File size: 69,683 Bytes
0150291 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 | {
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Data Preprocessing Analysis\n",
"\n",
"Data hygiene checks for entity-level sentiment classification.\n",
"\n",
"**Structure**\n",
"1. Load Data\n",
"2. Dataset Overview\n",
"3. Duplicate Texts\n",
"4. Entity Deduplication Pipeline\n",
"5. Label Validity\n",
"6. Position Text Mismatches\n",
"7. HTML Tags\n",
"8. Preprocessing Summary"
],
"id": "8cdbbbb3574744ab"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.500521Z",
"start_time": "2026-04-18T19:53:18.495644Z"
}
},
"source": [
"import os\n",
"import json\n",
"import warnings\n",
"from collections import Counter\n",
"\n",
"import numpy as np\n",
"import pandas as pd\n",
"import matplotlib.pyplot as plt\n",
"import matplotlib.patches as mpatches\n",
"import seaborn as sns\n",
"\n",
"warnings.filterwarnings(\"ignore\")\n",
"sns.set_theme(style=\"whitegrid\", palette=\"muted\")\n",
"plt.rcParams[\"figure.dpi\"] = 120\n",
"plt.rcParams[\"axes.titlesize\"] = 13\n",
"plt.rcParams[\"axes.labelsize\"] = 11\n",
"\n",
"VALID_LABELS = {\"positive\", \"neutral\", \"negative\"}\n",
"LABEL_ORDER = [\"negative\", \"neutral\", \"positive\"]\n",
"LABEL_COLORS = {\"positive\": \"#4CAF50\", \"neutral\": \"#90A4AE\", \"negative\": \"#EF5350\"}"
],
"id": "32a9ac93d2dec54",
"outputs": [],
"execution_count": 13
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1. Load Data"
],
"id": "5eabfa4c06905b4"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.574483Z",
"start_time": "2026-04-18T19:53:18.515917Z"
}
},
"source": [
"data_path = os.path.join(\"..\", \"data\", \"data_raw.json\")\n",
"with open(data_path, \"r\", encoding=\"utf-8\") as f:\n",
" data_raw = json.load(f)\n",
"\n",
"# Flat article-level dataframe\n",
"df_articles = pd.DataFrame([\n",
" {\"id\": s[\"id\"], \"text\": s[\"text\"], \"n_entities\": len(s[\"entities\"])}\n",
" for s in data_raw\n",
"])\n",
"\n",
"# Flat entity-level dataframe\n",
"df_entities = pd.DataFrame([\n",
" {\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_id\": e[\"entity_id\"],\n",
" \"entity_text\": e[\"entity_text\"],\n",
" \"entity_type\": e[\"entity_type\"],\n",
" \"label\": e[\"label\"],\n",
" \"n_positions\": len(e[\"positions\"]),\n",
" }\n",
" for s in data_raw\n",
" for e in s[\"entities\"]\n",
"])\n",
"\n",
"print(f\"Articles : {len(df_articles):,}\")\n",
"print(f\"Entities : {len(df_entities):,}\")\n",
"print(f\"Avg entities per article: {len(df_entities)/len(df_articles):.2f}\")\n",
"df_articles.head()"
],
"id": "463eb928d9be91b0",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Articles : 1,637\n",
"Entities : 12,627\n",
"Avg entities per article: 7.71\n"
]
},
{
"data": {
"text/plain": [
" id text n_entities\n",
"0 0 Amazon sold unauthorized mole removers, and th... 4\n",
"1 1 Russia Announces Response Measures to New US S... 11\n",
"2 2 India's richest man takes on Amazon, Walmart i... 12\n",
"3 3 Govt may impose anti-dumping duty on chemical ... 10\n",
"4 4 Apollo Go: AI-Powered Autonomous Ride-Hailing ... 13"
],
"text/html": [
"<div>\n",
"<style scoped>\n",
" .dataframe tbody tr th:only-of-type {\n",
" vertical-align: middle;\n",
" }\n",
"\n",
" .dataframe tbody tr th {\n",
" vertical-align: top;\n",
" }\n",
"\n",
" .dataframe thead th {\n",
" text-align: right;\n",
" }\n",
"</style>\n",
"<table border=\"1\" class=\"dataframe\">\n",
" <thead>\n",
" <tr style=\"text-align: right;\">\n",
" <th></th>\n",
" <th>id</th>\n",
" <th>text</th>\n",
" <th>n_entities</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th>0</th>\n",
" <td>0</td>\n",
" <td>Amazon sold unauthorized mole removers, and th...</td>\n",
" <td>4</td>\n",
" </tr>\n",
" <tr>\n",
" <th>1</th>\n",
" <td>1</td>\n",
" <td>Russia Announces Response Measures to New US S...</td>\n",
" <td>11</td>\n",
" </tr>\n",
" <tr>\n",
" <th>2</th>\n",
" <td>2</td>\n",
" <td>India's richest man takes on Amazon, Walmart i...</td>\n",
" <td>12</td>\n",
" </tr>\n",
" <tr>\n",
" <th>3</th>\n",
" <td>3</td>\n",
" <td>Govt may impose anti-dumping duty on chemical ...</td>\n",
" <td>10</td>\n",
" </tr>\n",
" <tr>\n",
" <th>4</th>\n",
" <td>4</td>\n",
" <td>Apollo Go: AI-Powered Autonomous Ride-Hailing ...</td>\n",
" <td>13</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n",
"</div>"
]
},
"execution_count": 14,
"metadata": {},
"output_type": "execute_result"
}
],
"execution_count": 14
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 2. Dataset Overview"
],
"id": "6812878d46314b80"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.601499Z",
"start_time": "2026-04-18T19:53:18.597184Z"
}
},
"source": [
"print(\"=\" * 45)\n",
"print(\"DATASET OVERVIEW\")\n",
"print(\"=\" * 45)\n",
"print(f\" Total samples : {len(data_raw):,}\")\n",
"print(f\" Total entities : {len(df_entities):,}\")\n",
"print(f\" Unique entity texts : {df_entities['entity_text'].nunique():,}\")\n",
"print(f\" Unique entity types : {df_entities['entity_type'].nunique()}\")\n",
"print()\n",
"print(\"Entity type breakdown:\")\n",
"for etype, cnt in df_entities[\"entity_type\"].value_counts().items():\n",
" print(f\" {etype:12s}: {cnt:6,} ({cnt / len(df_entities) * 100:.1f}%)\")"
],
"id": "c7f3929c5b35b77c",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"=============================================\n",
"DATASET OVERVIEW\n",
"=============================================\n",
" Total samples : 1,637\n",
" Total entities : 12,627\n",
" Unique entity texts : 5,815\n",
" Unique entity types : 2\n",
"\n",
"Entity type breakdown:\n",
" company : 10,351 (82.0%)\n",
" location : 2,276 (18.0%)\n"
]
}
],
"execution_count": 15
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 3. Duplicate Article Texts\n",
"\n",
"Two samples sharing the same article text would allow the model to see identical context\n",
"in both training and evaluation, artificially inflating metrics."
