Sensory Descriptor Analysis of Whisky Lexicons through the Use of Deep Learning
Bibliographic record
Abstract
This paper is concerned with extracting relevant terms from a text corpus on whisk(e)y. "Relevant" terms are usually contextually defined in their domain of use. Arguably, every domain has a specialized vocabulary used for describing things. For example, the field of Sensory Science, a sub-field of Food Science, investigates human responses to food products and differentiates "descriptive" terms for flavors from "ordinary", non-descriptive language. Within the field, descriptors are generated through Descriptive Analysis, a method wherein a human panel of experts tastes multiple food products and defines descriptors. This process is both time-consuming and expensive. However, one could leverage existing data to identify and build a flavor language automatically. For example, there are thousands of professional and semi-professional reviews of whisk(e)y published on the internet, providing abundant descriptors interspersed with non-descriptive language. The aim, then, is to be able to automatically identify descriptive terms in unstructured reviews for later use in product flavor characterization. We created two systems to perform this task. The first is an interactive visual tool that can be used to tag examples of descriptive terms from thousands of whisky reviews. This creates a training dataset that we use to perform transfer learning using GloVe word embeddings and a Long Short-Term Memory deep learning model architecture. The result is a model that can accurately identify descriptors within a corpus of whisky review texts with a train/test accuracy of 99% and precision, recall, and F1-scores of 0.99. We tested for overfitting by comparing the training and validation loss for divergence. Our results show that the language structure for descriptive terms can be programmatically learned.
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 | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| 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 itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
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".