather than exact keyword overlap. It is useful when users express a concept differently from the language used in the source documents. Retrieval systems should preserve links to evidence so that users can verify the answer.

3.7 Trend, anomaly, and crisis detection

Time-based analysis identifies changes in volume, language, sentiment, or network activity. An anomaly is an observation or pattern that differs materially from an expected baseline. Detection can support early warning, but a statistical anomaly is not necessarily a business crisis. It may result from a campaign, a platform change, a news event, a bot attack, or a data pipeline error.

3.8 Social network analysis

Social platforms contain relational data. Nodes may represent users, organizations, posts, or hashtags; edges may represent follows, replies, mentions, shares, or co-occurrences. Network analysis examines centrality, communities, brokerage, diffusion, and structural change.

Degree centrality counts direct connections, while betweenness centrality reflects how often a node lies on shortest paths between other nodes. A highly connected account is not automatically influential: influence also depends on audience quality, credibility, topic relevance, and the likelihood that others act on its messages.

4. Characteristics and Quality of Text and Social Data

4.1 High dimensionality and sparsity

A corpus may contain tens of thousands of possible terms, while each document uses only a small subset. The resulting representation is high-dimensional and sparse. Synonyms, spelling variation, morphology, abbreviations, and multilingual content increase the challenge.

4.2 Context dependence and ambiguity

Words do not carry a single stable meaning. Meaning depends on surrounding words, speaker intention, domain, culture, time, and interaction. A customer may say “great, another outage” sarcastically. A keyword match may classify this as positive unless context is modeled.

4.3 Velocity, volume, and volatility

Social content can arrive continuously and change rapidly. A model trained on last year’s vocabulary may fail when a new product, meme, political event, or crisis changes language. Systems therefore require monitoring for concept drift, data drift, and changes in platform behavior.

4.4 Veracity and representativeness

Online content may be inaccurate, duplicated, automated, coordinated, anonymous, or strategically posted. Public social data does not represent the entire customer population. Individuals who post frequently may differ systematically from silent customers. Analysts must not generalize from online conversations to all stakeholders without evidence.

4.5 Multimodality and multimodality-related context

Meaning may be distributed across text, images, video, audio, emojis, links, and metadata. A textual caption may be unintelligible without an attached image. Multimodal analysis can improve interpretation but introduces additional privacy, labeling, infrastructure, and bias concerns.

4.6 Data quality dimensions

Important quality dimensions include completeness, accuracy, consistency, timeliness, uniqueness, validity, and provenance. A missing timestamp can invalidate trend analysis. Duplicate reposts can inflate apparent demand. Deleted content may create survivorship bias. API restrictions may produce an incomplete sample.

A professional data audit should document source coverage, collection dates, access method, field definitions, missingness, deduplication rules, language detection, and known changes in platform policy.

5. The Text and Social Analytics Lifecycle

A practical lifecycle connects managerial purpose with technical execution.

1. Frame the decision. Define the business problem, decision owner, population, time horizon, and expected action.
2. Specify the construct. Clarify what is meant by concepts such as satisfaction, influence, risk, or churn intention.
3. Acquire and govern data. Establish lawful access, permissions, provenance, retention, and security.
4. Prepare the corpus. Remove or preserve duplicates deliberately, detect language, normalize text, handle missing fields, and document transformations.
5. Explore and label. Inspect samples, identify vocabulary, develop a coding scheme, and create reliable annotations where supervised learning is required.
6. Represent and model. Select features or embeddings, choose an analytical method, and establish a baseline.
7. Evaluate. Use appropriate metrics, error analysis, temporal or external validation, and business-cost analysis.
8. Interpret and communicate. Explain uncertainty, evidence, limitations, and implications for the decision.
9. Deploy and monitor. Track performance, drift, data quality, fairness, user feedback, and unintended effects.
10. Review and improve. Update the model and the business process as language, platforms, and objectives change.

The lifecycle is iterative. For example, error analysis may reveal that a supposedly simple sentiment task actually requires aspect-level labels or domain-specific rules.

6. Role in Business Intelligence, Strategy, and Digital Transformation

6.1 Business intelligence

Traditional business intelligence emphasizes structured measures such as revenue, cost, volume, and conversion. Text and social analytics enriches these measures with explanations and context. A dashboard may show that returns increased; customer narratives may reveal that packaging damage, rather than product quality, is the main driver.

Text can also serve as a leading indicator. Search queries, support conversations, and online discussions may change before formal sales or survey measures become available. Such indicators should be tested against reliable outcomes rather than assumed to be predictive.

