Two-paragraph reflection

Draft: personalize these paragraphs after running the analysis and completing the review.

This newly designed lab gives me a useful way to connect data preparation, text analysis, sentiment analysis, and visualization in one reproducible report. Working with AI also shows why I need to describe my files and constraints clearly. My workbook contains a 100-row first scrape, a 15-row second sample, a third file with the same 100 post IDs as the first scrape, and supplied sentiment samples of 20 and 15 posts. Treating these as independent collections or counting the sentiment samples as additional posts would produce misleading results. The code below documents missing text, checks duplicates, and separates the source records from the analysis. I will use its checks to verify the results before interpreting them. A worked example of timestamp validation and a clearer distinction between a sample and total conversation volume would make this lab easier to follow.

Human judgment remains important when interpreting multilingual posts, product promotions, sarcasm, and comments about athletes rather than the brand itself. An English sentiment dictionary can label individual words without understanding the intended meaning or target of a post. For that reason, this analysis restricts dictionary scoring to posts marked as English, preserves unmatched posts as a separate category, and provides the existing manual labels for comparison. I will inspect the original wording and matched words before accepting any automated interpretation. The unequal sample sizes, missing language information, and relative timestamps also limit comparisons over time. AI can help organize the workflow and write code, but I remain responsible for checking assumptions, explaining disagreements, and making recommendations supported by the evidence.

1. Purpose and source boundaries

This version includes a snapshot of the supplied workbook data inside the file, so knitting does not depend on the workbook filename or its folder. To refresh from a later workbook, set params: workbook to its actual path; otherwise leave it blank. Embedded data are visible in the source and the resulting HTML includes tables of public post text, so review the report before publishing.

This individual report uses the uploaded workbook to examine recurring words, engagement, and sentiment. The three supplied batches remain separate versions. Scrape 3 is not assumed to be a new collection session. The other workbook tabs contain copies or summaries of these records and are not additional independent observations.

  • Scrape 1: 100 source records. The All Posts sheet supplies the readable text and metrics; Scrape 1 Raw supplies identifiers, language, and timestamp metadata.
  • Scrape 2: 15 manually compiled records. The supplied notes identify the collection date as October 26, 2026, and explain that visible zero counts were transcribed from listings rather than verified with an API.
  • Scrape 3: 100 source records from the additional workbook. All IDs also occur in Scrape 1; a separate collection time is not documented. Supplied metrics differ for some IDs.
  • Manual sentiment: 20 existing labels and reasons for Scrape 1, plus a separate 15-row supplied sample in Scrape 3. They are supplied labels, not a new review conducted for this report. Random sampling is not documented.

The script does not claim that 100 versus 15 records means conversation volume fell. Search coverage, sampling methods, and time windows differ.

Supplied collection notes
Field Value
Collection source BlueFacts live Bluesky hashtag pages: #Nike and #sneakers
Source URL 1 https://bluefacts.app/tag/nike
Source URL 2 https://bluefacts.app/tag/sneakers
Collection date October 26, 2026
Important limitation This is a smaller manually compiled public sample, not a full API export. Relative timestamps are preserved as displayed. Rows with counts shown as 0 reflect no visible engagement in the listing, not independently verified API counts.
batch source_records
Scrape 1 100
Scrape 2 15
Scrape 3 100

Check overlap before interpreting the third batch

ids1 <- unique(na.omit(batch1$post_id))
ids3 <- unique(na.omit(batch3$post_id))
overlap_check <- tibble(
  measure = c("Scrape 3 distinct supplied IDs", "IDs also in Scrape 1", "IDs absent from Scrape 1"),
  value = c(length(ids3), length(intersect(ids1, ids3)), length(setdiff(ids3, ids1))))
kable(overlap_check)
measure value
Scrape 3 distinct supplied IDs 100
IDs also in Scrape 1 100
IDs absent from Scrape 1 0
metric_versions <- batch1 %>% select(post_id, likes_1 = likes, reposts_1 = reposts) %>%
  inner_join(batch3 %>% select(post_id, likes_3 = likes, reposts_3 = reposts), by = "post_id") %>%
  mutate(comparable = !is.na(likes_1) & !is.na(reposts_1) & !is.na(likes_3) & !is.na(reposts_3),
         metrics_differ = comparable & (likes_1 != likes_3 | reposts_1 != reposts_3))
kable(tibble(shared_ids = nrow(metric_versions),
             comparable_metrics = sum(metric_versions$comparable),
             different_metric_values = sum(metric_versions$metrics_differ)))
shared_ids comparable_metrics different_metric_values
100 100 42

Scrape 3 is retained as requested so its supplied version can be inspected. Its shared post IDs are not new independent observations. Different metrics or restored text do not establish when or how a file was collected. Do not infer growth or a new collection session from the filename alone.

2. Clean records without inventing missing values

Blank text and spreadsheet-error strings cannot support text analysis. Exclude them from the analytic dataset and export them for inspection. Missing metrics remain missing rather than becoming zero. Duplicate removal is confined to each batch: use a post URI when available, otherwise normalized handle plus text. This fallback cannot prove that two similar posts are identical.

