Conan-embedding-v1オープンソース中国語テキスト埋め込みモデル

ホーム

Conan Embedding V1 Q4 K M GGUF

KenLi315によって開発

Conan-embedding-v1はTencent BACチームによって開発された中国語テキスト埋め込みモデルで、中国語テキストの意味表現と類似度計算に特化しています。

テキスト埋め込み中国語#中国語テキスト埋め込み #マルチタスク評価 #高精度意味類似度

ダウンロード数 48

リリース時間 : 1/28/2025

モデル概要

このモデルは主に中国語テキストの意味埋め込み表現に使用され、テキスト類似度計算、分類、クラスタリング、検索などのさまざまな自然言語処理タスクをサポートします。

モデル特徴

中国語最適化

中国語テキストに特化して最適化されており、中国語の意味的特徴をより効果的に捉えることができます。

マルチタスクサポート

テキスト類似度計算、分類、クラスタリング、検索など、さまざまな自然言語処理タスクをサポートします。

高性能

複数の中国語ベンチマークテストで優れた性能を発揮し、特に意味類似度タスクで顕著です。

モデル能力

テキスト埋め込み

意味類似度計算

テキスト分類

テキストクラスタリング

情報検索

再ランキング

使用事例

情報検索

医療QA検索

医療分野のQA検索システムに使用され、ユーザーが関連する医療情報を迅速に見つけるのを支援します。

CMedqaRetrievalデータセットで良好な性能を発揮し、map@100は42.495に達しました

テキスト類似度

QAペアマッチング

質問と回答の間の意味的関連性を判断します。

BQデータセットでcos_sim_spearmanは74.507に達しました

テキスト分類

商品レビュー分類

ECプラットフォームの商品レビューを感情分類します。

JDReview分類タスクで精度は90.319%に達しました

🚀 conan-embedding

このモデルは、様々な自然言語処理タスクにおいて優れた性能を発揮する埋め込みモデルです。TencentBAC/Conan-embedding-v1をベースに構築され、MTEBベンチマークでの評価結果が提供されています。

📚 詳細ドキュメント

モデル情報

属性	詳情
モデルタイプ	sentence-transformers
ベースモデル	TencentBAC/Conan-embedding-v1
タグ	mteb, llama-cpp, gguf-my-repo
ライセンス	cc-by-nc-4.0

モデル評価結果

1. STSタスク

MTEB AFQMC (validation)
- cos_sim_pearson: 56.613572467148856
- cos_sim_spearman: 60.66446211824284
- euclidean_pearson: 58.42080485872613
- euclidean_spearman: 59.82750030458164
- manhattan_pearson: 58.39885271199772
- manhattan_spearman: 59.817749720366734
MTEB ATEC (test)
- cos_sim_pearson: 56.60530380552331
- cos_sim_spearman: 58.63822441736707
- euclidean_pearson: 62.18551665180664
- euclidean_spearman: 58.23168804495912
- manhattan_pearson: 62.17191480770053
- manhattan_spearman: 58.22556219601401
MTEB BQ (test)
- cos_sim_pearson: 72.6472074172711
- cos_sim_spearman: 74.50748447236577
- euclidean_pearson: 72.51833296451854
- euclidean_spearman: 73.9898922606105
- manhattan_pearson: 72.50184948939338
- manhattan_spearman: 73.97797921509638
MTEB LCQMC (test)
- cos_sim_pearson: 73.18906216730208
- cos_sim_spearman: 79.44570226735877
- euclidean_pearson: 78.8105072242798
- euclidean_spearman: 79.15605680863212
- manhattan_pearson: 78.80576507484064
- manhattan_spearman: 79.14625534068364
MTEB PAWSX (test)
- cos_sim_pearson: 40.01998290519605
- cos_sim_spearman: 46.5989769986853
- euclidean_pearson: 45.37905883182924
- euclidean_spearman: 46.22213849806378
- manhattan_pearson: 45.40925124776211
- manhattan_spearman: 46.250705124226386
MTEB QBQTC (test)
- cos_sim_pearson: 42.719516197112526
- cos_sim_spearman: 44.57507789581106
- euclidean_pearson: 35.73062264160721
- euclidean_spearman: 40.473523909913695
- manhattan_pearson: 35.69868964086357
- manhattan_spearman: 40.46349925372903
MTEB STS22 (zh) (test)
- cos_sim_pearson: 62.340118285801104
- cos_sim_spearman: 67.72781908620632
- euclidean_pearson: 63.161965746091596
- euclidean_spearman: 67.36825684340769
- manhattan_pearson: 63.089863788261425
- manhattan_spearman: 67.40868898995384
MTEB STSB (test)
- （具体的なメトリクス値は省略）