6.2 Strategic analysis

Text and social data support environmental scanning, competitor intelligence, market segmentation, brand positioning, innovation discovery, and reputation management. Analysts can compare how customers describe competing products, identify unmet needs, and detect language associated with emerging use cases.

Strategic interpretation requires caution. High visibility is not the same as strategic importance, and competitor content may be deliberately misleading. Findings should be triangulated with sales, customer research, operational data, and expert knowledge.

6.3 Digital transformation

Digital transformation involves changes in capabilities, processes, operating models, and customer experiences—not simply the adoption of new software. Text analytics can become an organizational capability when it is embedded in workflows. Examples include automatically routing service cases, generating searchable knowledge from documents, assisting compliance review, and providing frontline employees with evidence from prior cases.

Successful transformation depends on process redesign, data governance, employee adoption, integration, and accountability. A technically accurate model can fail if employees do not trust it, if recommendations arrive too late, or if responsibility for action is unclear.

6.4 Managerial value and measurement

Potential benefits include lower service cost, faster response, improved retention, better product design, and reduced risk. Benefits should be linked to measurable outcomes. For example, a complaint-routing system may be evaluated by routing accuracy, resolution time, escalation rate, customer satisfaction, and cost per case—not accuracy alone.

7. Applied Case: Voice of the Customer in a Subscription Technology Firm

The following is an illustrative case designed for educational purposes.

Background

A subscription software firm receives support tickets, app-store reviews, community posts, and public comments. Customer churn has increased, but conventional exit surveys have low response rates. Senior management wants to identify emerging causes and prioritize product interventions.

Problem

The firm initially proposes a single monthly sentiment score. The analytics team argues that this would conceal differences among product modules, customer segments, and stages of the service journey. It also might confuse dissatisfaction with temporary frustration.

Analysis and approach

The team defines the analytical population as English-language customer-generated content associated with active or recently departed business accounts over a six-month period. It records source, timestamp, account segment, product module, and interaction type. Personally identifying information is masked before general analyst access.

A stratified sample is manually coded for issue category, product aspect, sentiment, urgency, and whether the author reports an observed problem or merely expresses a general opinion. Two analysts independently code part of the sample and resolve disagreements through a documented codebook. This step reveals that “slow” may refer to product performance, support response time, or billing delays.

The team then combines rules and machine learning. Rules identify explicit security terms and regulatory issues requiring immediate escalation. A multilabel classifier assigns issue categories, while aspect-based sentiment identifies whether the opinion concerns reliability, usability, pricing, or support. Topic discovery is used to identify themes not included in the initial codebook. A time-series dashboard compares issue rates with release dates, support workload, and account churn.

The evaluation uses a held-out temporal sample rather than only a random split. This better reflects future deployment because language and product conditions change over time. Errors are reviewed by segment and issue type. The team finds weaker performance for short posts, mixed-language messages, and sarcasm, so these cases are routed for human review.

Outcome

The analysis shows that overall negative sentiment is not the principal explanation for churn. A smaller but rapidly increasing cluster of enterprise customers reports authentication failures after a release. Separately, smaller customers express dissatisfaction with onboarding documentation. Product engineering addresses the authentication defect, while customer success redesigns onboarding materials. The company monitors subsequent issue rates, time to resolution, renewal behavior, and customer feedback.

Learning points

1. The business decision determined the analytical design.
2. Aspect-level analysis was more actionable than one overall sentiment score.
3. Human coding was necessary to define the construct and expose ambiguity.
4. Temporal validation was more realistic than relying exclusively on random sampling.
5. Automated escalation required governance because false negatives in security-related content could be costly.
6. Analytics created value only when connected to accountable operational owners.

8. Challenges, Ethics, and Governance

Privacy and lawful use

Public availability does not eliminate privacy obligations. Collection and use should have a lawful basis, a defined purpose, data minimization, retention limits, access controls, and appropriate safeguards. Sensitive attributes, private messages, location data, and inferred characteristics require heightened care. Analysts should avoid collecting information merely because it is technically accessible.

Consent and user expectations

Users may not expect their posts to be used for employee evaluation, credit decisions, or detailed profiling. Purpose limitation and transparent communication are essential. When feasible, organizations should aggregate results and avoid exposing identifiable quotations.

