Bridging the Gap between Intrinsic and Perceived Relevance in Snippet Generation
Bibliographic record
Abstract
Snippet generation plays an important role in a search engine. Good snippets provide users a good indication on the main content of a search result related to the query and on whether one can find relevant information in it. Previous studies on snippet generation focused on selecting sentences that are related to the query and to the document. However, resulting snippet may look highly relevant while the document itself is not. A missing factor that has not been considered is the consistency between the perceived relevance by the user in reading the snippet and the intrinsic relevance of the document. This factor is important to avoid generating a seemingly relevant snippet for an irrelevant document and vice versa. In this paper, we incorporate this factor in a snippet generation method that imposes the constraint that the snippet of a more relevant document should also be more relevant to the query. We derive a set of pairwise preferences between sentences from relevance judgments. We then use this set to train a gradient boosting decision tree to model a sentence scoring function used in snippet generation. Compared to the existing snippet generation methods and to the snippets generated by a commercial search engine, our snippets are more consistent with the true relevance of the documents. When the snippets are incorporated into a document ranking function, we also observe a significant improvement in retrieval effectiveness. This study shows the importance to generate snippets indicating the right level of relevance to the search results.
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 | 0.006 | 0.037 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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".