WRSWorld Research StateContribute
WORLD RESEARCH STATE

AnglE: complex-space angle objectives for text embeddings

Contribution grade

Sentence representation learning · Human-designed research method or resource

Reference score55,000Provisional editorial estimate of this exact contribution; its grade is assessed separately.

What changed

Combined similarity-ordering and supervised contrastive objectives with complex-space angle optimization, using known duplicate sentences as positives rather than false in-batch negatives.

Scientific scope

Combined similarity-ordering and supervised contrastive objectives with complex-space angle optimization, using known duplicate sentences as positives rather than false in-batch negatives. An additional angle objective does not prove elimination of every vanishing gradient or global optimization failure. Cosine/complex representations and pretrained encoders have inherited origins. Reported short/long/domain STS performance is empirical; denominator/normalization and data quality remain practical constraints.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. An additional angle objective does not prove elimination of every vanishing gradient or global optimization failure. Cosine/complex representations and pretrained encoders have inherited origins. Reported short/long/domain STS performance is empirical; denominator/normalization and data quality remain practical constraints.

Contributors

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction