Federating natural language question answering services of a cognitive enterprise data platform
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
An enterprise data lake (EDL) combines big data storage, governance, and query abilities for structured and unstructured data with a navigable, searchable data catalogue. We define a cognitive enterprise data platform (CEDP) to be an EDL that is further equipped with a scalable deployment platform and an extensible catalogue of deployable cognitive computing services as well as a data science and data engineering environment to develop and train the cognitive computing services and publish them to the CEDP catalogue. A natural language question answering (NLQA) service is a CEDP cognitive computing service trained to recognize natural language questions and respond using CEDP data queries or cognitive computing services. In order to scale this form of cognition to the enterprise, business units must be able to crowd source the catalogue of trained NLQA that the CEDP must then deploy and federate automatically. However, the machine learned models that contribute to answer confidence values are separately trained, so the answer confidence values from any two NLQA services are not directly comparable. Therefore, federating separately trained NLQA services requires an answer ranking methodology. This paper includes a solution that is based on two insights. The first is that the problem of answering ranking across separately trained NLQA services is analogous to the left side of Bayes' formula. The second insight is that the factors in the right side of Bayes' formula can be automatically machine learned using the test sets of the NLQA services. Thus, calibrated answer ranking across separately trained NLQA services is achieved via Bayesian inferences on their answer confidence values. In turn, this baseline answer ranking methodology enables a cognitive enterprise data platform to automatically federate a dynamic changeable crowd-sourced catalogue of NLQA services.
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.
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 | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.000 | 0.000 |
| 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 it