flagged <- combined_raw %>%
  mutate(
    text_clean = normalize_text(text_original),
    text_invalid = text_clean == "" |
      text_clean %in% c("#NAME?", "#REF!", "#VALUE!", "#N/A", "NA"),
    content_key = paste(handle, str_to_lower(text_clean), sep = " :: "),
    dedup_key = if_else(!is.na(post_id) & post_id != "", post_id, content_key),
    encoding_flag = str_detect(text_clean, "\u00c3|\u00c2|\u00e2\u20ac|\ufffd")
  )
valid_before_dedup <- flagged %>% filter(!text_invalid)
clean <- valid_before_dedup %>% distinct(batch, dedup_key, .keep_all = TRUE) %>%
  mutate(engagement = if_else(is.na(likes) | is.na(reposts), NA_real_, likes + reposts))
audit <- flagged %>% group_by(batch) %>%
  summarise(source_records = n(), unusable_text = sum(text_invalid), .groups = "drop") %>%
  left_join(clean %>% count(batch, name = "retained"), by = "batch") %>%
  mutate(retained = replace_na(retained, 0L),
         duplicates_removed = source_records - unusable_text - retained)
kable(audit, caption = "Source-to-analysis reconciliation")
Source-to-analysis reconciliation
batch source_records unusable_text retained duplicates_removed
Scrape 1 100 4 96 0
Scrape 2 15 0 15 0
Scrape 3 100 3 97 0
write_csv(flagged %>% filter(text_invalid), file.path(output_dir, "excluded_text_records.csv"))
write_csv(clean, file.path(output_dir, "clean_posts.csv"))
kable(clean %>% group_by(batch) %>% summarise(
  missing_likes = sum(is.na(likes)), missing_reposts = sum(is.na(reposts)),
  text_encoding_flags = sum(encoding_flag), .groups = "drop"))
batch missing_likes missing_reposts text_encoding_flags
Scrape 1 0 0 0
Scrape 2 0 0 0
Scrape 3 0 0 0

Encoding flags are review prompts, not automatic proof of corruption. Original text is retained, and the script does not guess replacements for damaged text.

3. Compare sample coverage and dates

sample_summary <- clean %>% group_by(batch) %>% summarise(
  retained_posts = n(),
  distinct_nonblank_handles = n_distinct(handle[handle != ""]),
  valid_post_dates = sum(!is.na(posted_at)),
  complete_engagement_rows = sum(!is.na(engagement)),
  known_engagement_total = if (all(is.na(engagement))) NA_real_ else sum(engagement, na.rm = TRUE),
  mean_known_engagement = if (all(is.na(engagement))) NA_real_ else mean(engagement, na.rm = TRUE),
  .groups = "drop")
kable(sample_summary, digits = 2)
batch retained_posts distinct_nonblank_handles valid_post_dates complete_engagement_rows known_engagement_total mean_known_engagement
Scrape 1 96 54 96 96 79 0.82
Scrape 2 15 8 0 15 46 3.07
Scrape 3 97 55 97 97 79 0.81
date_coverage <- clean %>% group_by(batch) %>% summarise(
  earliest_utc = if (all(is.na(posted_at))) NA_character_ else
    format(min(posted_at, na.rm = TRUE), tz = "UTC", usetz = TRUE),
  latest_utc = if (all(is.na(posted_at))) NA_character_ else
    format(max(posted_at, na.rm = TRUE), tz = "UTC", usetz = TRUE),
  .groups = "drop")
kable(date_coverage)
batch earliest_utc latest_utc
Scrape 1 2026-10-22 18:15:20 UTC 2026-10-22 22:30:14 UTC
Scrape 2 NA NA
Scrape 3 2026-10-22 18:15:20 UTC 2026-10-22 22:30:14 UTC

Totals describe the retained sample; missing metrics make them partial. Relative times in Scrape 2 remain as supplied. The collection date is not substituted for the date of every post. A comparison of weekly trends requires multiple comparable time windows and reliable timestamps, which this workbook does not establish.

plotly::plot_ly(audit, x = ~batch, y = ~retained, type = "bar",
        marker = list(color = "#245B89"),
        hovertemplate = "%{x}<br>Retained sample records: %{y}<extra></extra>") %>%
  plotly::layout(title = "Retained records by sample (not total platform volume)",
         xaxis = list(title = "Sample"), yaxis = list(title = "Retained records"))
weekly_observed <- clean %>% filter(!is.na(posted_at)) %>%
  mutate(week_start_utc = as.Date(floor_date(posted_at, "week", week_start = 1))) %>%
  count(batch, week_start_utc, name = "dated_records")
kable(weekly_observed, caption = "Observed dated records by Monday-start week; missing dates excluded")
Observed dated records by Monday-start week; missing dates excluded
batch week_start_utc dated_records
Scrape 1 2026-10-19 96
Scrape 3 2026-10-19 97