],
"id": "7623b7ffc328b9a7"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.626539Z",
"start_time": "2026-04-18T19:53:18.621684Z"
}
},
"source": [
"dup_mask = df_articles.duplicated(\"text\", keep=False)\n",
"dup_articles = df_articles[dup_mask]\n",
"\n",
"print(f\"Duplicate article texts: {len(dup_articles)}\")\n",
"\n",
"if len(dup_articles) > 0:\n",
" print(dup_articles.head(10))"
],
"id": "17313e6496ece22",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Duplicate article texts: 16\n",
" id text n_entities\n",
"48 48 Elon Musk buys Twitter for $44B and will priva... 13\n",
"254 254 Judge tosses Biden threats case against northe... 12\n",
"265 265 Dozens feared dead as Russian shell hits Ukrai... 9\n",
"322 322 US weekly jobless claims unexpectedly fall\\nWA... 4\n",
"393 393 Yemen's rebels launch drone and missile strike... 9\n",
"610 610 Dozens feared dead as Russian shell hits Ukrai... 8\n",
"667 667 Elon Musk buys Twitter for $44B and will priva... 14\n",
"975 975 Mercado Pago anuncia 'conta que mais rende no ... 3\n",
"1019 1019 Yemen's rebels launch drone and missile strike... 8\n",
"1166 1166 US weekly jobless claims unexpectedly fall\\nWA... 4\n"
]
}
],
"execution_count": 16
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 4. Entity Deduplication Pipeline\n",
"\n",
"Within each sample, multiple entity records can refer to the same real-world entity\n",
"with overlapping or identical position spans. The pipeline below merges and cleans\n",
"these in four steps:\n",
"\n",
"1. **Merge entity records** by `(entity_text, label)` → all positions for that entity in one list\n",
"2. **Deduplicate exact-duplicate positions** (same offset + length)\n",
"3. **Resolve overlapping positions** (same offset, different length → keep longest)\n",
"4. **Handle partial overlaps** (rare edge cases)"
],
"id": "612bfc4a2968aa02"
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Step 1 — Merge entity records by `(entity_text, label)`\n",
"\n",
"Multiple entity records with the same `entity_text` and `label` inside one sample\n",
"are merged into a single entity with a unified position list. Conflicting labels\n",
"for the same entity text are flagged as annotation errors."
],
"id": "7765b7af17c9b50d"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.656417Z",
"start_time": "2026-04-18T19:53:18.650897Z"
}
},
"source": [
"merge_stats = {\"merged_entities\": 0, \"different_label\": 0, \"affected_samples\": set()}\n",
"conflict_examples = []\n",
"\n",
"for s in data_raw:\n",
" seen = {}\n",
" for e in s[\"entities\"]:\n",
" key = e[\"entity_text\"].lower()\n",
" if key in seen:\n",
" merge_stats[\"affected_samples\"].add(s[\"id\"])\n",
" if seen[key] == e[\"label\"]:\n",
" merge_stats[\"merged_entities\"] += 1\n",
" else:\n",
" merge_stats[\"different_label\"] += 1\n",
" conflict_examples.append({\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_text\": e[\"entity_text\"],\n",
" \"labels\": [seen[key], e[\"label\"]],\n",
" })\n",
" else:\n",
" seen[key] = e[\"label\"]\n",
"\n",
"total_merged = merge_stats[\"merged_entities\"] + merge_stats[\"different_label\"]\n",
"print(\"Step 1: Merge entity records by (entity_text, label)\")\n",
"print(\"=\" * 55)\n",
"print(f\"Duplicate (entity_text, sample) pairs: {total_merged:,}\")\n",
"print(f\"Same label: {merge_stats['merged_entities']:,}\")\n",
"print(f\"Different_labels: {merge_stats['different_label']:,}\")\n",
"print(f\"Affected samples: {len(merge_stats['affected_samples']):,}\")\n",
"print()\n",
"\n",
"if merge_stats[\"different_label\"] == 0:\n",
" print(\"All duplicates have consistent labels.\")\n",
" print(f\"Merging will consolidate {total_merged:,} redundant entity records ({total_merged / len(df_entities) * 100:.1f}% of total).\")\n",
"else:\n",
" print(f\"{merge_stats['different_label']} different-label duplicates.\")\n",
" print(pd.DataFrame(conflict_examples).head())"
],
"id": "e58872333d152f15",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Step 1: Merge entity records by (entity_text, label)\n",
"=======================================================\n",
"Duplicate (entity_text, sample) pairs: 2,014\n",
"Same label: 2,014\n",
"Different_labels: 0\n",
"Affected samples: 838\n",
"\n",
"All duplicates have consistent labels.\n",
"Merging will consolidate 2,014 redundant entity records (15.9% of total).\n"
]
}
],
"execution_count": 17
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Step 2 — Deduplicate exact-duplicate positions\n",
"\n",
"After merging, a single entity may carry duplicate position entries (identical offset\n",
"and length). These are removed, keeping one copy per unique span."
