UPI Study › Courses › Ethics in Technology › Free lessons › Lesson 5.5 Free with a free account
Applying Data Mining Techniques to Understand Behavior and Trends on Social Platforms
Ethics in Technology · Chapter 5: Making Sense of Massive Datasets: Big Data, Mining, and the Privacy Trade-OffSection 5.5 · lesson 21 of 11614 min · free with a free account
Free preview
Create a free account to keep watching
It takes a minute and costs nothing. Your free account unlocks all of Chapter 1 and 3 more lessons of Ethics in Technology.
Lesson complete
Earn the 3 college credits for Ethics in Technology
Enrolling adds a quiz after each lesson, the graded work and the official transcript. Recommended for credit by NCCRS.
Next lesson: Why Collecting and Analyzing Massive Amounts of Data Raises Serious Privacy Questions →← How Algorithms Dig Through Raw Data to Discover Hidden Patterns and RelationshipsNext lesson: Why Collecting and Analyzing Massive Amounts of Data Raises Serious Privacy Questions →
In this lesson
Sentiment, relationships, trends, influence, and recommendations.
- Score sentiment and recognize context errors
- Compare degree with network bridges
- Calculate a trend score and engagement rates
- Explain recommendation approaches and limits
Free study notes
Get this lesson's notes by email
The key points and 4 practice questions with answers, in your inbox. One email, free.
One email. Unsubscribe in one click.
Transcript
Everything said in the video, section by section (1,836 words). Read along, or search it.
Chapter 5
Patterns on Social Platforms
Social platforms record posts, follows, shares, and clicks, providing many opportunities to study patterns. Those records can support customer service, research, marketing, and decisions about personalized feeds. They remain a partial view of behavior, rather than a complete account of human motives. This lesson introduces five techniques, each suited to a different question about that activity. Sentiment analysis estimates whether language expresses, a positive, negative, or neutral attitude. Social network analysis examines relationships, including the importance of bridges between groups. Trend detection looks for activity rising, unusually far above an appropriate baseline. Influencer analysis compares audience indicators, with engagement, relevance, authenticity, and network position. Recommendation systems use past behavior or item features, to suggest what may interest someone next. Worked examples let us calculate scores, while recognizing what the numbers cannot establish. Throughout the lesson, useful patterns require context, evaluation, and judgment about the resulting action.
A Shared Process
Collect, Clean, Mine, Then Choose a Responsible Action
The source organizes social-platform mining, into collection, cleaning, mining, and action. Collection gathers the relevant posts, interactions, relationships, or features needed for a defined question. The question helps determine which records matter, rather than treating every available field as essential. Cleaning addresses duplicates, spam, misspellings, and differences in the way records are formatted. Slang and emojis also need interpretation, because changing their representation can lose useful meaning. Automated accounts deserve examination, but automation alone does not establish that an account is malicious. Mining applies a suitable technique, and checks the quality and limitations of its results. Action can mean investigating complaints, adjusting a campaign, or changing a recommendation. Poor preparation can create misleading patterns, even when a calculation itself is performed correctly. Remember that a click or follow records activity, without revealing every reason behind it. Information boundaries and review matter, when a pattern becomes a decision affecting an audience.
Sentiment Analysis
A Word Score Can Miss the Meaning of a Post
Sentiment analysis classifies expressed attitude, usually as positive, negative, or neutral. A simple lexicon gives words scores, then adds those scores within a post. In the source example, love and great each receive, positive two points. Good receives positive one, while slow receives negative one and broken receives negative two. Other words receive zero, and the total determines the simple classification. Loving the camera and calling its photographs great produces, positive four points overall. A slow application with broken login produces, negative three points in the same scheme. An order arriving on Tuesday contains no scored word, so its total is zero. The sarcastic complaint about another crash includes great, giving a misleading positive score of two. The method follows its dictionary correctly, while failing to capture what the person actually means. A learned model can recognize additional patterns, but its performance still depends on training and evaluation.
Language and Context
Track Signals Without Treating Every Label as Certain
Language often requires more context, than the sentiment of an isolated word provides. Not bad can express approval, although bad alone would commonly suggest a negative attitude. Great screen and terrible battery express, different judgments about two parts of the same product. Slang and emojis can change meaning, across communities, situations, and surrounding language. Sarcasm can reverse the apparent sentiment, while leaving positive words visible in the text. Learned methods may handle some of these patterns, without becoming reliable in every situation. The source illustrates an app launch, followed by an increase in negative posts around an update. That fictional chart demonstrates monitoring, rather than providing measurements from a real product. A rising complaint share can be an early signal, prompting inspection of the posts and affected topics. Check which posts form the denominator, and whether collection or classification changed during the period. Treat the pattern as a reason to investigate, rather than proof of a cause or every individual's feeling.
