{"paper":{"title":"Beyond Cosine Similarity: Zero-Initialized Residual Complex Projection for Aspect-Based Sentiment Analysis","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Projecting ABSA features into complex space via zero-initialized residuals separates opposing polarities by phase.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Fandi Sun, Haoyu Wen, Yijin Wang","submitted_at":"2026-03-30T09:23:04Z","abstract_excerpt":"Aspect-Based Sentiment Analysis (ABSA) faces critical challenges due to representation entanglement and false-negative collisions in real-valued embedding spaces. In this paper, we propose a novel framework featuring a Zero-Initialized Residual Complex Projection (ZRCP) and an Anti-collision Masked Angle Loss. Our approach projects textual features into a complex semantic space, utilizing the phase to isolate sentiment polarities while regularizing the amplitude to ensure structural consistency within aspect categories. To mitigate this, we introduce an anti-collision mask that preserves intra"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our framework achieves a state-of-the-art Macro-F1 score of 0.8923 on the ASAP dataset, outperforming robust baselines by projecting textual features into a complex semantic space and using an anti-collision mask to expand the discriminative margin between opposing polarities.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That mapping features to complex space via zero-initialized residual projection plus the masked angle loss will reliably isolate polarities and generalize beyond the ASAP dataset without introducing new representation artifacts or requiring extensive hyperparameter tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Zero-Initialized Residual Complex Projection with anti-collision masked angle loss reaches 0.8923 Macro-F1 on the ASAP dataset for aspect-based sentiment analysis.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Projecting ABSA features into complex space via zero-initialized residuals separates opposing polarities by phase.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"906fe689a03a09675ce42f0bd44c24a9cf2df4558e2373c170dafc02ba2b8cd7"},"source":{"id":"2603.28205","kind":"arxiv","version":2},"verdict":{"id":"74f0f157-0e03-4a44-b02d-958f56de4009","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:38:29.810879Z","strongest_claim":"Our framework achieves a state-of-the-art Macro-F1 score of 0.8923 on the ASAP dataset, outperforming robust baselines by projecting textual features into a complex semantic space and using an anti-collision mask to expand the discriminative margin between opposing polarities.","one_line_summary":"Zero-Initialized Residual Complex Projection with anti-collision masked angle loss reaches 0.8923 Macro-F1 on the ASAP dataset for aspect-based sentiment analysis.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That mapping features to complex space via zero-initialized residual projection plus the masked angle loss will reliably isolate polarities and generalize beyond the ASAP dataset without introducing new representation artifacts or requiring extensive hyperparameter tuning.","pith_extraction_headline":"Projecting ABSA features into complex space via zero-initialized residuals separates opposing polarities by phase."},"references":{"count":22,"sample":[{"doi":"","year":2020,"title":"Issues and challenges of aspect-based sentiment analysis: A comprehensive survey","work_id":"2093f27a-95a9-44e9-a8c8-35dc7dbd188f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Survey on aspect-level sentiment analysis","work_id":"7cba367a-13de-4daf-ac5d-37c80119b189","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Contrastive sentence representation learning with adaptive false negative cancellation","work_id":"af903eef-19f3-40b9-a00e-a740137971b7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"QPEN: quantum projection and quantum entanglement enhanced network for cross-lingual aspect-based sentiment analysis","work_id":"82ee21c3-bf67-40a8-9e1f-b31be9c5dc80","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1912,"title":"Encoding word order in complex embeddings","work_id":"8bd34535-8adc-4697-a88c-3cfeef78c54b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":22,"snapshot_sha256":"c3168477490be76dc80b0d6979ab6c8ca87574297cb2d7efddad498042089222","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}