],
"id": "b0eb6a25890d61b6"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.694440Z",
"start_time": "2026-04-18T19:53:18.675611Z"
}
},
"source": [
"exact_dup_positions = []\n",
"\n",
"for s in data_raw:\n",
" merged = {}\n",
" for e in s[\"entities\"]:\n",
" key = (e[\"entity_text\"].lower(), e[\"label\"])\n",
" if key not in merged:\n",
" merged[key] = {\"entity_text\": e[\"entity_text\"], \"label\": e[\"label\"], \"positions\": []}\n",
" merged[key][\"positions\"].extend(e[\"positions\"])\n",
"\n",
" for key, ent in merged.items():\n",
" seen_spans = set()\n",
" for p in ent[\"positions\"]:\n",
" span = (p[\"offset\"], p[\"length\"])\n",
" if span in seen_spans:\n",
" exact_dup_positions.append({\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_text\": ent[\"entity_text\"],\n",
" \"position_text\": p[\"position_text\"],\n",
" \"offset\": p[\"offset\"],\n",
" \"length\": p[\"length\"],\n",
" })\n",
" else:\n",
" seen_spans.add(span)\n",
"\n",
"print(\"Step 2: Deduplicate exact-duplicate positions\")\n",
"print(\"=\" * 55)\n",
"print(f\"Exact-duplicate positions found: {len(exact_dup_positions):,}\")\n",
"if exact_dup_positions:\n",
" df_exact_dup = pd.DataFrame(exact_dup_positions)\n",
" n_samples = df_exact_dup[\"sample_id\"].nunique()\n",
" print(f\"Across {n_samples:,} sample(s)\")\n",
" print()\n",
" print(df_exact_dup.head(20).to_string(index=False))\n",
"else:\n",
" print(\"No exact-duplicate positions after merging.\")"
],
"id": "d5acf8bf642c044",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Step 2: Deduplicate exact-duplicate positions\n",
"=======================================================\n",
"Exact-duplicate positions found: 9,634\n",
"Across 938 sample(s)\n",
"\n",
" sample_id entity_text position_text offset length\n",
" 0 Verge Verge 2082 5\n",
" 1 SolarWinds SolarWinds 271 10\n",
" 1 SolarWinds SolarWinds 2767 10\n",
" 2 Grofers Grofers 2184 7\n",
" 3 Reliance industries Ltd Reliance industries Ltd 70 23\n",
" 3 Reliance industries Ltd Reliance industries Ltd 630 23\n",
" 3 Reliance industries Ltd Reliance industries Ltd 70 23\n",
" 3 Reliance industries Ltd Reliance 70 8\n",
" 3 Reliance industries Ltd Reliance industries Ltd 630 23\n",
" 3 Reliance industries Ltd Reliance 630 8\n",
" 3 India Glycols Limited India Glycols Limited 924 21\n",
" 3 India Glycols Limited India Glycols Limited 924 21\n",
" 4 Baidu Baidu 459 5\n",
" 4 Baidu Baidu 980 5\n",
" 4 Baidu Baidu 1289 5\n",
" 4 Baidu Baidu 2915 5\n",
" 4 Baidu Baidu 4180 5\n",
" 4 global industries global industries 175 17\n",
" 5 Nokian Tyres plc Nokian Tyres plc 51 16\n",
" 5 Nokian Tyres plc Nokian Tyres plc 51 16\n"
]
}
],
"execution_count": 18
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Step 3 — Resolve overlapping positions (same offset, different length)\n",
"\n",
"Two positions for the same entity starting at the same offset but with different\n",
"lengths represent the same mention captured at different granularity.\n",
"Resolution: keep the longest span (it captures the full entity mention)."
