Data and code for: Aligned, obscured, or unanswered — Measuring how research on AI and communication tracks public concerns voiced in YouTube comments

Data and code accompanying the manuscript "Aligned, obscured, or unanswered: Measuring how research on artificial intelligence and communication tracks public concerns voiced in YouTube comments" (Liyao Xiao, Yi He). The deposit contains all materials needed to reproduce the reported results: 01_literature_corpus – Web of Science record identifiers for the 8,238 retained articles (from 11,574 retrieved), screening flags, LDA topic assignments (k = 10, seed 42), the communication-tagged subset (n = 1,512), abstract-level embeddings, and top-15 term weights per topic. 02_comment_corpus – the 293 modelled video identifiers, the 1,225-video sampling frame, 18,479 comment records (hashed comment IDs, video membership, BERTopic topic assignments, k = 18), and comment-level embeddings. 03_topic_units – the 86 topic units (LDA and BERTopic term bags, 18 academic-register paraphrase bags, shuffled bags) with recorded hashes and their embeddings. 04_alignment_channels – similarity matrices, best matches and 180,000-value permutation null distributions for the bag-of-words, paraphrase and communication-subset channels; shuffled-text inputs; q-value tables; period-by-channel results. 05_external_checks – the 7,741-comment unseeded independent sample (16 queries, 94 videos: kNN assignments, topic and group shares, slop annotations, embeddings, frozen-partition gate report); 22 perturbation runs (video exclusions, comment caps, k/seed variations); expert coding by topic; family-drop and self-seeding results. 06_code – runnable verifier, archived analysis scripts, study configuration (64 seeded search queries, collection caps, model settings), and recovery scripts. 07_audit – numeric verification of reported figures, source provenance, privacy/secret scan, recovery audits, and the locked environment (Python 3.12; scikit-learn 1.9.0; BERTopic 0.17.4; sentence-transformers 5.6.0; umap-learn 0.5.12). README.md gives a directory-to-manuscript-section map, verification instructions, and known limitations (including which historical cross-validation and correction figures could not be regenerated). MANIFEST_SHA256.txt lists checksums for every file. Raw YouTube comment text and full Web of Science records are not included because they cannot be redistributed under the YouTube API Terms of Service and the Clarivate license. They can be re-obtained through the YouTube Data API using the video identifiers provided, and through a Web of Science subscription using the search strings provided. Comment identifiers are hashed; no human subjects were involved. License: data and documentation under CC BY 4.0; code under the MIT License (see LICENSE in the archive).