Temporal pattern classification of internet meme propagation: A hybrid machine learning approach

Probably not a substantial claim: may not pass S3 (load-bearing).

The prior study [6] manually accrued a dataset of 100 elements using keywords from KnowYourMeme.com and Google Trends time-series.

“The prior study [6] manually accrued a dataset of 100 elements, using keywords from a variety of meme pages on KnowYourMeme.com.”

• cites [6] A. Lonnberg, P. Xiao, K. Wolfinger, The growth, spread, and mutation of internet phenomena: a study of memes, <i>Results Appl. Math.</i>, <b>6</b> (2020), 100092. https://doi.org/10.1016/j.rinam.2020.100092 sourcePartly supported

  Section 2.1 (Data collection) says the authors selected memes using KnowYourMeme.com and collected Google Trends data using the memes’ exact names. It states that 66 memes were selected, not 100; the number 100 refers to Google Trends’ normalized maximum popularity value. (LLM, conf=0.90)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim.

A cluster size of four was chosen based on intuition from [6] and silhouette analysis.

“A cluster size of four was chosen based on intuition from [6], as there are four popularity patterns expected to take shape, in addition to the results provided by a silhouette analysis.”

• cites [6] A. Lonnberg, P. Xiao, K. Wolfinger, The growth, spread, and mutation of internet phenomena: a study of memes, <i>Results Appl. Math.</i>, <b>6</b> (2020), 100092. https://doi.org/10.1016/j.rinam.2020.100092 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Silhouette analysis was performed to evaluate the number of clusters and interpret the resulting grouping structure.

“Silhouette analysis [23,24] was performed to evaluate the number of clusters and interpret the resulting grouping structure.”

• cites [23] P. J. Rousseeuw, Silhouettes: a graphical aid to the interpretation and validation of cluster analysis, <i>J. Comput. Appl. Math.</i>, <b>20</b> (1987), 53–65. https://doi.org/10.1016/0377-0427(87)90125-7 sourceSupported
• cites [24] L. Kaufman, P. J. Rousseeuw, <i>Finding Groups in Data: An Introduction to Cluster Analysis</i>, John Wiley &amp; Sons, New York, 1990. sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

A meme's date-of-origin, minus two years, was factored into each Pytrends request, allowing for a more accurate start date of the time-series data.

“Differing from the original paper [6], a meme's date-of-origin, minus two years for caution, was factored into each Pytrends request, allowing for a more accurate start date of the time-series data.”

• cites [6] A. Lonnberg, P. Xiao, K. Wolfinger, The growth, spread, and mutation of internet phenomena: a study of memes, <i>Results Appl. Math.</i>, <b>6</b> (2020), 100092. https://doi.org/10.1016/j.rinam.2020.100092 sourceSupported

Classified as probably not a substantial claim.

Previous work identified four distinct post-peak patterns: smooth decay, oscillatory decay, plateau, and sustained growth.

“Previous work identified four distinct post-peak patterns: smooth decay, oscillatory decay, plateau, and sustained growth.”

• Also stated in: § 1. Introduction · plain

  “Lonnberg et al. [6] distilled four canonical post-peak patterns that we adopt as a baseline taxonomy.”

• cites [6] A. Lonnberg, P. Xiao, K. Wolfinger, The growth, spread, and mutation of internet phenomena: a study of memes, <i>Results Appl. Math.</i>, <b>6</b> (2020), 100092. https://doi.org/10.1016/j.rinam.2020.100092 sourcePartly supported

  Section 2.1 (page not provided) classifies memes into four patterns corresponding to smooth decay, oscillatory or spikey decay, leveling off, and long-term growth. But it describes long-term growth as a continuous rise throughout the dataset, not a post-peak pattern. (LLM, conf=0.87)

  Experimental — accuracy not yet measured

May combine two claims.

The machine learning framework offers greater scalability for large datasets and flexibility for handling noisy time-series data compared to previous work.

“While we built upon the theoretical foundation established by [6], our machine learning framework offers greater scalability for large datasets and flexibility for handling noisy time-series data.”

