Bibliographic record
Abstract
Our daily life is increasingly in uenced by data-driven decision processes, both for good (e.g., machine learning algorithms being used to detect cancer ) and bad purposes (e.g., Cambridge Analytica in uencing the U.S. presidential elections ).These decision processes rely on a large amount of (input) data.To create high-quality decision processes, a large amount of high-quality data is needed.However, it is infeasible to integrate large amounts of diverse data manually.This is one of the reasons that a set of technologies, namely, Semantic Web technologies, are proposed.Semantic Web technologies allow generating and processing so-called knowledge graphs: a means to exchange data values, and additionally their meaningful (i.e., semantic) relationships, between multiple agents.These agents generate and use data in di erent ways on a large scale.Knowledge graphs thus ease (automatic) data integration, and can thus be an essential element of future data-driven decision processes.However, automatic processing (of knowledge graphs) lacks manual inspection (i.e., human supervision that can help to interpret and nuance results), and thus requires a strong level of quality assessment.Without quality assessment, the past has shown that data can be incorrectly integrated due to semantic di erences (e.g., mixing the metric and imperial system, which led to the crash of a NASA Mars orbiter ) or processed badly (e.g., introducing rounding errors, halving the value of the Vancouver Stock Exchange in less than a year ).If we want to use knowledge graphs in a real-world context that does not result in crashes or massive money loss, we need to improve and assess its quality, both on the level of data values and semantic relationships.This dissertation focuses exactly on this: improving data quality and assessing semantic quality of knowledge graphs.We speci cally investigate following two challenges that complementary tackle knowledge graph quality.The rst challenge is including data transformations in knowledge graphs as this can help cleaning the data (e.g., by including data nor-
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 imitationNot 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.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.038 | 0.042 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.000 | 0.004 |
| Open science | 0.007 | 0.004 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
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".