4. Engagement and account concentration

top_posts <- clean %>% filter(!is.na(engagement)) %>%
  group_by(batch) %>% arrange(desc(engagement), record_id, .by_group = TRUE) %>%
  slice_head(n = 5) %>% ungroup() %>%
  select(batch, record_id, handle, text_clean, likes, reposts, engagement)
kable(top_posts, caption = "Top five retained records per sample; ties ordered by record ID")
Top five retained records per sample; ties ordered by record ID
batch record_id handle text_clean likes reposts engagement
Scrape 1 Scrape1-row2 realmadrid-fans.bsky.social ⚜ One to watch: Spotless: Barcelona unveils special WHITE boots for El Clasico Agree or disagree? #RealMadrid #HalaMadrid #ElClasico 25 1 26
Scrape 1 Scrape1-row3 fcbarcelona-news.bsky.social ⚜ One to watch: Spotless: Barcelona unveils special WHITE boots for El Clasico Agree or disagree? #FCBarcelona #Barca #ElClasico 4 1 5
Scrape 1 Scrape1-row4 swooshsupply.bsky.social Nike SB’s skate-ready Air Force 1 returns in a clean color combination of Cool Grey and Court Green, releasing next Tuesday, October 27. 4 1 5
Scrape 1 Scrape1-row5 hnkbae.bsky.social Ah, sim, meus queridos Nike, Adidas e
 4 0 4
Scrape 1 Scrape1-row6 goods.wouldcop.com Nike Tech Fleece (M), L — $11 all-in. The lowest of 1224 current size-L listings in the same condition on the same site (oldest listed ~28 months ago). (sized from a single reference garment) 4 0 4
Scrape 2 Scrape2-row7 butterflygirl24.bsky.social Nike’s apology for the Converse Ad. There wasn’t any information of disciplinary action towards those responsible for the ad. The apology is not an apology if it’s empty words w no acknowledgment of the harm of racism. So how will Nike n Converse Do Better? #nike 30 7 37
Scrape 2 Scrape2-row8 solefeen.bsky.social Jalen Brunson x Nike Kobe 3 Low Protro “Arancia” Availability: December 11th, 2026. Retail Price: $190 USD. #Nike #Kobe3 #JalenBrunson #Sneakers 5 3 8
Scrape 2 Scrape2-row2 jbjaso3.bsky.social Nike Crater Impact ‘Black Off Noir’ #WDYWT #KOTD #Sneakers #Sneakerhead #Nike 1 0 1
Scrape 2 Scrape2-row10 vibecircuit.bsky.social .SWOOSH x Nike Air Max 1 ‘Low Poly’: Future tech meets retro aesthetic, and it actually works. #nike #airmax1 #sneakers #sneakerhead #kicks 0 0 0
Scrape 2 Scrape2-row11 toppnews.bsky.social News post linking Nike with Paris and London and an external Financial Times article. #Nike 0 0 0
Scrape 3 Scrape3-row71 roryblank.bsky.social I mean I don’t know that Travis Knight, Nike scion, is hurting for money, but Laika movies are extremely time and labor intensive endeavors and they spent \(100 million dollars making a stop motion movie, and by they, I mean also he's the producer and head of the company that produced it. | 25| 1| 26| |Scrape 3 |Scrape3-row58 |ricardooliveria2.bsky.social |đŸ”„ TĂȘnis Nike Mc Trainer 3 🚹 De R\) 599 por R$ 289 đŸŽŸïž Use o Cupom: MODA2310 🔗 https://meli.la/2qL9VCU #amazon #Afazenda #Jonatas #desconto #promo 4 0 4
Scrape 3 Scrape3-row60 promoatomicas.bsky.social đŸ”„ TĂȘnis Nike Mc Trainer 3 🚹 De R$ 599 por R$ 289 đŸŽŸïž Use o Cupom: MODA2310 🔗 https://meli.la/2qL9VCU #mercadolivre #amazon #Afazenda #oferta #blackfriday 4 0 4
Scrape 3 Scrape3-row84 ricardooliveria2.bsky.social đŸ”„ TĂȘnis Nike Uplift Sc Masculino 🚹 R$ 266 đŸŽŸïž Cupom: MODA2310 🔗 https://meli.la/1PZ54Me #Revelando #amazon #SDV #desconto #oferta 4 0 4
Scrape 3 Scrape3-row86 promoatomicas.bsky.social đŸ”„ TĂȘnis Nike Uplift Sc Masculino 🚹 R$ 266 đŸŽŸïž Cupom: MODA2310 🔗 https://meli.la/1PZ54Me #Afazenda #SDV #Entregou #promoção #oferta 4 0 4
account_counts <- clean %>% filter(handle != "") %>% count(batch, handle, name = "posts") %>%
  group_by(batch) %>% mutate(share_of_named_posts = posts / sum(posts)) %>%
  arrange(desc(posts), handle, .by_group = TRUE) %>% slice_head(n = 5) %>% ungroup()
kable(account_counts, digits = 3, caption = "Most frequent accounts; denominator excludes blank handles")
Most frequent accounts; denominator excludes blank handles
batch handle posts share_of_named_posts
Scrape 1 kickterminal.bsky.social 15 0.156
Scrape 1 garfieldelric.bsky.social 8 0.083
Scrape 1 sneakerscouts.bsky.social 8 0.083
Scrape 1 promobit-promo.bsky.social 5 0.052
Scrape 1 ofertandopromocoes.bsky.social 4 0.042
Scrape 2 solefeen.bsky.social 5 0.333
Scrape 2 vibecircuit.bsky.social 3 0.200
Scrape 2 dealzbro.bsky.social 2 0.133
Scrape 2 backthefudgeoff.bsky.social 1 0.067
Scrape 2 butterflygirl24.bsky.social 1 0.067
Scrape 3 kickterminal.bsky.social 15 0.155
Scrape 3 garfieldelric.bsky.social 8 0.082
Scrape 3 sneakerscouts.bsky.social 8 0.082
Scrape 3 promobit-promo.bsky.social 5 0.052
Scrape 3 ofertandopromocoes.bsky.social 4 0.041

Compare original wording before describing what top posts have in common. Engagement counts reflect different collection methods and observation times. They do not measure sales or establish the reasons a post attracted attention.

5. Text analysis

All readable posts are tokenized. Frequency results are separated by batch and language. English stop words are removed only for posts explicitly marked as English. Non-English or unknown-language frequencies need language-specific review. URLs are removed for tokenization; original text remains available.

token_input <- clean %>%
  mutate(text_for_tokens = str_remove_all(text_clean, "https?://\\S+|www\\.\\S+"),
         english_eligible = str_detect(language, "^en(?:$|[-;,_ ])"))
tokens <- token_input %>% select(record_id, batch, language, english_eligible, text_for_tokens) %>%
  unnest_tokens(word, text_for_tokens)
stopword_lookup <- tidytext::stop_words %>% distinct(word) %>% mutate(is_stopword = TRUE)
content_tokens <- tokens %>% left_join(stopword_lookup, by = "word") %>%
  filter(str_detect(word, "\\p{L}"),
         !(english_eligible & replace_na(is_stopword, FALSE)))
word_counts <- content_tokens %>% count(batch, language, word, sort = TRUE)
kable(word_counts %>% group_by(batch) %>% slice_max(n, n = 10, with_ties = FALSE) %>% ungroup(),
      caption = "Common retained words; language shown to avoid mixing vocabularies")
Common retained words; language shown to avoid mixing vocabularies
batch language word n
Scrape 1 en nike 39
Scrape 1 fr nike 35
Scrape 1 unknown nike 24
Scrape 1 fr chez 15
Scrape 1 unknown the 15
Scrape 1 en tĂȘnis 13
Scrape 1 fr ad 13
Scrape 1 fr sneakers 12
Scrape 1 en cupom 11
Scrape 1 fr le 11
Scrape 2 unknown nike 27
Scrape 2 unknown sneakers 10
Scrape 2 unknown air 9
Scrape 2 unknown the 8
Scrape 2 unknown force 6
Scrape 2 unknown low 6
Scrape 2 unknown airforce1 5
Scrape 2 unknown availability 5
Scrape 2 unknown price 5
Scrape 2 unknown retail 5
Scrape 3 en nike 39
Scrape 3 fr nike 35
Scrape 3 unknown nike 24
Scrape 3 fr chez 15
Scrape 3 unknown the 15
Scrape 3 en tĂȘnis 13
Scrape 3 fr ad 13
Scrape 3 fr sneakers 12
Scrape 3 en cupom 11
Scrape 3 fr le 11
write_csv(word_counts, file.path(output_dir, "word_counts.csv"))
english_words <- content_tokens %>% filter(english_eligible, batch == "Scrape 1") %>% count(word, sort = TRUE) %>%
  slice_head(n = 15) %>% arrange(n, word)
if (nrow(english_words) > 0) {
  english_words$word <- factor(english_words$word, levels = english_words$word)
  plotly::plot_ly(english_words, x = ~n, y = ~word, type = "bar", orientation = "h",
          marker = list(color = "#32867C"),
          hovertemplate = "%{y}<br>Occurrences: %{x}<extra></extra>") %>%
    plotly::layout(title = "Frequent English words in Scrape 1 (shared posts not pooled)",
           xaxis = list(title = "Occurrences"), yaxis = list(title = ""),
           margin = list(l = 120))
} else cat("No posts are explicitly marked as English; no English word chart was produced.")