• cites [6] A. Lonnberg, P. Xiao, K. Wolfinger, The growth, spread, and mutation of internet phenomena: a study of memes, <i>Results Appl. Math.</i>, <b>6</b> (2020), 100092. https://doi.org/10.1016/j.rinam.2020.100092 sourceSupported

The quote only partly states this claim.

The average silhouette coefficient was highest at k=4, with a score of 0.27, among cluster sizes from one to ten.

“With the highest average score, 0.27, given for a cluster size of four, the decision to keep four clusters as the optimal number gained credence.”

• Also stated in: § 3.3. Silhouette analysis · plain

  “We observe the highest output at k = 4 for all k ∈ [3...10].”

Probably not a substantial claim: may not pass S3 (load-bearing).

Cluster 1 representative examples include Approval Guy, 4-Panel Cringe, and Are You Sure About That?.

“Representative examples include Approval Guy, 4-Panel Cringe, and Are You Sure About That?.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Cluster 3 representative examples include 30-50 Feral Hogs, 2 Girls 1 Cup, and 6ix9ine Snitch.

“Representative examples include 30-50 Feral Hogs, 2 Girls 1 Cup, and 6ix9ine Snitch.”

May combine two claims.

Visual inspection reveals most curves within each cluster follow similar trajectories, with some outliers and borderline cases.

“Visual inspection of the clusters reveals that while most curves within each cluster follow similar trajectories, there are some outliers and borderline cases.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Cluster 0 representative examples include Elmo Rise, Ankha Zone, and Bitconnect Carlos.

“Representative examples include Elmo Rise, Ankha Zone, and Bitconnect Carlos.”

May combine two claims.

The chi-square test for independence between categories and clusters yielded a p-value of 0.023, indicating the category-cluster relationship is unlikely to be due to random chance.

“The chi-square test for independence between categories and clusters yielded a p-value of 0.023, indicating that the category-cluster relationship is unlikely to be due to random chance.”

• AssumptionRests on an assumption: KnowYourMeme.com categories are valid and sufficiently exhaustive for meme taxonomic analysis.

May combine two claims.

Exploitable memes show a relatively even distribution across clusters, with slight preferences for spikey decay (29.7%) and long-term growth (29.2%) patterns.

“Exploitable memes show a relatively even distribution across clusters, with slight preferences for spikey decay (29.7%) and long-term growth (29.2%)patterns”

• AssumptionRests on an assumption: KnowYourMeme.com categories are valid and sufficiently exhaustive for meme taxonomic analysis.

May combine two claims.

Relatively low intra-cluster variance within clusters 0, 2, and 3 suggests these patterns are more consistently defined, while higher variance in cluster 1 confirms its noisier composition.

“The relatively low intra-cluster variance within clusters 0, 2, and 3 suggests that these patterns are more consistently defined, while the higher variance in cluster 1 confirms our observation about its relatively noisier composition.”

Probably not a substantial claim: may not pass S3 (load-bearing).

k-means clustering on a few hundred samples yielded four clusters.

“The four clusters resulting from k-means clustering on a few hundred samples.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Performing the metric analysis for DTW yielded negligible differences between scores.

“Performing this metric analysis for DTW yielded negligible differences between scores.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Label 0 has 76 samples.

“some labels such as label 0 with 76 samples have relatively few samples compared to label 3, which has 225.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The dummy classifier achieved an accuracy of 0.35 under the most frequent strategy.

“| | Accuracy | | 0.35 | | 638 |”

May combine two claims.

Ankha's popularity curve exhibits a sharp surge to peak popularity followed by a steep decline, characteristic of Cluster 0 (short-lived burst).

“Ankha exhibits a sharp surge to peak popularity followed by a steep decline, characteristic of Cluster 0 (short-lived burst).”

Probably not a substantial claim: may not pass S3 (load-bearing).

The model achieved a cross-entropy loss of 0.0685.

“Cross-Entropy Loss | 0.0685”

Probably not a substantial claim: may not pass S3 (load-bearing).

The popularity data for all memes was not able to be successfully requested due to networking errors or otherwise.