Social Network Analysis
A Bridge Can Matter More Than Its Connection Count
Social network analysis represents people or accounts as nodes, with relationships shown as edges. The meaning of an edge must be defined, such as friendship, following, or replying. Following and replying can have direction, while this source example uses undirected connections. The left cluster has Ana, Ben, Cai, and Dee, with Kim connected only to Ana. The right cluster has Eli, Fay, Gus, and Hal, with several internal connections. Jo connects Dee on the left, with Eli on the right through two edges. Within this recorded network, removing Jo separates the two clusters from each other. Jo therefore matters as a bridge, despite having fewer connections than several other people. Ana and Dee each have four direct connections, illustrating a different kind of prominence. A network map can reveal structure, that an ordinary list of accounts would hide. Its boundaries still matter, because recorded shortest paths do not prove control over all real communication.
Different Measures
Degree and Betweenness Answer Different Questions
Degree here means the raw number of direct connections, rather than a normalized degree score. Betweenness examines how frequently a node participates, in shortest paths between other nodes. The source reports normalized values, for this undirected network using a particular calculation convention. Ana has four connections, with betweenness rounded to zero point two two. Dee also has four connections, with betweenness rounded to zero point five six. Jo has just two connections, while matching Dee's reported betweenness of zero point five six. Eli has three connections and betweenness of zero point five, serving the right cluster's gateway. Kim has one connection and zero betweenness, because no shortest path between others requires Kim. The comparison shows why many connections, and an important bridging position are different properties. Community detection finds relatively dense clusters, which may represent shared interests or coordinated activity. Density can motivate investigation of suspicious coordination, but does not by itself establish fake accounts.
Trend Detection
Compare the Spike With a Suitable Baseline
Trend detection looks for unusual activity, relative to a relevant topic's normal level. A constantly popular topic differs, from one whose activity has suddenly increased substantially. The source's example averages two hundred mentions per hour, with a standard deviation of forty. The current hour contains six hundred twenty mentions, exceeding the mean by four hundred twenty. Dividing that difference by forty yields, a z-score of ten point five. The example rule alerts above three, so this count is comfortably beyond that threshold. The score measures distance from the baseline, in units of its standard deviation. An alert does not prove genuine public interest, or establish why the count rose. Real analysis should compare relevant times and conditions, because activity patterns vary by audience. Check distinct accounts and automation, rather than assuming many posts mean many independent participants. Group equivalent hashtags and spellings carefully, then investigate the content behind the apparent spike.
Influencer Identification
Follower Count and Interaction Rate Tell Different Stories
Influencer identification asks whose voice fits, the purpose of a campaign or message. Follower count indicates a potential audience, without guaranteeing how many people view a post. The source compares three fictional accounts, using interactions per post divided by followers. Account A has two million followers, and twenty thousand interactions per post on average. That gives one percent, using the follower denominator specified for this example. Account B has fifty thousand followers and thirty-five hundred interactions, giving seven percent. Account C has eight thousand followers and nine hundred sixty interactions, giving twelve percent. The smaller account's higher rate can be useful, when its audience matches the message. It does not guarantee more purchases, a more trusting audience, or better overall return. Examine relevance, network position, and authentic participation, alongside the follower count and interaction rate. Actual outcomes should guide the choice, rather than treating one convenient metric as influence itself.
Recommendations
Similar Behavior and Similar Items Are Different Inputs
Recommendation systems estimate which item, connection, or piece of content may interest a person next. They support suggested videos, shopping recommendations, possible connections, and personalized feeds. Collaborative approaches use patterns in interactions, across users and items to generate suggestions. The source's simplified version compares users, who share recorded preferences with the target person. Content-based approaches instead examine item features, alongside that person's preferences and past actions. A suggested item can resemble content, that the person has previously liked or selected. Hybrid approaches combine signals, and may incorporate additional context into their ranking. The approaches differ in their inputs, although real systems can contain several stages and methods. A recommendation is a prediction about interest, rather than knowledge of someone's inevitable next choice. Past behavior provides evidence, while circumstances and preferences may still change. The worked example next shows a simple comparison, without claiming to represent every modern recommendation algorithm.