6. Sentiment analysis with an explicit language boundary

The Bing lexicon labels English words as positive or negative. The score is the number of positive word occurrences minus negative word occurrences. The scoring tokens are taken before stop-word removal, so frequency filtering cannot silently change sentiment inputs.

  • Positive/Negative: net score above/below zero.
  • Balanced matches: positive and negative counts tie, with at least one match.
  • No matches: no dictionary words matched; this does not establish neutrality.
  • Not scored: the supplied metadata do not identify the post as English.
bing <- tidytext::get_sentiments("bing")
matched_words <- tokens %>% filter(english_eligible) %>% inner_join(bing, by = "word")
per_post_matches <- matched_words %>% group_by(record_id) %>% summarise(
  positive_words = sum(sentiment == "positive"),
  negative_words = sum(sentiment == "negative"), .groups = "drop")
scored <- token_input %>% left_join(per_post_matches, by = "record_id") %>%
  mutate(across(c(positive_words, negative_words), ~replace_na(.x, 0L)),
         positive_words = if_else(english_eligible, positive_words, NA_integer_),
         negative_words = if_else(english_eligible, negative_words, NA_integer_),
         matched_count = positive_words + negative_words,
         score = positive_words - negative_words,
         automated_label = case_when(
           !english_eligible ~ "Not scored: language",
           matched_count == 0 ~ "No matches",
           score > 0 ~ "Positive", score < 0 ~ "Negative", TRUE ~ "Balanced matches"))
language_coverage <- scored %>% count(batch, language, english_eligible, name = "posts")
kable(language_coverage, caption = "Language coverage and scoring eligibility")
Language coverage and scoring eligibility
batch language english_eligible posts
Scrape 1 de FALSE 1
Scrape 1 en TRUE 45
Scrape 1 fr FALSE 15
Scrape 1 nl FALSE 1
Scrape 1 pt FALSE 6
Scrape 1 pt-br FALSE 4
Scrape 1 unknown FALSE 24
Scrape 2 unknown FALSE 15
Scrape 3 de FALSE 2
Scrape 3 en TRUE 45
Scrape 3 fr FALSE 15
Scrape 3 nl FALSE 1
Scrape 3 pt FALSE 6
Scrape 3 pt-br FALSE 4
Scrape 3 unknown FALSE 24
coverage <- scored %>% group_by(batch) %>% summarise(
  retained = n(), eligible = sum(english_eligible),
  with_dictionary_matches = sum(english_eligible & matched_count > 0, na.rm = TRUE),
  not_scored = sum(!english_eligible), .groups = "drop")
kable(coverage)
batch retained eligible with_dictionary_matches not_scored
Scrape 1 96 45 21 51
Scrape 2 15 0 0 15
Scrape 3 97 45 21 52
sentiment_summary <- scored %>% filter(english_eligible) %>%
  count(batch, automated_label, name = "posts") %>% group_by(batch) %>%
  mutate(percent_of_eligible = 100 * posts / sum(posts)) %>% ungroup()
kable(sentiment_summary, digits = 1,
      caption = "Percentages use all English-eligible retained posts in the corresponding sample")
Percentages use all English-eligible retained posts in the corresponding sample
batch automated_label posts percent_of_eligible
Scrape 1 Negative 5 11.1
Scrape 1 No matches 24 53.3
Scrape 1 Positive 16 35.6
Scrape 3 Negative 5 11.1
Scrape 3 No matches 24 53.3
Scrape 3 Positive 16 35.6
write_csv(scored, file.path(output_dir, "automated_sentiment.csv"))
write_csv(matched_words, file.path(output_dir, "sentiment_word_matches.csv"))

Scrape 2 has no language metadata in the supplied file and is not automatically scored. Read its posts and document their language before extending English scoring to them. Scrape 3 repeats first-scrape IDs and has no documented collection time. This report therefore does not claim a sentiment trend across the three supplied versions.

if (nrow(sentiment_summary) > 0) {
  plotly::plot_ly(sentiment_summary, x = ~batch, y = ~percent_of_eligible,
          color = ~automated_label, type = "bar",
          text = ~paste0(posts, " posts; ", round(percent_of_eligible, 1), "%"),
          hovertemplate = "%{x}<br>%{text}<extra>%{fullData.name}</extra>") %>%
    plotly::layout(title = "Dictionary sentiment among English-eligible posts",
           barmode = "stack", xaxis = list(title = "Sample with language coverage"),
           yaxis = list(title = "Percent of English-eligible posts", range = c(0, 100)))
} else cat("No English-eligible posts were available for dictionary sentiment analysis.")