“As previously mentioned, the popularity data for all memes was not able to be successfully requested whether due to networking errors or otherwise.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The dummy classifier achieved an accuracy of 0.28 under the stratified strategy.

“| | Accuracy | | 0.28 | | 638 |”

May combine two claims.

The average of all accuracy scores is approximately .976, or .98, which matches the performance found in the classification report listed by Table 1.

“The average of all accuracy scores is approximately .976, or .98,which matches the performance found in the classification report listed by Table 1.”

May combine two claims.

SVC achieved the highest accuracy among the alternative classification models, followed closely by XGBoost and random forest.

“All models substantially outperformed the dummy classifiers, with SVC achieving the highest accuracy, followed closely by XGBoost and random forest.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The dummy classifier achieved an accuracy of 0.26 under the uniform strategy.

“| | Accuracy | | 0.26 | | 638 |”

Probably not a substantial claim: may not pass S3 (load-bearing).

Label 3 takes up about 35% of the samples.

“With label 3 taking up about 35% of the samples, the dummy classifier sees a jump in performance to that accuracy.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Label 3 has 225 samples.

“some labels such as label 0 with 76 samples have relatively few samples compared to label 3, which has 225.”

May combine two claims.

Some memes were described by keywords frequently used to refer to other popular, homonymous terms, and therefore the obtained Google Trends data is not accurate in reflecting the month-by-month popularity of the expected meme.

“Some memes were also described by keywords frequently used to refer to other popular, homonymous terms and therefore the obtained Google Trends data is not accurate in reflecting the month-by-month popularity of the expected meme.”

Probably not a substantial claim: may not pass S3 (load-bearing).

MinMax scaling helps reduce potential bias by scaling outliers into the same fixed range between 0 and 100.

“The feature normalization provided by MinMax scaling helps to reduce any potential bias, scaling outliers into the same fixed range, i.e., between 0 and 100 in our project.”

• cites [22] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, et al., Scikit-learn: machine learning in Python, <i>J. Mach. Learn. Res.</i>, <b>12</b> (2011), 2825–2830. https://doi.org/10.5555/1953048.2078195 sourceNot checked — no full text of the source

May combine two claims.

k-means was selected for its computational efficiency, interpretability, and effectiveness with normalized time-series data.

“After comparative analysis, k-means was selected for its computational efficiency, interpretability, and effectiveness with our normalized time-series data.”

May combine two claims.

XGBoost showed comparable accuracy in preliminary tests, but SVC provided more stable decision boundaries and better generalization on validation sets.

“While XGBoost showed comparable accuracy in our preliminary tests, SVC provided more stable decision boundaries and better generalization on our validation sets.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Cluster 2 patterns typically emerge from memes that achieve cultural permanence in specific communities.

“Memes that achieve cultural permanence in specific communities”

Probably not a substantial claim: may not pass S3 (load-bearing).

Dimensionality reduction such as PCA was unnecessary due to the feature size of 2.

“Dimensionality reduction such as principal component analysis (PCA) was unnecessary due to the feature size of 2.”

May combine two claims.

The date-of-origin adjustment helped limit noise and the popularity of homonymous terms that could skew results.

“This helped to limit noise and the popularity of homonymous terms-that could drastically skew results-early 2000's data in Google Trends typically returns, before improvements were made in its data collection system.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). May combine two claims.

This sustained utility explains their persistence in online communities and their role as foundational elements of internet culture.

“This sustained utility explains their persistence in online communities and their role as foundational elements of internet culture.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The tendency of exploitable image macros toward leveling-off patterns (Cluster 2), as illustrated by 60s Spider-Man, aligns with their inherent adaptability.

“The tendency of exploitable image macros toward leveling-off patterns (Cluster 2), as illustrated by 60s Spider-Man, aligns with their inherent adaptability.”

• AssumptionRests on an assumption: KnowYourMeme.com categories are valid and sufficiently exhaustive for meme taxonomic analysis.

The quote only partly states this claim.

Cluster 2 patterns typically emerge from established meme templates with consistent baseline usage.

“Established meme templates with consistent baseline usage”

Classified as probably not a substantial claim. May combine two claims. The quote only partly states this claim.