Bias and fairness

Language models may perform differently across dialects, languages, demographic groups, or communication styles. A moderation or risk model can disproportionately flag certain communities. Fairness evaluation should examine subgroup performance, error costs, and the consequences of deployment. Removing demographic variables does not guarantee fairness because language can act as a proxy.

Transparency and explainability

Decision makers need to know what data was used, what the model predicts, how reliable it is, and when human review is required. Explanations should be appropriate to the audience. A feature-importance chart may explain a model statistically but not justify a high-impact decision by itself.

Security and manipulation

Social data may contain malicious links, prompt injection attempts, coordinated campaigns, bots, and fabricated engagement. Systems that use retrieved text with generative AI should separate instructions from untrusted content and preserve source citations. Bot detection itself is uncertain and should not be treated as an infallible identity judgment.

Governance practices

A mature governance program maintains a data dictionary, model card or system description, annotation guidelines, version history, access logs, performance reports, incident procedures, and named accountability. High-impact uses should include human oversight, appeal mechanisms, and periodic independent review.

9. Best Practices for Postgraduate Analysts

• Begin with a decision and a measurable outcome, not with available data alone.
• Define the unit of analysis: document, sentence, author, account, conversation, or event.
• Separate exploration from confirmation; do not present an exploratory topic as proven causation.
• Establish a transparent baseline before adopting a complex model.
• Preserve provenance and representative examples for every important finding.
• Evaluate by time period, language, segment, and business consequence—not only by average score.
• Treat social-media volume as a sample of behavior, not a census of public opinion.
• Combine automated methods with domain experts and human review for ambiguous or high-risk cases.
• Monitor drift, platform changes, and changes in vocabulary after deployment.
• Communicate uncertainty and specify what action the analysis supports.

Contemporary Developments

Current practice is moving toward retrieval-augmented systems, multimodal models, privacy-preserving analytics, synthetic data for testing, and smaller domain-adapted language models. These developments can improve accessibility and productivity, but they also increase the importance of provenance, evaluation, and security. Generative systems are particularly useful for drafting summaries and assisting analysts, yet factual claims should be traceable to source evidence.

Another important development is the movement from isolated dashboards toward decision intelligence. In this model, text and social signals are integrated with operational data, experiments, workflow systems, and human judgment. The goal is not to automate every decision, but to improve the quality, speed, and accountability of decisions.

Chapter Summary

Text and social media analytics transforms language and online interaction data into evidence for organizational decisions. Text analytics focuses on extracting patterns and meaning from documents and messages, while social media analytics additionally studies interaction, diffusion, engagement, and networks. The field has developed from rule-based language processing through statistical learning, web-scale analysis, deep learning, and contemporary generative systems.

Major tasks include description, classification, sentiment and emotion analysis, topic discovery, information extraction, summarization, semantic search, anomaly detection, and network analysis. Text and social data are high-dimensional, contextual, rapidly changing, noisy, multimodal, and often unrepresentative. These characteristics make data quality assessment, careful construct definition, appropriate validation, and human interpretation essential.

The analytics lifecycle connects business framing, governed data acquisition, preparation, labeling, modeling, evaluation, communication, deployment, and monitoring. In business intelligence, text supplies explanations and early signals alongside structured metrics. In strategy, it supports market understanding, competitive analysis, innovation, and reputation management. In digital transformation, value arises when analytics is embedded in accountable processes rather than treated as an isolated technical project.

Ethical and governance responsibilities include privacy, lawful use, consent, fairness, transparency, security, and human oversight. The central professional principle is that analytical sophistication does not compensate for a poorly defined decision, biased data, weak validation, or irresponsible use.

Key Terms

Text analytics: Systematic extraction and interpretation of useful information from written or transcribed language.

Social media analytics: Analysis of content, interactions, networks, and behavior generated through social platforms.

Natural language processing: Computational processing and generation of human language.

Corpus: A collection of documents or text items assembled for analysis.

Tokenization: Division of text into words, subwords, sentences, or other units.

Lemmatization: Reduction of related word forms to a dictionary base form.

Feature: A measurable representation used by an analytical model.

Metadata: Descriptive information about a content item or interaction.

Classification: Assignment of data items to predefined categories.

Sentiment analysis: Estimation of evaluative polarity or attitude in text.

Aspect-based sentiment: Analysis of sentiment toward particular subjects or attributes.

Topic analysis: Identification of recurring themes or semantic areas.

Named-entity recognition: Identification of entities such as people, organizations, products, and places.

Semantic search: Retrieval based on meaning rather than exact keyword matching.