7. Check the supplied manual sentiment sample

The workbook’s labels are retained as supplied. A favorable promotional description is not necessarily a customer’s endorsement, and a negative running experience is not necessarily criticism of the shoe brand. Inspect the stated reasons rather than treating the labels as a gold standard.

manual <- manual_raw %>% transmute(
  manual_row = row_number(), author = `Author name`,
  text_original = `Post text`, text_key = str_to_lower(normalize_text(`Post text`)),
  manual_label = str_to_title(str_trim(Sentiment)),
  supplied_reason = `Reason (manual tag)`)
stopifnot(all(manual$manual_label %in% c("Positive", "Negative", "Neutral")))
manual_summary <- manual %>% count(manual_label, name = "posts") %>%
  mutate(percent = 100 * posts / sum(posts))
kable(manual_summary, digits = 1, caption = "Existing supplied labels, recomputed from the 20-row sample")
Existing supplied labels, recomputed from the 20-row sample
manual_label posts percent
Negative 1 5
Neutral 10 50
Positive 9 45
# A text-only match is used because the manual sheet has no stable post ID.
# Match only when exactly one first-scrape record has that normalized text.
manual_lookup <- scored %>% filter(batch == "Scrape 1") %>%
  mutate(text_key = str_to_lower(text_clean)) %>% group_by(text_key) %>%
  filter(n() == 1) %>% ungroup() %>%
  select(text_key, record_id, language, english_eligible, automated_label, score)
manual_comparison <- manual %>% left_join(manual_lookup, by = "text_key") %>%
  mutate(comparison_status = case_when(
    is.na(record_id) ~ "No unique retained text match",
    !english_eligible ~ "Not comparable: language",
    !automated_label %in% c("Positive", "Negative") ~ "Review no-match or balanced label",
    manual_label == automated_label ~ "Same label",
    TRUE ~ "Different label: inspect wording"))
kable(manual_comparison %>% select(manual_row, manual_label, automated_label,
                                   language, comparison_status, supplied_reason))
manual_row manual_label automated_label language comparison_status supplied_reason
1 Neutral Positive en Different label: inspect wording Product listing/discount code; mainly informational promotion, no clear emotional opinion.
2 Positive Not scored: language unknown Not comparable: language Highlights a large discount, framing the product deal favorably.
3 Neutral No matches en Review no-match or balanced label Simple link prompt with no clear positive or negative opinion.
4 Positive No matches en Review no-match or balanced label Uses favorable wording such as “newest” and “limited edition” to promote a release.
5 Positive No matches en Review no-match or balanced label Expresses enthusiastic affection for the art/design.
6 Neutral No matches en Review no-match or balanced label Product name only; no clear opinion.
7 Positive Not scored: language unknown Not comparable: language Frames the price as a deal by comparing it with retail price.
8 Neutral NA NA No unique retained text match Post text is missing, so sentiment cannot be inferred; counted as neutral/unclear.
9 Neutral No matches en Review no-match or balanced label Hashtags only; no clear opinion.
10 Neutral Not scored: language unknown Not comparable: language News-style product headline and a question; does not clearly express an opinion.
11 Positive Not scored: language pt Not comparable: language Promotional wording emphasizes a low price and a coupon.
12 Negative Positive en Different label: inspect wording Uses strongly critical language and explicitly calls for a boycott.
13 Positive Not scored: language fr Not comparable: language Promotional product post invites opinions and presents the item for sale.
14 Positive No matches en Review no-match or balanced label Positive styling suggestion pairing a business suit with Nike sneakers.
15 Positive NA NA No unique retained text match Promotional post highlights the product and price.
16 Positive NA NA No unique retained text match Emphasizes a low price compared with many listings.
17 Neutral Positive en Different label: inspect wording Product listing with no clear evaluative language.
18 Neutral No matches en Review no-match or balanced label Product listing with no clear evaluative language.
19 Neutral Positive en Different label: inspect wording Short product announcement without a clear positive or negative opinion.
20 Neutral Not scored: language unknown Not comparable: language Reports a legal/trademark development in a factual tone.
write_csv(manual_comparison, file.path(output_dir, "manual_vs_automated.csv"))

imported_check <- imported_manual %>% filter(Sentiment != "Total") %>%
  transmute(manual_label = Sentiment, imported_posts = metric_number(`Number of posts`)) %>%
  full_join(manual_summary %>% select(manual_label, posts), by = "manual_label") %>%
  mutate(matches = !is.na(imported_posts) & !is.na(posts) & imported_posts == posts)
kable(imported_check, caption = "Independent reconciliation to the separately supplied sentiment totals")
Independent reconciliation to the separately supplied sentiment totals
manual_label imported_posts posts matches
Positive 9 9 TRUE
Negative 1 1 TRUE
Neutral 10 10 TRUE

Additional supplied Scrape 3 sentiment sample

manual3 <- manual3_raw %>% transmute(
  raw_row = as.integer(`Raw row`), author = Author, text_original = `Post text`,
  manual_label = str_to_title(str_trim(Sentiment)),
  supplied_reason = `Coding rationale`, sampling_note = `Random method`)
stopifnot(all(manual3$manual_label %in% c("Positive", "Negative", "Neutral")))
manual3_summary <- manual3 %>% count(manual_label, name = "posts") %>%
  mutate(percent = 100 * posts / sum(posts))
kable(manual3_summary, digits = 1, caption = "Separate supplied Scrape 3 sample (15 labels)")
Separate supplied Scrape 3 sample (15 labels)
manual_label posts percent
Negative 2 13.3
Neutral 5 33.3
Positive 8 53.3
manual3_comparison <- manual3 %>% left_join(
  scored %>% filter(batch == "Scrape 3") %>%
    select(raw_row = source_row, record_id, automated_label, score, language), by = "raw_row")
kable(manual3_comparison %>% select(raw_row, manual_label, automated_label, language,
                                     text_original, supplied_reason))
raw_row manual_label automated_label language text_original supplied_reason
83 Negative Negative en Check out what I’m selling. OT Sports blank Minor League Baseball Tacoma Rainers jersey (size M, w/stains): Get up to $20 off* when you use my code JCGUWH to sign up for Mercari. *Terms apply #mercari www.mercari.com/us/item/m441
 Complaint, negative affect, or unfavorable cue.
16 Positive Not scored: language fr #ad

