File size: 7,151 Bytes
399f588
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import json
from pathlib import Path
from src.schemas.labels import LABEL_REMAP, SENTIMENT_LABELS
VALID_LABELS = set(SENTIMENT_LABELS.classes)


def _deduplicate_articles(samples: list[dict]) -> list[dict]:
    """Remove samples with duplicate article text, keeping the first."""
    seen_texts = set()
    deduped = []
    removed = 0
    for s in samples:
        if s["text"] in seen_texts:
            removed += 1
            continue
        seen_texts.add(s["text"])
        deduped.append(s)
    print(f"Deduplicated articles: removed {removed}, kept {len(deduped)}")
    return deduped


def _remap_labels(samples: list[dict]) -> list[dict]:
    """Remap non-standard labels and drop entities with unmappable labels."""
    mapping = LABEL_REMAP.mapping
    remapped_count = 0
    dropped_count = 0

    for s in samples:
        cleaned_entities = []
        for e in s["entities"]:
            label = e["label"]
            if label in VALID_LABELS:
                cleaned_entities.append(e)
            elif label in mapping:
                e["label"] = mapping[label]
                cleaned_entities.append(e)
                remapped_count += 1
            else:
                dropped_count += 1
        s["entities"] = cleaned_entities

    print(f"Labels remapped: {remapped_count}, dropped (unmappable): {dropped_count}")
    return samples


def _fix_position_text(samples: list[dict]) -> list[dict]:
    """Overwrite position_text with actual span from article text when case differs."""
    fixed = 0
    for s in samples:
        text = s["text"]
        for e in s["entities"]:
            for p in e["positions"]:
                end = p["offset"] + p["length"]
                if end > len(text):
                    continue
                actual = text[p["offset"]:end]
                if actual != p["position_text"] and actual.lower() == p["position_text"].lower():
                    p["position_text"] = actual
                    fixed += 1
    print(f"Position text case mismatches fixed: {fixed}")
    return samples


def _merge_entities(samples: list[dict]) -> list[dict]:
    """Merge entity records sharing (entity_text, label) within a sample.

    Collects all positions into a single entity record per unique
    (entity_text.lower(), label) pair.
    """
    total_merged = 0
    different_label = 0

    for s in samples:
        merged: dict[str, dict] = {}
        label_seen: dict[str, str] = {}

        for e in s["entities"]:
            key = e["entity_text"].lower()

            if key in label_seen and label_seen[key] != e["label"]:
                different_label += 1

            merge_key = (key, e["label"])
            if merge_key in merged:
                merged[merge_key]["positions"].extend(e["positions"])
                total_merged += 1
            else:
                merged[merge_key] = {
                    "entity_id": e["entity_id"],
                    "entity_text": e["entity_text"],
                    "entity_type": e["entity_type"],
                    "positions": list(e["positions"]),
                    "label": e["label"],
                }
                label_seen[key] = e["label"]

        s["entities"] = list(merged.values())

    print(f"Entities merged: {total_merged}, different-label pairs: {different_label}")
    return samples


def _deduplicate_positions(samples: list[dict]) -> list[dict]:
    """Remove exact-duplicate positions (same offset + length) within each entity."""
    removed = 0
    for s in samples:
        for e in s["entities"]:
            seen = set()
            unique = []
            for p in e["positions"]:
                span = (p["offset"], p["length"])
                if span in seen:
                    removed += 1
                    continue
                seen.add(span)
                unique.append(p)
            e["positions"] = unique
    print(f"Exact-duplicate positions removed: {removed}")
    return samples


def _resolve_same_offset(samples: list[dict]) -> list[dict]:
    """For positions sharing the same offset, keep the longest span."""
    resolved = 0
    for s in samples:
        for e in s["entities"]:
            by_offset: dict[int, dict] = {}
            for p in e["positions"]:
                off = p["offset"]
                if off not in by_offset or p["length"] > by_offset[off]["length"]:
                    if off in by_offset:
                        resolved += 1
                    by_offset[off] = p
            e["positions"] = sorted(by_offset.values(), key=lambda p: p["offset"])
    print(f"Same-offset positions resolved (kept longest): {resolved}")
    return samples


def _resolve_partial_overlaps(samples: list[dict]) -> list[dict]:
    """Resolve positions that partially overlap (different offset, shared characters).

    Positions are already sorted by offset. When two positions overlap,
    the longer span is kept and the shorter one is discarded.
    """
    resolved = 0
    for s in samples:
        for e in s["entities"]:
            positions = sorted(e["positions"], key=lambda p: p["offset"])
            kept: list[dict] = []
            for p in positions:
                if not kept:
                    kept.append(p)
                    continue
                prev = kept[-1]
                prev_end = prev["offset"] + prev["length"]
                if p["offset"] < prev_end:
                    resolved += 1
                    if p["length"] > prev["length"]:
                        kept[-1] = p
                else:
                    kept.append(p)
            e["positions"] = kept
    print(f"Partial overlaps resolved (kept longest): {resolved}")
    return samples


def preprocess(samples: list[dict]) -> list[dict]:
    """Run the full preprocessing pipeline on raw samples."""
    samples = _deduplicate_articles(samples)
    samples = _remap_labels(samples)
    samples = _fix_position_text(samples)
    samples = _merge_entities(samples)
    samples = _deduplicate_positions(samples)
    samples = _resolve_same_offset(samples)
    samples = _resolve_partial_overlaps(samples)

    total_entities = sum(len(s["entities"]) for s in samples)
    print(f"Preprocessing complete: {len(samples)} samples, {total_entities} entities")
    return samples


def save_jsonl(samples: list[dict], path: str | Path) -> None:
    """Write preprocessed samples as JSONL (one JSON object per line)."""
    path = Path(path)
    path.parent.mkdir(parents=True, exist_ok=True)
    with open(path, "w", encoding="utf-8") as f:
        for s in samples:
            f.write(json.dumps(s, ensure_ascii=False) + "\n")
    print(f"Saved {len(samples)} samples to {path}")


def main(
    input_path: str | Path = "data/data_raw.json",
    output_path: str | Path = "data/data_preprocessed.jsonl",
) -> list[dict]:
    """Load raw data, preprocess, and save."""
    with open(input_path, "r", encoding="utf-8") as f:
        raw = json.load(f)

    print(f"Loaded {len(raw)} raw samples from {input_path}")
    samples = preprocess(raw)
    save_jsonl(samples, output_path)
    return samples


if __name__ == "__main__":
    main()