],
"id": "31dd5cc13f612c0b"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.721435Z",
"start_time": "2026-04-18T19:53:18.703478Z"
}
},
"source": [
"same_offset_diff_len = []\n",
"\n",
"for s in data_raw:\n",
" merged = {}\n",
" for e in s[\"entities\"]:\n",
" key = (e[\"entity_text\"].lower(), e[\"label\"])\n",
" if key not in merged:\n",
" merged[key] = {\"entity_text\": e[\"entity_text\"], \"positions\": []}\n",
" merged[key][\"positions\"].extend(e[\"positions\"])\n",
"\n",
" for key, ent in merged.items():\n",
" # Deduplicate exact spans first\n",
" unique_positions = {}\n",
" for p in ent[\"positions\"]:\n",
" span = (p[\"offset\"], p[\"length\"])\n",
" if span not in unique_positions:\n",
" unique_positions[span] = p\n",
"\n",
" # Group by offset\n",
" by_offset = {}\n",
" for (off, length), p in unique_positions.items():\n",
" by_offset.setdefault(off, []).append(p)\n",
"\n",
" for off, positions in by_offset.items():\n",
" if len(positions) > 1:\n",
" positions.sort(key=lambda p: p[\"length\"], reverse=True)\n",
" keep = positions[0]\n",
" for discard in positions[1:]:\n",
" same_offset_diff_len.append({\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_text\": ent[\"entity_text\"],\n",
" \"keep_text\": keep[\"position_text\"],\n",
" \"keep_span\": f'[{keep[\"offset\"]}:{keep[\"offset\"]+keep[\"length\"]})',\n",
" \"discard_text\": discard[\"position_text\"],\n",
" \"discard_span\": f'[{discard[\"offset\"]}:{discard[\"offset\"]+discard[\"length\"]})',\n",
" })\n",
"\n",
"print(\"Step 3: Resolve same-offset, different-length positions\")\n",
"print(\"=\" * 55)\n",
"print(f\"Cases found: {len(same_offset_diff_len):,}\")\n",
"if same_offset_diff_len:\n",
" df_same_off = pd.DataFrame(same_offset_diff_len)\n",
" n_samples = df_same_off[\"sample_id\"].nunique()\n",
" print(f\"Across {n_samples:,} sample(s)\")\n",
" print()\n",
" print(df_same_off.head(10).to_string(index=False))\n",
"else:\n",
" print(\"No same-offset length conflicts after deduplication.\")"
],
"id": "2ec34f05729dd40c",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Step 3: Resolve same-offset, different-length positions\n",
"=======================================================\n",
"Cases found: 805\n",
"Across 377 sample(s)\n",
"\n",
" sample_id entity_text keep_text keep_span discard_text discard_span\n",
" 3 Reliance industries Ltd Reliance industries Ltd [70:93) Reliance [70:78)\n",
" 3 Reliance industries Ltd Reliance industries Ltd [630:653) Reliance [630:638)\n",
" 7 Charles Schwab Charles Schwab Corporation [171:197) Charles Schwab [171:185)\n",
" 7 Charles Schwab Charles Schwab's [823:839) Charles Schwab [823:837)\n",
" 7 Charles Schwab Charles Schwab's [1208:1224) Charles Schwab [1208:1222)\n",
" 7 Charles Schwab Charles Schwab's [1404:1420) Charles Schwab [1404:1418)\n",
" 10 Cracker Barrel Cracker Barrel Old Country Store, Inc [130:167) Cracker Barrel Old Country Store [130:162)\n",
" 10 Cracker Barrel Cracker Barrel Old Country Store, Inc [130:167) Cracker Barrel [130:144)\n",
" 14 American Water Works Association American Water Works Association [1299:1331) American Water [1299:1313)\n",
" 15 Sainsbury's Sainsbury's [232:243) Sainsbury [232:241)\n"
]
}
],
"execution_count": 19
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Step 4 — Handle partial overlaps\n",
"\n",
"Remaining overlaps where two positions for the same entity share some characters\n",
"but start at different offsets. These are rare and may indicate tokenisation\n",
"differences in the original annotation."
],
"id": "abdbd0fa2842058c"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.748741Z",
"start_time": "2026-04-18T19:53:18.726218Z"
}
},
"source": [
"partial_overlaps = []\n",
"\n",
"for s in data_raw:\n",
" merged = {}\n",
" for e in s[\"entities\"]:\n",
" key = (e[\"entity_text\"].lower(), e[\"label\"])\n",
" if key not in merged:\n",
" merged[key] = {\"entity_text\": e[\"entity_text\"], \"positions\": []}\n",
" merged[key][\"positions\"].extend(e[\"positions\"])\n",
"\n",
" for key, ent in merged.items():\n",
" # Deduplicate exact spans\n",
" unique_positions = {}\n",
" for p in ent[\"positions\"]:\n",
" span = (p[\"offset\"], p[\"length\"])\n",
" if span not in unique_positions:\n",
" unique_positions[span] = p\n",
"\n",
" # Resolve same-offset conflicts (keep longest)\n",
" by_offset = {}\n",
" for (off, length), p in unique_positions.items():\n",
" if off not in by_offset or length > by_offset[off][\"length\"]:\n",
" by_offset[off] = p\n",
" resolved = sorted(by_offset.values(), key=lambda p: p[\"offset\"])\n",
"\n",
" # Check remaining partial overlaps\n",
" for i in range(len(resolved)):\n",
" for j in range(i + 1, len(resolved)):\n",
" si_off = resolved[i][\"offset\"]\n",
" si_end = si_off + resolved[i][\"length\"]\n",
" sj_off = resolved[j][\"offset\"]\n",
" sj_end = sj_off + resolved[j][\"length\"]\n",
" if sj_off >= si_end:\n",
" break\n",
" partial_overlaps.append({\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_text\": ent[\"entity_text\"],\n",
" \"position_text_a\": resolved[i][\"position_text\"],\n",
" \"position_text_b\": resolved[j][\"position_text\"],\n",
" \"span_a\": f'[{si_off}:{si_end})',\n",
" \"span_b\": f'[{sj_off}:{sj_end})',\n",
" \"overlap_chars\": si_end - sj_off,\n",
" })\n",
"\n",
"print(\"Step 4: Handle partial overlaps\")\n",
"print(\"=\" * 55)\n",
"print(f\"Partial overlaps remaining: {len(partial_overlaps):,}\")\n",
"if partial_overlaps:\n",
" df_partial = pd.DataFrame(partial_overlaps)\n",
" n_samples = df_partial[\"sample_id\"].nunique()\n",
" print(f\"Across {n_samples:,} sample(s)\")\n",
" print()\n",
