Adding a Capability to Extract Sentiment from Text Using HanDles
Bibliographic record
Abstract
Abstract : HanDles is a document visualization tool developed by Ohio State University for DRDC Toronto. One aspect of documents that might be of interest to analysts is the extent to which they express positive or negative opinion or sentiment toward some issue or group. In this report, we describe how HanDles was extended to include the ability to classify documents as containing predominantly positive or negative sentiment. The capability was added to the tool so that it could be used in Influence Operations contexts. As a test case, we trained HanDles to distinguish good and poor film reviews, and then tested it three times to see how well it classified documents. The first test was conducted on reviews of the Amazon Kindle. The second test was run on text segments of the original training set of movie reviews, and finally, it was tested on a set of movie reviews that it had not seen before. In general, HanDles did a poor job detecting the sentiment associated with the reviews of the Amazon Kindle. We attribute the poor performance to the fact that movie and product reviews discuss different issues, and as such, there is limited similarity in the two classes of document. Not surprisingly, HanDles did a good job classifying text segments of the original training set. Also, the finding demonstrated that, unlike many other sentiment analysis tools that only classify text at the whole-document level, HanDles can be used effectively to extract the issues being discussed within documents, and assign sentiment to those. For example, a review of a film might be classified as negative overall, but HanDles can determine that the acting was good, but the directing was poor. Finally, when we tested HanDles on a new set of movie reviews it had not seen before, it performed with 93.3% accuracy. The results of our trial suggest that there must be some similarity between the documents used during training and those used in the operational context for HanDles to work properly.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.008 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.005 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.015 | 0.007 |
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 source (direct Gemma or distilled Codex), 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".