Anomaly: A pattern or observation that differs materially from an expected baseline.

Centrality: A measure of the structural importance of a node in a network.

Concept drift: Change over time in the relationship between inputs and the target being predicted.

Precision: The proportion of predicted positive cases that are truly positive.

Recall: The proportion of actual positive cases that are correctly identified.

Data provenance: A documented record of data origin, transformations, and use.

Review Questions

Short Answer Questions

1. What is the difference between text analytics and social media analytics?
2. Why can a high volume of online mentions fail to indicate high business importance?
3. Distinguish precision from recall.
4. What is aspect-based sentiment analysis?
5. Why is social-media data not necessarily representative of all customers?

Long Answer Questions

1. Explain the historical development of text and social media analytics from rule-based systems to generative AI.
2. Discuss the major analytical tasks in text analytics and identify a business use for each.
3. Explain the principal characteristics that make text and social data difficult to analyze.
4. Describe the complete analytics lifecycle and explain why business framing should precede model selection.
5. Evaluate the contribution of text and social analytics to business intelligence, strategy, and digital transformation.

Analytical / Application Questions

1. A retailer observes a sudden increase in negative social posts. Design an analytical investigation that distinguishes a genuine service crisis from a viral event, bot activity, or data-pipeline error.
2. A bank wants to use customer messages to predict account closure. Identify the data, modeling, validation, fairness, privacy, and governance issues that should be addressed before deployment.
3. A university receives thousands of student comments about online learning. Propose a method combining classification, aspect-based sentiment, topic discovery, and human review. Explain how you would evaluate whether the resulting insights improve decisions.

Multiple Choice Questions with Answers

Question 1

Which statement best defines text analytics?

A. The manual editing of documents for grammatical correctness

B. The systematic extraction and interpretation of useful information from language data

C. The measurement of internet connection speed

D. The storage of numerical data in spreadsheets

Correct Answer: B. The systematic extraction and interpretation of useful information from language data

Question 2

Which task assigns a document to one or more predefined categories?

A. Classification

B. Tokenization

C. Summarization

D. Network visualization

Correct Answer: A. Classification

Question 3

If a model identifies most actual security-related messages but also flags many harmless messages, which measure is likely high and which may be low?

A. High recall and potentially low precision

B. High precision and necessarily low recall

C. Low recall and necessarily high precision

D. Neither precision nor recall can apply

Correct Answer: A. High recall and potentially low precision

Question 4

Why is aspect-based sentiment often more actionable than an overall sentiment score?

A. It removes the need for human interpretation

B. It identifies sentiment toward specific attributes or issues

C. It guarantees that sarcasm is detected

D. It measures only the number of documents

Correct Answer: B. It identifies sentiment toward specific attributes or issues

Question 5

Which feature most clearly distinguishes social media analytics from analysis of a static document archive?

A. The use of spelling correction

B. The presence of relational and interaction data

C. The use of alphabetical ordering

D. The absence of timestamps

Correct Answer: B. The presence of relational and interaction data

Question 6

What is the main purpose of temporal validation in a deployed text model?

A. To ensure every document has the same length

B. To simulate performance on future data under changing conditions

C. To eliminate the need for labeled data

D. To guarantee causal conclusions

Correct Answer: B. To simulate performance on future data under changing conditions

Question 7

Which statement about online social data is most defensible?

A. It is always a representative census of public opinion

B. It is generally free from automated or coordinated activity

C. It may be valuable but can be biased, incomplete, and unrepresentative

D. It cannot contain metadata

Correct Answer: C. It may be valuable but can be biased, incomplete, and unrepresentative

Question 8

In a social network, what does betweenness centrality primarily capture?

A. The number of words used by a user

B. The frequency with which a node lies on shortest paths between other nodes

C. The sentiment polarity of a post

D. The age of an account

Correct Answer: B. The frequency with which a node lies on shortest paths between other nodes

Question 9

Which practice best supports responsible use of generative AI for text analysis?

A. Accepting every generated summary without checking sources

B. Removing all human review from high-impact decisions

C. Preserving evidence links and validating generated claims

D. Treating fluent language as proof of factual accuracy

Correct Answer: C. Preserving evidence links and validating generated claims

Question 10

A company finds many complaints about reliability but positive comments about support staff. Which method is most suitable for preserving this distinction?

A. Overall document length analysis

B. Aspect-based sentiment analysis

C. Alphabetical sorting of posts

D. Counting only positive words

Correct Answer: B. Aspect-based sentiment analysis