👀 EN STOCK CHEZ NIKE

👟 Chaussures de running sur route Nike Journey Run pour homme - Noir

En stock chez Nike Ă  70€ đŸ”„

Celle-lĂ  mĂ©rite un coup d’Ɠil

#Nike #Sneakers

https://www.kickterminal.com/cop/nike/2q3KgQ?c=bluesky |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 5|Neutral |Not scored: language |pt |Ah, sim, meus queridos Nike, Adidas e
 |Primarily factual/ambiguous; no clear positive or negative evaluation. | | 96|Negative |Positive |en |The Golden Gate Bridge taken from an outlook near the Nike Missile Silo. â˜č |Complaint, negative affect, or unfavorable cue. | | 37|Positive |Not scored: language |unknown |Nike G.T. Future ‘DMV’

Available via @nikestore for $110.97 (retail $200) #ad

https://ow.ly/G2B550ZY5qE |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 33|Neutral |Not scored: language |unknown |7-Eleven’s trademark case over Nike’s convenience-store Air Max 95 ended with prejudice and no explanation. Here’s where the shoe stands now. |Primarily factual/ambiguous; no clear positive or negative evaluation. | | 30|Positive |Not scored: language |unknown |Nike Ja 3 ‘Animal Pack’

Available via @footlocker |$130| #ad @JaMorant

https://ow.ly/Nhop50ZY5Pa |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 19|Positive |Not scored: language |fr |#ad

👀 REPÉRÉ CHEZ NIKE

👟 Chaussure de tennis pour surface dure Nike Vapor Lite 3 pour homme - Bleu

Disponible sur le site, 59€ 👀

Elle vous tente ?

#Nike #Sneakers

https://www.kickterminal.com/cop/nike/4rxAyV?c=bluesky |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 15|Positive |Not scored: language |fr |#ad

👀 REPÉRÉ CHEZ NIKE

👟 Chaussure d’entraünement Nike Metcon 10 AMP pour homme - Rouge

En stock chez Nike Ă  105€ đŸ”„

Celle-lĂ  mĂ©rite un coup d’Ɠil

#Nike #Sneakers

https://www.kickterminal.com/cop/nike/9df2he?c=bluesky |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 88|Positive |Not scored: language |unknown |đŸ”„ TĂȘnis Nike Uplift Sc Masculino

đŸ’” Por: R$ 266,79 đŸŽŸïž Cupom: MODA2310 https://meli.la/2Y4c5g9

‱ AnĂșncio Mercado Livre |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 71|Neutral |Negative |en |I mean I don’t know that Travis Knight, Nike scion, is hurting for money, but Laika movies are extremely time and labor intensive endeavors and they spent $100 million dollars making a stop motion movie, and by they, I mean also he’s the producer and head of the company that produced it. |Primarily factual/ambiguous; no clear positive or negative evaluation. | | 13|Positive |Not scored: language |fr |#ad

👀 31 OCTOBRE CHEZ ASPHALTGOLD

👟 Nike Air Jordan 14 Retro Forest Green — Gorge Green Light Graphite

PrĂ©vu pour le 31 octobre đŸ”„

Vous comptez tenter ? 👀

#Jordan #AirJordan14 #AJ14 #Sneakers #SneakerRelease

https://www.kickterminal.com/cop/asphaltgold/3STrjq?c=bluesky |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 77|Positive |Not scored: language |unknown |Nike x SKIMS just dropped a bold red collection, these are the styles I’d shop first |Promotional/enthusiastic language, product availability, deal, or favorable shopping cue. | | 56|Neutral |No matches |en |TĂȘnis Nike Mc Trainer 3 Masculino |Primarily factual/ambiguous; no clear positive or negative evaluation. | | 6|Neutral |Not scored: language |unknown |Nike Tech Fleece (M), L — $11 all-in. The lowest of 1224 current size-L listings in the same condition on the same site (oldest listed ~28 months ago). (sized from a single reference garment) |Primarily factual/ambiguous; no clear positive or negative evaluation. |

write_csv(manual3_comparison, file.path(output_dir, "scrape3_manual_vs_automated.csv"))

The source labels report their sampling method as a fixed-seed random sample; this is a supplied claim rather than a reconstruction of that sampling process. The 15 labels differ from the earlier 20-row sample and should not be pooled as 35 unique people or used to estimate a sentiment change. Inspect each rationale. For example, a resale description mentioning a stained jersey may describe a product’s condition without expressing dissatisfaction with the brand.

8. Verify the AI-written analysis

Sanity checks

percent_check <- sentiment_summary %>% group_by(batch) %>%
  summarise(total = sum(percent_of_eligible), .groups = "drop")
checks <- tibble(
  check = c("Source row accounting balances for each batch",
            "No repeated within-batch post keys remain",
            "All analytic text is nonblank",
            "One sentiment output per retained record",
            "Every retained post has an automated status",
            "Matched dictionary words reconcile to per-post counts",
            "Eligible sentiment percentages total 100 per sample",
            "Supplied sentiment totals reconcile to sample labels"),
  passed = c(
    all(audit$source_records == audit$unusable_text + audit$duplicates_removed + audit$retained),
    !anyDuplicated(paste(clean$batch, clean$dedup_key)),
    all(nchar(clean$text_clean) > 0),
    nrow(scored) == nrow(clean) && !anyDuplicated(scored$record_id),
    all(!is.na(scored$automated_label)),
    sum(scored$matched_count, na.rm = TRUE) == nrow(matched_words),
    all(abs(percent_check$total - 100) < 1e-8),
    all(imported_check$matches)))
kable(checks)
check passed
Source row accounting balances for each batch TRUE
No repeated within-batch post keys remain TRUE
All analytic text is nonblank TRUE
One sentiment output per retained record TRUE
Every retained post has an automated status TRUE
Matched dictionary words reconcile to per-post counts TRUE
Eligible sentiment percentages total 100 per sample TRUE
Supplied sentiment totals reconcile to sample labels TRUE
stopifnot(all(checks$passed))

These tests check internal consistency, not the truth of opinions or the representativeness of the collection.