2. Classificationタスク

MTEB AmazonReviewsClassification (zh) (test)
- accuracy: 50.308
- f1: 46.927458607895126
MTEB IFlyTek (validation)
- accuracy: 51.94305502116199
- f1: 39.82197338426721
MTEB JDReview (test)
- accuracy: 90.31894934333957
- ap: 63.89821836499594
- f1: 85.93687177603624
MTEB MassiveIntentClassification (zh-CN) (test)
- accuracy: 78.13718897108272
- f1: 74.07613180855328
MTEB MassiveScenarioClassification (zh-CN) (test)
- accuracy: 86.20040349697376
- f1: 85.05282136519973
MTEB MultilingualSentiment (validation)
- accuracy: 78.57666666666667
- f1: 78.23373528202681
MTEB OnlineShopping (test)
- accuracy: 95.06999999999998
- ap: 93.45104559324996
- f1: 95.06036329426092

3. Clusteringタスク

MTEB CLSClusteringP2P (test)
- v_measure: 60.63545326048343
MTEB CLSClusteringS2S (test)
- v_measure: 52.64834762325994

4. Rerankingタスク

MTEB CMedQAv1 (test)
- map: 91.38528814655234
- mrr: 93.35857142857144
MTEB CMedQAv2 (test)
- map: 89.72084678877096
- mrr: 91.74380952380953
MTEB MMarcoReranking (dev)
- map: 41.58107192600853
- mrr: 41.37063492063492

5. Retrievalタスク

MTEB CmedqaRetrieval (dev)
- map_at_1: 26.987
- map_at_10: 40.675
- map_at_100: 42.495
- map_at_1000: 42.596000000000004
- map_at_3: 36.195
- map_at_5: 38.704
- mrr_at_1: 41.21
- mrr_at_10: 49.816
- mrr_at_100: 50.743
- mrr_at_1000: 50.77700000000001
- mrr_at_3: 47.312
- mrr_at_5: 48.699999999999996
- ndcg_at_1: 41.21
- ndcg_at_10: 47.606
- ndcg_at_100: 54.457
- ndcg_at_1000: 56.16100000000001
- ndcg_at_3: 42.108000000000004
- ndcg_at_5: 44.393
- precision_at_1: 41.21
- precision_at_10: 10.593
- precision_at_100: 1.609
- precision_at_1000: 0.183
- precision_at_3: 23.881
- precision_at_5: 17.339
- recall_at_1: 26.987
- recall_at_10: 58.875
- recall_at_100: 87.023
- recall_at_1000: 98.328
- recall_at_3: 42.265
- recall_at_5: 49.334
MTEB CovidRetrieval (dev)
- map_at_1: 83.693
- map_at_10: 90.098
- map_at_100: 90.145
- map_at_1000: 90.146
- map_at_3: 89.445
- map_at_5: 89.935
- mrr_at_1: 83.878
- mrr_at_10: 90.007
- mrr_at_100: 90.045
- mrr_at_1000: 90.046
- mrr_at_3: 89.34
- mrr_at_5: 89.835
- ndcg_at_1: 84.089
- ndcg_at_10: 92.351
- ndcg_at_100: 92.54599999999999
- ndcg_at_1000: 92.561
- ndcg_at_3: 91.15299999999999
- ndcg_at_5: 91.968
- precision_at_1: 84.089
- precision_at_10: 10.011000000000001
- precision_at_100: 1.009
- precision_at_1000: 0.101
- precision_at_3: 32.28
- precision_at_5: 19.789
- recall_at_1: 83.693
- recall_at_10: 99.05199999999999
- recall_at_100: 99.895
- recall_at_1000: 100
- recall_at_3: 95.917
- recall_at_5: 97.893
MTEB DuRetrieval (dev)
- map_at_1: 26.924
- map_at_10: 81.392
- map_at_100: 84.209
- map_at_1000: 84.237
- map_at_3: 56.998000000000005
- map_at_5: 71.40100000000001
- mrr_at_1: 91.75
- mrr_at_10: 94.45
- mrr_at_100: 94.503
- mrr_at_1000: 94.505
- mrr_at_3: 94.258
- mrr_at_5: 94.381
- ndcg_at_1: 91.75
- ndcg_at_10: 88.53
- ndcg_at_100: 91.13900000000001
- ndcg_at_1000: 91.387
- ndcg_at_3: 87.925
- ndcg_at_5: 86.461
- precision_at_1: 91.75
- precision_at_10: 42.05
- precision_at_100: 4.827
- precision_at_1000: 0.48900000000000005
- precision_at_3: 78.55
- precision_at_5: 65.82000000000001
- recall_at_1: 26.924
- recall_at_10: 89.338
- recall_at_100: 97.856
- recall_at_1000: 99.11
- recall_at_3: 59.202999999999996
- recall_at_5: 75.642
MTEB EcomRetrieval (dev)
- map_at_1: 54.800000000000004
- map_at_10: 65.613
- map_at_100: 66.185
- map_at_1000: 66.191
- map_at_3: 62.8
- map_at_5: 64.535
- mrr_at_1: 54.800000000000004
- mrr_at_10: 65.613
- mrr_at_100: 66.185
- mrr_at_1000: 66.191
- mrr_at_3: 62.8
- mrr_at_5: 64.535
- ndcg_at_1: 54.800000000000004
- ndcg_at_10: 70.991
- ndcg_at_100: 73.434
- ndcg_at_1000: 73.587
- ndcg_at_3: 65.324
- ndcg_at_5: 68.431
- precision_at_1: 54.800000000000004
- precision_at_10: 8.790000000000001
- precision_at_100: 0.9860000000000001
- precision_at_1000: 0.1
- precision_at_3: 24.2
- precision_at_5: 16.02
- recall_at_1: 54.800000000000004
- recall_at_10: 87.9
- recall_at_100: 98.6
- recall_at_1000: 99.8
- recall_at_3: 72.6
- recall_at_5: 80.10000000000001
MTEB MMarcoRetrieval (dev)
- map_at_1: 68.33
- map_at_10: 78.261
- map_at_100: 78.522
- map_at_1000: 78.527
- map_at_3: 76.236
- map_at_5: 77.557
- mrr_at_1: 70.602
- mrr_at_10: 78.779
- mrr_at_100: 79.00500000000001
- mrr_at_1000: 79.01
- mrr_at_3: 77.037
- mrr_at_5: 78.157
- ndcg_at_1: 70.602
- ndcg_at_10: 82.254
- ndcg_at_100: 83.319
- ndcg_at_1000: 83.449
- ndcg_at_3: 78.46
- ndcg_at_5: 80.679
- precision_at_1: 70.602
- precision_at_10: 9.989
- precision_at_100: 1.05
- precision_at_1000: 0.106
- precision_at_3: 29.598999999999997
- precision_at_5: 18.948
- recall_at_1: 68.33
- recall_at_10: 94.00800000000001
- recall_at_100: 98.589
- recall_at_1000: 99.60799999999999
- recall_at_3: 84.057
- recall_at_5: 89.32900000000001
MTEB MedicalRetrieval (dev)
- map_at_1: 56.8
- map_at_10: 64.199
- map_at_100: 64.89
- map_at_1000: 64.917
- map_at_3: 62.383
- map_at_5: 63.378
- mrr_at_1: 56.8
- mrr_at_10: 64.199
- mrr_at_100: 64.89
- mrr_at_1000: 64.917
- mrr_at_3: 62.383
- mrr_at_5: 63.378
- ndcg_at_1: 56.8
- ndcg_at_10: 67.944
- ndcg_at_100: 71.286
- ndcg_at_1000: 71.879
- ndcg_at_3: 64.163
- ndcg_at_5: 65.96600000000001
- precision_at_1: 56.8
- precision_at_10: 7.9799999999999995
- precision_at_100: 0.954
- precision_at_1000: 0.1
- precision_at_3: 23.1
- precision_at_5: 14.74
- recall_at_1: 56.8
- recall_at_10: 79.80000000000001
- recall_at_100: 95.39999999999999
- recall_at_1000: 99.8
- recall_at_3: 69.3
- recall_at_5: 73.7