Collaborative Example
Two Similar Users Suggest Fitness Before Comedy
The source's table records your likes, for Cooking and Travel categories. Priya and Meera share both recorded likes, making them close matches in this simplified comparison. Arjun's recorded likes are Tech Reviews and Comedy, rather than either of your two categories. Both Priya and Meera also like Fitness, which has no known preference recorded for you. Only Priya has a recorded Comedy like, supplying a weaker additional signal. Fitness therefore comes first as a candidate, with Comedy a weaker second suggestion. This follows the example's simple matching rule, rather than a universal ranking formula. A category without a recorded like, should not automatically be interpreted as a dislike. It also does not establish, whether the person has seen the category or had an opportunity to respond. The suggestion remains a hypothesis, which can be evaluated through suitable feedback and outcomes. Larger systems use richer mathematical representations, but still need care when interpreting missing interaction evidence.
Limitations
No History and Narrow Exposure Need Deliberate Design
Cold start describes the difficulty, of recommending for users or items with little interaction history. Collaborative comparison has less evidence, when there are no established patterns to compare. That difficulty does not mean, every possible recommendation must fail until a long history exists. Other signals or additional preferences can help, depending on the system's design and available information. Repeatedly suggesting familiar content can narrow exposure, rather than encouraging variety or exploration. The risk is often called a filter bubble, but it is not an inevitable result of every recommendation. Hybrid methods can reduce particular weaknesses, by combining evidence with different strengths and limitations. They still need evaluation, because adding more signals does not automatically remove errors or harmful effects. What the platform displays influences, which future interactions it will be able to observe. That feedback loop deserves attention, when interpreting the history used to train or assess a system. Relevance, privacy, and audience effects remain ethical questions, beyond the accuracy of one ranking alone.
Practice and Review
Match the Technique to the Question
Choose the technique by the question, rather than applying a fashionable method to every dataset. Sentiment analysis examines expressed feeling, while network analysis studies relationships and structural positions. Trend detection examines unusual increases, relative to a relevant and defensible baseline. Influencer identification combines audience indicators, with suitability and evidence about participation and outcomes. Recommendation systems estimate interest, from interaction patterns, item features, or combined signals. For practice, suppose a topic averages five hundred mentions per hour, with standard deviation one hundred. The current count is seven hundred fifty, exceeding that mean by two hundred fifty. Dividing by one hundred gives, a z-score of two point five. With an alert threshold above three, this example would not yet trigger the rule. The numerical answer follows the stated baseline, rather than settling every reason for the observed activity. Across all five techniques, check the evidence and limits, before using a signal to justify an action.
About Ethics in Technology at UPI Study
Ethics in Technology (Computer Science 195) is a self-paced online course of 116 video lessons in 15 chapters, from cybersecurity basics, privacy and surveillance to big data, hacking, malware, network defense, intellectual property, cybercrime, encryption, emerging technology and digital citizenship. It carries a NCCRS credit recommendation for 3 semester credits.
Every lesson transcript is free on pages like this one, and the videos are free with a free UPI Study account. Enrolling adds a practice quiz after each lesson, the graded work and final exam that earn the credit, lifetime access, and an official transcript on completion. UPI Study has helped over 130,000 students with college credits. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
Enroll now → Course details Download the syllabus All Computer Science courses
Compare: Cybersecurity vs Ethics in Technology · Which ethics course?
Why take it here
Why students take Ethics in Technology with UPI Study
Where the credit goes
Universities that work with NCCRS
Ethics in Technology carries a NCCRS credit recommendation. These universities work with NCCRS; some are UPI Study's signed partners. ACE and NCCRS have over 2,100 cooperating colleges that will consider accepting credits; the final decision is with the universities.
Thomas Edison State UniversitySigned partner
University of the PeopleSigned partner
Charter Oak State CollegeSigned partner
Strayer UniversitySigned partner
Columbia Southern UniversitySigned partner
Indiana TechSigned partner
Penn StateWorks with NCCRS
Southern New Hampshire UniversityWorks with NCCRS
SUNYWorks with NCCRS
+many more cooperating universitiesFind my college →Find my college → Not listed? Ask Athena in the corner of this page, or book a 15-minute call.
FAQ
Questions students ask us
About Applying Data Mining Techniques to Understand Behavior and Trends on Social Platforms
How does lexicon-based sentiment analysis work?
What is betweenness centrality in social network analysis?
How do you use a z-score to detect a trending topic?
What is the difference between collaborative filtering and content-based recommendations?
About the Ethics in Technology course
How many college credits is Ethics in Technology?
How much does Ethics in Technology cost and what is included?
Which universities accept UPI Study credits?
How long does Ethics in Technology take?
How are the exams done?
How do my credits reach my college?
Ready for the full course? Earn 3 credits in Ethics in Technology.
116 video lessons, a quiz after each one, and the graded work that earns the credit. Recommended for credit by NCCRS.
3 credits · NCCRSEnroll →