" print(df_partial.to_string(index=False))\n",
"else:\n",
" print(\"No partial overlaps remain.\")"
],
"id": "af1fb86847fd23b3",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Step 4: Handle partial overlaps\n",
"=======================================================\n",
"Partial overlaps remaining: 19\n",
"Across 11 sample(s)\n",
"\n",
" sample_id entity_text position_text_a position_text_b span_a span_b overlap_chars\n",
" 35 The Walt Disney Company The Walt Disney Company Walt Disney Company's [0:23) [4:25) 19\n",
" 35 The Walt Disney Company The Walt Disney Company Walt Disney Company's [136:159) [140:161) 19\n",
" 35 The Walt Disney Company The Walt Disney Company Walt Disney Company's [462:485) [466:487) 19\n",
" 147 Shell Royal Dutch Shell's Shell [154:173) [166:171) 7\n",
" 147 Shell Royal Dutch Shell Shell [479:496) [491:496) 5\n",
" 397 the Walt Disney Company the Walt Disney Company Walt Disney Company [427:450) [431:450) 19\n",
" 397 Disney Walt Disney Co Disney [135:149) [140:146) 9\n",
" 397 Disney Walt Disney Disney [916:927) [921:927) 6\n",
" 537 Royal Dutch Shell Royal Dutch Shell Shell [991:1008) [1003:1008) 5\n",
" 548 Shell Royal Dutch Shell PLC Shell [1340:1361) [1352:1357) 9\n",
" 886 Disney Walt Disney Co Disney [70:84) [75:81) 9\n",
" 925 Royal Dutch Shell Royal Dutch Shell Shell [2720:2737) [2732:2737) 5\n",
" 938 Disney Walt Disney Disney [192:203) [197:203) 6\n",
" 1050 Disney The Walt Disney Company Walt Disney Company [61:84) [65:84) 19\n",
" 1050 Disney The Walt Disney Company Disney [61:84) [70:76) 14\n",
" 1050 Disney Walt Disney Company Disney [65:84) [70:76) 14\n",
" 1050 The Walt Disney Company The Walt Disney Company Walt Disney Company [61:84) [65:84) 19\n",
" 1400 Royal Dutch Shell Plc Royal Dutch Shell Plc Shell [2864:2885) [2876:2881) 9\n",
" 1410 Vale PT Vale Indonesia Vale [241:258) [244:248) 14\n"
]
}
],
"execution_count": 20
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 5. Label Validity\n",
"\n",
"The schema defines three valid labels: `positive`, `neutral`, `negative`.\n",
"Any other value is an annotation error. We check for remappable near-matches\n",
"(e.g. `very positive`) that can be salvaged rather than dropped."
],
"id": "fee7fd592186f064"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.765731Z",
"start_time": "2026-04-18T19:53:18.762683Z"
}
},
"source": [
"label_counts = df_entities[\"label\"].value_counts()\n",
"invalid = label_counts[~label_counts.index.isin(VALID_LABELS)]\n",
"\n",
"print(\"All label values and counts:\")\n",
"print(label_counts.to_frame(\"count\").to_string())\n",
"print()\n",
"\n",
"if len(invalid) == 0:\n",
" print(\"All labels are valid.\")"
],
"id": "123db84848d0be7f",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"All label values and counts:\n",
" count\n",
"label \n",
"neutral 8400\n",
"negative 2148\n",
"positive 2078\n",
"very positive 1\n",
"\n"
]
}
],
"execution_count": 21
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 6. Position Text Mismatches\n",
"\n",
"Each position stores `position_text`, `offset`, and `length`. The actual span in\n",
"the article is `text[offset : offset+length]`. We check whether these agree.\n",
"\n",
"- **Case-only mismatch** (`BRUSSELS` vs `Brussels`): offset is correct, stored string differs → fix by overwriting.\n",
"- **Content mismatch**: offset itself is wrong → requires manual review."
],
"id": "62d9dd3ff74f13c6"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.785834Z",
"start_time": "2026-04-18T19:53:18.775763Z"
}
},
"source": [
"case_mismatches, content_mismatches = [], []\n",
"\n",
"for s in data_raw:\n",
" txt = s[\"text\"]\n",
" for e in s[\"entities\"]:\n",
" for p in e[\"positions\"]:\n",
" end = p[\"offset\"] + p[\"length\"]\n",
" actual = txt[p[\"offset\"]:end]\n",
" stored = p[\"position_text\"]\n",
"\n",
" if actual != stored:\n",
" row = {\n",
" \"sample_id\": s[\"id\"],\n",
" \"entity_text\": e[\"entity_text\"],\n",
" \"stored\": stored,\n",
" \"actual\": actual,\n",
" }\n",
"\n",
" if actual.lower() == stored.lower():\n",
" case_mismatches.append(row)\n",
" else:\n",
" content_mismatches.append(row)\n",
"\n",
"print(f\"Case-only mismatches : {len(case_mismatches):,}\")\n",
"print(f\"Content mismatches : {len(content_mismatches):,}\")\n",
"print()\n",
"\n",
"if case_mismatches:\n",
" print(\"Sample case mismatches:\")\n",
" print(pd.DataFrame(case_mismatches).head()[[\"sample_id\", \"entity_text\", \"stored\", \"actual\"]].to_string(index=False))\n",
" print()\n",
" print(\"All case mismatches: overwrite position_text with actual span from text.\")\n",
"\n",
"if content_mismatches:\n",
" print(\"Content mismatches require manual review:\")\n",
" print(pd.DataFrame(content_mismatches).head().to_string(index=False))\n",
"else:\n",
" print(\"No content mismatches.\")"
],
"id": "ccbe961ef0ec6248",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Case-only mismatches : 224\n",
"Content mismatches : 0\n",
"\n",
"Sample case mismatches:\n",
" sample_id entity_text stored actual\n",
" 12 BRUSSELS BRUSSELS Brussels\n",
" 12 BRUSSELS BRUSSELS Brussels\n",
" 12 BRUSSELS BRUSSELS Brussels\n",
" 12 BRUSSELS BRUSSELS Brussels\n",
" 12 BRUSSELS BRUSSELS Brussels\n",
"\n",
"All case mismatches: overwrite position_text with actual span from text.\n",
"No content mismatches.\n"
]
}
],
"execution_count": 22
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 7. HTML Tags\n",
"\n",
"Raw text may contain residual HTML tags from web scraping (e.g. `<b>`, `<a href=...>`, `&`).\n",
"These add noise to the model input and should be stripped during preprocessing."
