MétaCan
Menu
Back to cohort
Record W2163542005 · doi:10.2196/medinform.4211

Context-Sensitive Spelling Correction of Consumer-Generated Content on Health Care

2015· article· en· W2163542005 on OpenAlexvenueno aff
An Zheng, Jiaheng Yin, Rudan Chen, Xianyang Zhao, Wei Xu, Wenqing Cheng, Tian Xia, Simon Lin

Bibliographic record

VenueJMIR Medical Informatics · 2015
Typearticle
Languageen
FieldComputer Science
TopicNatural Language Processing Techniques
Canadian institutionsnot available
Fundersnot available
KeywordsSpellingContext (archaeology)Perspective (graphical)Computer scienceSocial mediaHealth careContent (measure theory)User-generated contentInternet privacyWorld Wide WebAdvertisingArtificial intelligenceLinguisticsBusiness

Abstract

fetched live from OpenAlex

BACKGROUND: Consumer-generated content, such as postings on social media websites, can serve as an ideal source of information for studying health care from a consumer's perspective. However, consumer-generated content on health care topics often contains spelling errors, which, if not corrected, will be obstacles for downstream computer-based text analysis. OBJECTIVE: In this study, we proposed a framework with a spelling correction system designed for consumer-generated content and a novel ontology-based evaluation system which was used to efficiently assess the correction quality. Additionally, we emphasized the importance of context sensitivity in the correction process, and demonstrated why correction methods designed for electronic medical records (EMRs) failed to perform well with consumer-generated content. METHODS: First, we developed our spelling correction system based on Google Spell Checker. The system processed postings acquired from MedHelp, a biomedical bulletin board system (BBS), and saved misspelled words (eg, sertaline) and corresponding corrected words (eg, sertraline) into two separate sets. Second, to reduce the number of words needing manual examination in the evaluation process, we respectively matched the words in the two sets with terms in two biomedical ontologies: RxNorm and Systematized Nomenclature of Medicine -- Clinical Terms (SNOMED CT). The ratio of words which could be matched and appropriately corrected was used to evaluate the correction system's overall performance. Third, we categorized the misspelled words according to the types of spelling errors. Finally, we calculated the ratio of abbreviations in the postings, which remarkably differed between EMRs and consumer-generated content and could largely influence the overall performance of spelling checkers. RESULTS: An uncorrected word and the corresponding corrected word was called a spelling pair, and the two words in the spelling pair were its members. In our study, there were 271 spelling pairs detected, among which 58 (21.4%) pairs had one or two members matched in the selected ontologies. The ratio of appropriate correction in the 271 overall spelling errors was 85.2% (231/271). The ratio of that in the 58 spelling pairs was 86% (50/58), close to the overall ratio. We also found that linguistic errors took up 31.4% (85/271) of all errors detected, and only 0.98% (210/21,358) of words in the postings were abbreviations, which was much lower than the ratio in the EMRs (33.6%). CONCLUSIONS: We conclude that our system can accurately correct spelling errors in consumer-generated content. Context sensitivity is indispensable in the correction process. Additionally, it can be confirmed that consumer-generated content differs from EMRs in that consumers seldom use abbreviations. Also, the evaluation method, taking advantage of biomedical ontology, can effectively estimate the accuracy of the correction system and reduce manual examination time.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.000
metaresearch head score (Gemma)0.000
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Other design · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.869
Threshold uncertainty score0.408

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0000.000
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0000.000
Science and technology studies0.0000.000
Scholarly communication0.0000.000
Open science0.0000.000
Research integrity0.0000.000
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.044
GPT teacher head0.325
Teacher spread0.281 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one teacher head, not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designOther design
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations16
Published2015
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Medical InformaticsSame topicNatural Language Processing TechniquesFrench-language works237,207