Stratified k-fold cross-validation with k=5 was used, repeated 20 times with different random splits, and performance metrics were averaged.

“For the SVC model, we used stratified k-fold cross-validation with k = 5, which maintains the same class proportion across all folds.”

• Also stated in: § 2.5. Cross-validation and model evaluation · plain

  “The process was repeated 20 times (n = 20) with different random splits.”

May combine two claims.

The analysis window of 43 months is motivated because popularity patterns are defined primarily by the immediate rise to the peak and post-peak behavior, and all curves must be the same length for k-means Euclidean clustering.

“The final stage of the preprocessing pipeline is the use of an analysis window, with a period of 43 months, on each popularity curve.We are motivated to use such a method because popularity patterns are defined primarily by the immediate rise to the peak and the resulting post-peak behavior in the curve.”

May combine two claims.

These dynamics underscore the role of shock value, cultural references, and community-driven resurgences in prolonging meme relevance.

“These dynamics underscore the role of shock value, cultural references, and community-driven resurgences in prolonging meme relevance.”

May combine two claims.

SVC was chosen due to its strong performance on moderate-sized datasets with complex decision boundaries, high generalization capacity with regularization, and robustness to overfitting when properly tuned.

“SVC was ultimately chosen due to its strong performance on moderate-sized datasets with complex decision boundaries, high generalization capacity with regularization, and robustness to overfitting when properly tuned.”

Probably not a substantial claim: may not pass S3 (load-bearing).

This rigorous validation approach provides a more reliable estimate of model performance than a single train-test split, particularly for detecting potential overfitting.

“This rigorous validation approach provides a more reliable estimate of model performance than a single train-test split, particularly for detecting potential overfitting.”

May combine two claims.

This pattern likely reflects the ease of replication and modification of textual or event-driven memes, which accelerates saturation and contributes to audience fatigue.

“This pattern likely reflects the ease of replication and modification of textual or event-driven memes, which accelerates saturation and contributes to audience fatigue.”

Classified as probably not a substantial claim. May combine two claims.

The CHI likely regarded the three-cluster configuration as providing the most distinct and compact grouping at a global scale, while treating the fourth cluster as a refinement of existing groups that did not significantly enhance overall dispersion.

“In this case, the CHI likely regarded the three-cluster configuration as providing the most distinct and compact grouping at a global scale, while treating the fourth cluster as a refinement of existing groups that did not significantly enhance overall dispersion.”

• AssumptionRests on an assumption: The CHI is a global variance-based metric that evaluates the ratio of between-cluster dispersion to within-cluster dispersion.
• AssumptionRests on an assumption: The CHI tends to favor simpler models unless additional clusters lead to a substantial gain in inter-cluster separation.

May combine two claims.

DTW-based methods can handle temporal shifts more effectively, but fixed-length time windows mitigated this advantage, and k-means provided clearer cluster boundaries for the dataset.

“While DTW-based methods can handle temporal shifts more effectively, our preprocessing approach with fixed-length time windows mitigated this advantage, and k-means provided clearer cluster boundaries for our dataset.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

This endurance may stem from the higher barrier to creation, which limits oversaturation, and the richer sensory experience that sustains audience engagement over extended periods.

“This endurance may stem from the higher barrier to creation, which limits oversaturation, as well as the richer sensory experience that sustains audience engagement over extended periods.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Trying different splits on each iteration further reduces the possibility that the model performance is stemming from excessive overfitting.

“By trying different splits on each iteration, we can further reduce the possibility that the model performance is stemming from excessive overfitting.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

The methodology developed here could be adapted to study other forms of viral content beyond traditional internet memes.

“The methodology developed here could also be adapted to study other forms of viral content beyond traditional internet memes.”

Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

The developed classification model could assist in predicting meme trajectory patterns, with potential applications in digital marketing, content strategy, and social media analysis.

“Practically, the developed classification model could assist in predicting meme trajectory patterns, with potential applications in digital marketing, content strategy, and social media analysis.”

May combine two claims.

Higher production investment and multimedia richness contribute to content longevity, possibly due to higher replay value and emotional resonance.