],
"id": "93ccd719e8730f47"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.838764Z",
"start_time": "2026-04-18T19:53:18.795458Z"
}
},
"source": [
"from bs4 import BeautifulSoup\n",
"import html as html_lib\n",
"\n",
"articles_with_tags = []\n",
"articles_with_entities = []\n",
"tag_counts = Counter()\n",
"entity_counts = Counter()\n",
"\n",
"for s in data_raw:\n",
" txt = s[\"text\"]\n",
"\n",
" soup = BeautifulSoup(txt, \"html.parser\")\n",
" tags = soup.find_all()\n",
" if tags:\n",
" tag_names = [str(t) for t in tags]\n",
" articles_with_tags.append({\"id\": s[\"id\"], \"tags\": tag_names, \"count\": len(tag_names)})\n",
" for t in tags:\n",
" tag_counts[t.name] += 1\n",
"\n",
" decoded = html_lib.unescape(txt)\n",
" if decoded != txt:\n",
" diff_chars = [\n",
" txt[i:i+20] for i in range(len(txt))\n",
" if i < len(decoded) and txt[i] != decoded[i]\n",
" ]\n",
" entities_found = []\n",
" i = 0\n",
" while i < len(txt):\n",
" if txt[i] == \"&\":\n",
" end = txt.find(\";\", i)\n",
" if end != -1:\n",
" candidate = txt[i:end+1]\n",
" if html_lib.unescape(candidate) != candidate:\n",
" entities_found.append(candidate)\n",
" i = end + 1\n",
" continue\n",
" i += 1\n",
" if entities_found:\n",
" articles_with_entities.append({\"id\": s[\"id\"], \"entities\": entities_found, \"count\": len(entities_found)})\n",
" for e in entities_found:\n",
" entity_counts[e] += 1\n",
"\n",
"print(f\"Articles containing HTML tags : {len(articles_with_tags):,} / {len(data_raw):,}\")\n",
"print(f\"Articles containing HTML entities : {len(articles_with_entities):,} / {len(data_raw):,}\")\n",
"print()\n",
"\n",
"if tag_counts:\n",
" print(\"Most common HTML tags:\")\n",
" for tag, cnt in tag_counts.most_common(10):\n",
" print(f\" {tag:30s}: {cnt:,}\")\n",
"else:\n",
" print(\"\\u2713 No HTML tags found in article texts.\")\n",
"\n",
"print()\n",
"if entity_counts:\n",
" print(\"Most common HTML entities:\")\n",
" for ent, cnt in entity_counts.most_common(10):\n",
" print(f\" {ent:30s}: {cnt:,}\")\n",
"else:\n",
" print(\"\\u2713 No HTML entities found in article texts.\")\n",
"\n",
"if articles_with_tags:\n",
" print()\n",
" print(\"Sample articles with HTML tags:\")\n",
" df_html = pd.DataFrame(articles_with_tags).sort_values(\"count\", ascending=False)\n",
" print(df_html.head(10).to_string(index=False))"
],
"id": "51fe8e19c409a770",
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Articles containing HTML tags : 2 / 1,637\n",
"Articles containing HTML entities : 6 / 1,637\n",
"\n",
"Most common HTML tags:\n",
" azn.l : 1\n",
" jnj.n : 1\n",
" sasy.pa : 1\n",
" mrna.o : 1\n",
" cvac.o : 1\n",
" pfe.n : 1\n",
" bntx.o : 1\n",
" rena.pa : 1\n",
"\n",
"Most common HTML entities:\n",
" ’ : 13\n",
" ” : 10\n",
" / : 2\n",
" € : 2\n",
" — : 1\n",
" &P Global raises EDB credit rating to 'AA'\n",
"S&: 1\n",
" : : 1\n",
"\n",
"Sample articles with HTML tags:\n",
" id tags count\n",
"1046 [<azn.l> to secure at least 300 million doses of its potential COVID-19 vaccine, a spokesman said on Thursday. The deal covers development, liability and other costs faced by the vaccine maker. The EU has also secured an option to buy 100 million additional doses of the vaccine under development. The 27 EU states could buy it at a later stage, should the vaccine prove successful. The overall price they will pay to acquire the doses has not been revealed, but under an earlier deal struck in June with AstraZeneca by Germany, France, Italy and the Netherlands, all members of the EU, AstraZeneca agreed to sell 300 million doses for 750 million euros ($843 million). The EU deal completed the preliminary accord reached with the drug maker by the four countries, the Commission said in a statement. \"We cannot indicate at this stage the specific pricing per dose. However, a significant part of the overall costs are funded by a contribution from the overall ESI funding for vaccines,\" the commission spokesman said, referring to the 336 million euros paid through the bloc's so-called emergency support instrument. It is the first contract signed by the EU with a maker of potential COVID-19 vaccines. Brussels was previously said to be in advanced talks with Johnson & Johnson <jnj.n>, Sanofi <sasy.pa>, Moderna <mrna.o> and CureVac <cvac.o> for their potential vaccines. EU officials told Reuters in July the bloc was also talking with Pfizer <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n></cvac.o></mrna.o></sasy.pa></jnj.n></azn.l>, <jnj.n>, Sanofi <sasy.pa>, Moderna <mrna.o> and CureVac <cvac.o> for their potential vaccines. EU officials told Reuters in July the bloc was also talking with Pfizer <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n></cvac.o></mrna.o></sasy.pa></jnj.n>, <sasy.pa>, Moderna <mrna.o> and CureVac <cvac.o> for their potential vaccines. EU officials told Reuters in July the bloc was also talking with Pfizer <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n></cvac.o></mrna.o></sasy.pa>, <mrna.o> and CureVac <cvac.o> for their potential vaccines. EU officials told Reuters in July the bloc was also talking with Pfizer <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n></cvac.o></mrna.o>, <cvac.o> for