Hand calculation and direct source spot check

The first record in the supplied All Posts table lists 25 likes and 1 repost. The hand calculation is 25 + 1 = 26. Verify these values in Excel as well.

first_record <- clean %>% filter(batch == "Scrape 1", source_row == 2L)
stopifnot(nrow(first_record) == 1L)
hand_check <- tibble(record_id = first_record$record_id,
                     hand_expected = 26, code_total = first_record$engagement,
                     matches_expected = first_record$likes == 25 &
                       first_record$reposts == 1 & first_record$engagement == 26)
kable(hand_check)
record_id hand_expected code_total matches_expected
Scrape1-row2 26 26 TRUE

The expected value refers to this supplied workbook snapshot. A mismatch in a future refreshed workbook calls for a new source check rather than forcing the data to match an old expected number.

Inspect exact sentiment words and complete your own review

set.seed(4000)
eligible_review_ids <- scored$record_id[scored$english_eligible %in% TRUE]
review_ids <- if (length(eligible_review_ids) > 0L) {
  sample(eligible_review_ids, size = min(10L, length(eligible_review_ids)), replace = FALSE)
} else character(0)
review_records <- scored %>% filter(record_id %in% review_ids) %>%
  select(record_id, text_original, automated_label, positive_words, negative_words, score)
kable(review_records, caption = "Reproducible sample for your own contextual review")
Reproducible sample for your own contextual review
record_id text_original automated_label positive_words negative_words score
Scrape1-row21 Tottenham 2027/28 away kit leak reveals historic pink design for outfield players

www.spurs-web.com/spurs-news/t
 |Negative | 0| 1| -1| |Scrape1-row29 |If I’m Nike I’d be asking a lot of questions, too. |No matches | 0| 0| 0| |Scrape1-row35 |Business suit and Nike sneakers

For the working woman in your households |No matches | 0| 0| 0| |Scrape1-row42 |Please and thank you. |Positive | 1| 0| 1| |Scrape1-row96 |The Golden Gate Bridge taken from an outlook near the Nike Missile Silo. â˜č |Positive | 1| 0| 1| |Scrape3-row56 |TĂȘnis Nike Mc Trainer 3 Masculino |No matches | 0| 0| 0| |Scrape3-row66 |🛎 Mochila Masculina Nike Heritage Backpack cor preta por R$ 115

Ganhe R$30 no Mercado Pago: https://mpago.li/2gudrhL use o cupom mercado livre MODA2310 https://meli.la/1jf9P2M

AnĂșncio |📍Meus Grupos: entr.ai/lamorimpromos |No matches | 0| 0| 0| |Scrape3-row78 |Check out Nike NFL Seattle Seahawks Elite 51 blank navy blue home jersey (adult size 52) ebay.io/m/kgyj0X #eBay via @eBay |Positive | 1| 0| 1| |Scrape3-row81 |Check out what I’m selling. MLB Nike Los Angeles Dodgers Shohei Ohtani replica #17 jersey (NWT/adult sz XL): Get up to $20 off* when you use my code JCGUWH to sign up for Mercari. *Terms apply #mercari www.mercari.com/us/item/m588
 |No matches | 0| 0| 0| |Scrape3-row87 |đŸ”„ TĂȘnis Nike Uplift Sc Masculino

🚹 R$ 266 đŸŽŸïž Cupom: MODA2310

🔗 https://meli.la/1PZ54Me

#Revelando #Jonatas #Entregou #oferta #promo |Positive | 1| 0| 1|

kable(matched_words %>% filter(record_id %in% review_ids) %>%
        count(record_id, word, sentiment), caption = "Exact word matches for the review sample")
Exact word matches for the review sample
record_id word sentiment n
Scrape1-row21 leak negative 1
Scrape1-row42 thank positive 1
Scrape1-row96 golden positive 1
Scrape3-row78 elite positive 1
Scrape3-row87 uplift positive 1
# Complete reviewer_label and reviewer_note in lab_outputs/my_verification.csv.
# It is created once and then retained so re-knitting does not overwrite your work.
verification_path <- file.path(output_dir, "my_verification.csv")
if (!file.exists(verification_path)) {
  write_csv(review_records %>% mutate(reviewer_label = "", reviewer_note = ""), verification_path)
}
verification_notes <- read_csv(verification_path, show_col_types = FALSE,
                               col_types = cols(.default = col_character()))
stopifnot(all(c("record_id", "reviewer_label", "reviewer_note") %in% names(verification_notes)))
stopifnot(!anyDuplicated(verification_notes$record_id))
review_progress <- tibble(
  record_id = review_ids) %>%
  left_join(verification_notes %>% select(record_id, reviewer_label, reviewer_note), by = "record_id") %>%
  mutate(completed = !is.na(reviewer_label) & str_trim(reviewer_label) != "" &
                      !is.na(reviewer_note) & str_trim(reviewer_note) != "")
kable(review_progress)
record_id reviewer_label reviewer_note completed
Scrape3-row56 NA NA FALSE
Scrape1-row35 NA NA FALSE
Scrape3-row87 NA NA FALSE
Scrape1-row96 NA NA FALSE
Scrape1-row42 NA NA FALSE
Scrape3-row81 NA NA FALSE
Scrape3-row78 NA NA FALSE
Scrape3-row66 NA NA FALSE
Scrape1-row29 NA NA FALSE
Scrape1-row21 NA NA FALSE
cat("Completed contextual reviews:", sum(review_progress$completed), "of", nrow(review_progress))
## Completed contextual reviews: 0 of 10

Open lab_outputs/my_verification.csv, read the original wording, and enter Positive, Negative, Mixed, Neutral, or Unclear plus a short reason. Save and re-knit. Do not report that you personally verified labels until you have done so. The file is keyed by record ID; re-create it deliberately if a new source workbook changes the selected sample.