6. PairClassificationタスク

MTEB Cmnli (validation)
- cos_sim_accuracy: 85.91701743836441
- cos_sim_ap: 92.53650618807644
- cos_sim_f1: 86.80265975431082
- cos_sim_precision: 83.79025239338556
- cos_sim_recall: 90.039747486556
- dot_accuracy: 77.17378232110643
- dot_ap: 85.40244368166546
- dot_f1: 79.03038001481951
- dot_precision: 72.20502901353966
- dot_recall: 87.2808043020809
- euclidean_accuracy: 84.65423932651834
- euclidean_ap: 91.47775530034588
- euclidean_f1: 85.64471499723298
- euclidean_precision: 81.31567885666246
- euclidean_recall: 90.46060322656068
- manhattan_accuracy: 84.58208057726999
- manhattan_ap: 91.46228709402014
- manhattan_f1: 85.6631626034444
- manhattan_precision: 82.10075026795283
- manhattan_recall: 89.5487491232172
- max_accuracy: 85.91701743836441
- max_ap: 92.53650618807644
- max_f1: 86.80265975431082
MTEB Ocnli (validation)
- cos_sim_accuracy: 85.43584190579317
- cos_sim_ap: 90.76665640338129
- cos_sim_f1: 86.5021770682148
- cos_sim_precision: 79.82142857142858
- cos_sim_recall: 94.40337909186906
- dot_accuracy: 78.66811044937737
- dot_ap: 85.84084363880804
- dot_f1: 80.10075566750629
- dot_precision: 76.58959537572254
- dot_recall: 83.9493136219641
- euclidean_accuracy: 84.46128857606931
- euclidean_ap: 88.62351100230491
- euclidean_f1: 85.7709469509172
- euclidean_precision: 80.8411214953271
- euclidean_recall: 91.34107708553326
- manhattan_accuracy: 84.51543042772063
- manhattan_ap: 88.53975607870393
- manhattan_f1: 85.75697211155378
- manhattan_precision: 81.14985862393968
- manhattan_recall: 90.91869060190075
- max_accuracy: 85.43584190579317
- max_ap: 90.76665640338129
- max_f1: 86.5021770682148