their potential vaccines. EU officials told Reuters in July the bloc was also talking with Pfizer <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n></cvac.o>, <pfe.n> and BionTech <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o></pfe.n>, <bntx.o> for the shot they are developing together. The contract with AstraZeneca follows an advance purchase agreement signed by Brussels with the company earlier in August. Part of the money the EU pays for supply deals covers legal risks faced by vaccine makers if their shots have unexpected side effects. These risks are increased by the hastened process to develop a vaccine in the race against the COVID-19 pandemic. \"In order to compensate for such high risks taken by manufacturers, the Advanced Purchase Agreements provide for member states to indemnify the manufacturer for liabilities incurred under certain conditions,\" the commission said. \"Liability still remains with the companies,\" it added. This issue has been one of the stumbling blocs in talks with other vaccine makers, official told Reuters, as companies prefer to have a broader shield.</bntx.o>] 7\n",
"1585 [<rena.pa> and thrown Nissan into disarray as it finds itself on course to book its lowest operating profit in 11 years. The sources said Nissan will likely kill loss-making variants for the Titan full-size pickup. Unprofitable variants include the single-cab and diesel versions. A planned shuttering of under-utilised production lines will most probably hit plants in emerging markets building Datsun and other small cars hardest, they added. \"We need to chart a recovery but the rot goes deep,\" one of the sources said of the many problems facing Nissan. The second source said all markets with factories except China were being looked at for possible reductions in production capacity. That source also said, however, that there were no plans to close an entire plant or withdraw completely from any country. In the United States, one of Nissan's biggest markets, the plan calls for fresh efforts to weed out the practice of buying market share by selling vehicles to rental car and other fleet operators at heavy discounts - a practice which destroyed profitability and undermined Nissan's brand image. \"We're trying to clean up what had happened in the past,\" one of the sources said, adding that under Ghosn, Nissan sought to meet sales objectives at any cost, including \"practically giving away cars\" to fleet customers. A team led by Jun Seki, a senior vice president and incoming vice chief operating officer, is expected to unveil the wide-ranging plan this month though some aspects are still being finalised, said the sources, who were not authorised to speak to media and declined to be identified. Nissan declined to comment. Seki is part of a new management team that will see Makoto Uchida, Nissan's head of China operations, take the helm - an appointment that is expected to take effect by Jan. 1. The new steps follow plans unveiled in July to cut headcount by 12,500 globally by early 2023 and which also flagged cuts to production capacity. At the time, then-CEO Hiroto Saikawa said 14 facilities would be affected. EMERGING MARKET WOES Overall, the plan's aim is to free up resources to focus more on the United States and China, the sources said. To that end it will roll back an aggressive expansionist strategy Ghosn set in motion under a five-year plan called Power 88 which aimed to raise profit margins and global market share to 8 percent by fiscal 2016 - goals which were never achieved. The Datsun brand - revived for emerging markets under Ghosn after being phased out in the 1980s - will likely bear the brunt of the restructuring. The models are manufactured in Indonesia, India and Russia. The sources said problems emerged after Nissan began deploying the no-frills cars in 2014 in small markets such as Indonesia, India, Russia and South Africa where it also sells vehicles under its mainstay Nissan brand. In Indonesia, for example, after a relatively good start, Datsun cars soon began eating into Nissan sales. \"We ended up pushing two mainstream brands in a market where you have a one or two percent market share. You cannot do that,\" one of the sources said, adding that there had been similar outcomes in India, South Africa and Russia. In its bigger markets, a steady supply of new or significantly redesigned models – starting with the redesigned Altima which was launched in the United States late last year – is expected to help Nissan reset the way it prices its vehicles. \"Still, it takes about a year to get any sort of tangible results,\" one of the sources said, adding that until then the Japanese automaker would continue to see sales by volume fall in the U.S. market. (Reporting by Norihiko Shirouzu; Additional reporting by Aditi Shah in New Delhi, Paul Lienert in Detroit and Naomi Tajitsu in Tokyo; Editing by Edwina Gibbs)</rena.pa>] 1\n"
]
}
],
"execution_count": 23
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 8. Preprocessing Summary\n",
"\n",
"All hygiene findings mapped to the preprocessing steps that resolve them."