“The prevalence of viral videos in this cluster suggests that higher production investment and multimedia richness contribute to content longevity, possibly due to higher replay value and emotional resonance.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Stratification ensures that the cross-validation maintains the percentage of samples for each class.

“Stratification ensures that the CV maintains the percentage of samples for each class.”

May combine two claims.

Although the CHI points to a globally optimal structure at k=3, the silhouette and DBI scores suggest that the finer structure at k=4 provides better local separation.

“Although the CHI points to a globally optimal structure at k = 3, the silhouette and DBI scores suggest that the finer structure at k = 4 provides better local separation.”

Probably not a substantial claim: may not pass S2 (checkable).

The findings advance understanding of how different types of digital content propagate through online networks.

“Theoretically, they advance our understanding of how different types of digital content propagate through online networks.”

Probably not a substantial claim: may not pass S4 (based).

The spikey decay pattern (Cluster 3) indicates content that generates intensive but intermittent engagement, possibly reflecting complex or niche themes that sustain prolonged community discussion and periodic revival.

“Finally, the spikey decay pattern (Cluster 3) indicates content that generates intensive but intermittent engagement, possibly reflecting complex or niche themes that sustain prolonged community discussion and periodic revival.”

Classified as probably not a substantial claim.

Additional classification experiments using alternative algorithms including random forest, XGBoost, and k-means with the DTW metric were conducted to address potential concerns about the validity of the approach.

“To address potential concerns about the validity of our approach, we conducted additional classification experiments using alternative algorithms including random forest, XGBoost, and k-means with the DTW metric.”

Classified as probably not a substantial claim. May combine two claims.

Stratified k-fold cross-validation was used to eliminate the possibility of nonnegligible overfitting, with the accuracies of each fold taken.

“To eliminate the possibility of nonnegligible overfitting, the stratified k-fold cross-validation (CV) technique was also used, with the accuracies of each fold being taken.”

May combine two claims.

Cross-validation was performed on the model to alleviate concerns of overfitting and confirm results.

“To alleviate concerns of overfitting, we perform cross-validation on the model to help confirm our results thus far.”

May combine two claims.

Specificity and temporal relevance drive initial virality but may limit long-term cultural impact.

“Our analysis suggests this pattern is most common among catchphrase memes and event-driven content, indicating that specificity and temporal relevance, while driving initial virality, may limit long-term cultural impact.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The stratified dummy classifier's accuracy of .28 is only marginally better than the uniform strategy.

“This strategy sees an accuracy score of .28,or only marginally better than the uniform strategy.”

Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

Each cluster represents fundamentally different approaches to cultural transmission in digital spaces.

“Each cluster represents fundamentally different approaches to cultural transmission in digital spaces, with implications extending beyond academic classification to practical applications in content strategy and digital marketing.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The most frequent dummy classifier's accuracy of .35 is due to label 3 comprising about 35% of samples.

“With label 3 taking up about 35% of the samples, the dummy classifier sees a jump in performance to that accuracy.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

There may be hybrid or transitional patterns not fully captured by the four-pattern framework.

“Additionally, while our clustering approach revealed four distinct patterns, there may be hybrid or transitional patterns not fully captured by this framework.”

Probably not a substantial claim: may not pass S3 (load-bearing).

There will always be memes that do not appear to follow any of the four popularity patterns.

“Four clusters may be objectively best for classification, but there will always be memes that simply do not appear to follow any of the four popularity patterns.”

Probably not a substantial claim: may not pass S3 (load-bearing).

RepeatedStratifiedKFold was used to further confirm training performance.

“To ensure that the training performance is further confirmed, we use RepeatedStratifiedKFold which repeats the stratified k-fold CV n times.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

Reducing data sparsity has the potential to better the output of any machine learning algorithms that could be applied going forward.

“In addition, it has shown that reducing data sparsity has the potential to better the output of any machine learning algorithms that could be applied going forward.”

Classified as probably not a substantial claim.

The high performance is expected to some degree because SVC is particularly effective at learning the decision boundaries that separate clusters created by k-means.

“This high performance is expected to some degree, as SVC is particularly effective at learning the decision boundaries that separate clusters created by k-means.”