9. Findings, limitations, and recommendation

The report imports 215 source records and retains 208 records for text analysis after the stated exclusions. The supplied manual sample contains 20 labels, with 15 additional supplied labels in the third file. These are separate measures and should not be added together.

Complete after inspecting the computed tables and original posts:

  1. Recurring themes: Name two or three themes and identify specific posts supporting each. Word frequency alone does not establish a theme. Compare batches separately so repeated Scrape 1/3 posts are not mistaken for independent evidence.
  2. Engagement: Explain a common feature among the highest-engagement records and whether a few accounts dominate the retained sample.
  3. Sentiment: Explain the Scrape 1/3 ID overlap and do not interpret repeated posts as independent observations. Report the automated distribution with its English-only denominator. Separately describe the existing manual sample; do not apply its percentages to every post.
  4. AI critique: Describe a concrete mismatch or assumption you checked, what the original code or dictionary missed, and how your judgment changes the interpretation.
  5. Recommendation: Propose one cautious marketing action tied to actual text. Avoid claiming a platform-wide trend, customer behavior change, or sales effect.

My final interpretation: [Replace with your results, one checked example, one limitation, and one recommendation.]

10. Actual AI requests and a more specific follow-up

Two requests made in this conversation were:

Request 1: “can you add this data into my excel sheet”

Response summary: The assistant produced an updated workbook with separate tabs for both scrapes, collection notes, and the supplied sentiment summary, preserving the four original analysis tabs. It noted that Scrape 2 has 15 posts. This is a summary, not a verbatim transcript; screenshot the actual exchange.

Request 2 (opening phrase): “with that can you give me a rmarkdown script”

The second request also included the lab requirements: a beginning reflection, cleaning, text and sentiment analysis, Plotly, verification, AI critique, and prompt screenshots. This report is the assistant’s response. Screenshot that request and the response as well.

Suggested improved follow-up — send after you run the report:

Review my R Markdown results using readxl, dplyr, tidytext, and plotly. My input workbook is Social_Media_Analysis_Updated.xlsx; the data come from All Posts, Scrape 1 Raw, Scrape 2, and Sentiment Sample. Verify that duplicate representations are not counted twice, missing text is excluded transparently, and English Bing sentiment is not applied to posts with other or unknown languages. Preserve relative timestamps and do not interpret 100 versus 15 source records as a change in platform volume. Inspect the checks, matched words, and my manual review below; identify one unsupported interpretation and provide corrected R code and an explanation. Do not claim to have run anything you did not run. [Paste my actual output, review notes, and any error messages here.]

This suggested follow-up is not a completed exchange. Record the actual prompt you send and the actual response received. Post screenshots of the exchanges to the discussion forum; do not fabricate screenshots or responses.

11. Submission checklist

  • Complete your contextual review and final interpretation.
  • Revise the two-paragraph reflection to describe your actual experience and feedback on the new lab. Remove the draft notice after doing so.
  • Knit this report to HTML and verify the interactive charts respond to hover.
  • In the HTML preview choose Publish, select RPubs, and sign in.
  • Submit the specific published document URL, not only your profile URL.
  • Attach screenshots of your actual AI prompts and responses to the discussion forum.

References and reproducibility

sessionInfo()
## R version 4.6.2 (2026-10-29)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.5 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0
## 
## locale:
##  [1] LC_CTYPE=C.UTF-8       LC_NUMERIC=C           LC_TIME=C.UTF-8       
##  [4] LC_COLLATE=C.UTF-8     LC_MONETARY=C.UTF-8    LC_MESSAGES=C.UTF-8   
##  [7] LC_PAPER=C.UTF-8       LC_NAME=C              LC_ADDRESS=C          
## [10] LC_TELEPHONE=C         LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C   
## 
## time zone: UTC
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] knitr_1.52      plotly_4.12.1   ggplot2_4.0.4   tidytext_0.4.3 
##  [5] lubridate_1.9.5 readr_2.2.0     tibble_3.3.1    stringr_1.6.0  
##  [9] tidyr_1.3.2     dplyr_1.2.1    
## 
## loaded via a namespace (and not attached):
##  [1] janeaustenr_1.0.0   sass_0.4.10         generics_0.1.4     
##  [4] stringi_1.8.9       lattice_0.23-1      hms_1.1.4          
##  [7] digest_0.6.39       magrittr_2.0.5      evaluate_1.0.5     
## [10] grid_4.6.2          timechange_0.4.0    RColorBrewer_1.1-3 
## [13] fastmap_1.2.0       jsonlite_2.0.0      Matrix_1.7-6       
## [16] httr_1.4.9          purrr_1.2.2         crosstalk_1.2.2    
## [19] viridisLite_0.4.3   scales_1.4.0        jquerylib_0.1.4    
## [22] cli_3.6.6           crayon_1.5.3        rlang_1.3.0        
## [25] tokenizers_0.3.0    bit64_4.8.6         withr_3.0.3        
## [28] cachem_1.1.0        yaml_2.3.12         otel_0.3.0         
## [31] parallel_4.6.2      tools_4.6.2         tzdb_0.5.0         
## [34] vctrs_0.7.3         R6_2.6.1            lifecycle_1.0.5    
## [37] bit_4.6.0           htmlwidgets_1.6.4   vroom_1.7.1        
## [40] pkgconfig_2.0.3     pillar_1.11.1       bslib_0.12.0       
## [43] gtable_0.3.6        glue_1.8.1          data.table_1.18.6.1
## [46] Rcpp_1.1.2          xfun_0.62           tidyselect_1.2.1   
## [49] farver_2.1.2        htmltools_0.5.9     SnowballC_0.7.1    
## [52] rmarkdown_2.32      compiler_4.6.2      S7_0.2.2