],
"id": "aac93d52102c5dcf"
},
{
"cell_type": "code",
"metadata": {
"ExecuteTime": {
"end_time": "2026-04-18T19:53:18.856223Z",
"start_time": "2026-04-18T19:53:18.851227Z"
}
},
"source": [
"summary = [\n",
" {\n",
" \"Issue\": \"Duplicate article texts\",\n",
" \"Count\": len(dup_articles),\n",
" \"Action\": \"Deduplicate at sample level before splitting\",\n",
" },\n",
" {\n",
" \"Issue\": \"Duplicate entities within sample (same label)\",\n",
" \"Count\": merge_stats[\"merged_entities\"],\n",
" \"Action\": \"Remove — redundant, no information loss\",\n",
" },\n",
" {\n",
" \"Issue\": \"Non-standard label ('very positive')\",\n",
" \"Count\": int((df_entities[\"label\"] == \"very positive\").sum()),\n",
" \"Action\": \"Remap to 'positive'\",\n",
" },\n",
" {\n",
" \"Issue\": \"position_text case mismatches (e.g. BRUSSELS vs Brussels)\",\n",
" \"Count\": len(case_mismatches),\n",
" \"Action\": \"Overwrite position_text with actual span from article text\",\n",
" },\n",
" {\n",
" \"Issue\": \"position_text content mismatches\",\n",
" \"Count\": len(content_mismatches),\n",
" \"Action\": \"Manual review; drop affected positions if unresolvable\",\n",
" },\n",
"]\n",
"\n",
"df_summary = pd.DataFrame(summary)\n",
"display(\n",
"df_summary.style\n",
".set_table_styles([{\n",
" \"selector\": \"th\",\n",
" \"props\": [(\"font-weight\", \"bold\"), (\"background-color\", \"#ECEFF1\")]\n",
"}])\n",
")\n",
"\n",
"print()\n",
"total_removed = merge_stats[\"merged_entities\"] + int((df_entities[\"label\"] == \"very positive\").sum())\n",
"print(f\"Entities before preprocessing : {len(df_entities):,}\")\n",
"print(f\"Entities removed : {total_removed:,} ({total_removed / len(df_entities) * 100:.1f}%)\")\n",
"print(f\"Entities after preprocessing : {len(df_entities) - total_removed:,}\")"
],
"id": "77d44d31f2a37305",
"outputs": [
{
"data": {
"text/plain": [
"<pandas.io.formats.style.Styler at 0x1141db4d0>"
],
"text/html": [
"<style type=\"text/css\">\n",
"#T_b0e0e th {\n",
" font-weight: bold;\n",
" background-color: #ECEFF1;\n",
"}\n",
"</style>\n",
"<table id=\"T_b0e0e\">\n",
" <thead>\n",
" <tr>\n",
" <th class=\"blank level0\" > </th>\n",
" <th id=\"T_b0e0e_level0_col0\" class=\"col_heading level0 col0\" >Issue</th>\n",
" <th id=\"T_b0e0e_level0_col1\" class=\"col_heading level0 col1\" >Count</th>\n",
" <th id=\"T_b0e0e_level0_col2\" class=\"col_heading level0 col2\" >Action</th>\n",
" </tr>\n",
" </thead>\n",
" <tbody>\n",
" <tr>\n",
" <th id=\"T_b0e0e_level0_row0\" class=\"row_heading level0 row0\" >0</th>\n",
" <td id=\"T_b0e0e_row0_col0\" class=\"data row0 col0\" >Duplicate article texts</td>\n",
" <td id=\"T_b0e0e_row0_col1\" class=\"data row0 col1\" >16</td>\n",
" <td id=\"T_b0e0e_row0_col2\" class=\"data row0 col2\" >Deduplicate at sample level before splitting</td>\n",
" </tr>\n",
" <tr>\n",
" <th id=\"T_b0e0e_level0_row1\" class=\"row_heading level0 row1\" >1</th>\n",
" <td id=\"T_b0e0e_row1_col0\" class=\"data row1 col0\" >Duplicate entities within sample (same label)</td>\n",
" <td id=\"T_b0e0e_row1_col1\" class=\"data row1 col1\" >2014</td>\n",
" <td id=\"T_b0e0e_row1_col2\" class=\"data row1 col2\" >Remove — redundant, no information loss</td>\n",
" </tr>\n",
" <tr>\n",
" <th id=\"T_b0e0e_level0_row2\" class=\"row_heading level0 row2\" >2</th>\n",
" <td id=\"T_b0e0e_row2_col0\" class=\"data row2 col0\" >Non-standard label ('very positive')</td>\n",
" <td id=\"T_b0e0e_row2_col1\" class=\"data row2 col1\" >1</td>\n",
" <td id=\"T_b0e0e_row2_col2\" class=\"data row2 col2\" >Remap to 'positive'</td>\n",
" </tr>\n",
" <tr>\n",
" <th id=\"T_b0e0e_level0_row3\" class=\"row_heading level0 row3\" >3</th>\n",
" <td id=\"T_b0e0e_row3_col0\" class=\"data row3 col0\" >position_text case mismatches (e.g. BRUSSELS vs Brussels)</td>\n",
" <td id=\"T_b0e0e_row3_col1\" class=\"data row3 col1\" >224</td>\n",
" <td id=\"T_b0e0e_row3_col2\" class=\"data row3 col2\" >Overwrite position_text with actual span from article text</td>\n",
" </tr>\n",
" <tr>\n",
" <th id=\"T_b0e0e_level0_row4\" class=\"row_heading level0 row4\" >4</th>\n",
" <td id=\"T_b0e0e_row4_col0\" class=\"data row4 col0\" >position_text content mismatches</td>\n",
" <td id=\"T_b0e0e_row4_col1\" class=\"data row4 col1\" >0</td>\n",
" <td id=\"T_b0e0e_row4_col2\" class=\"data row4 col2\" >Manual review; drop affected positions if unresolvable</td>\n",
" </tr>\n",
" </tbody>\n",
"</table>\n"
]
},
"metadata": {},
"output_type": "display_data"
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"Entities before preprocessing : 12,627\n",
"Entities removed : 2,015 (16.0%)\n",
"Entities after preprocessing : 10,612\n"
]
}
],
"execution_count": 24
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3",
"language": "python",
"name": "python3"
},
"language_info": {
"name": "python",
"version": "3.10.0"
}
},
"nbformat": 4,
"nbformat_minor": 5
}
|