File size: 7,309 Bytes
fef33ce | 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 | """
Data preprocessing for Amazon Reviews 2023.
Loads raw JSONL files, filters low-activity users and items,
encodes user/item IDs as sequential integers, and saves clean
parquet files to data/processed/.
"""
from __future__ import annotations
import pickle
from pathlib import Path
from typing import Tuple
import pandas as pd
from sklearn.preprocessing import LabelEncoder
from src.config import DataConfig, PROCESSED_DIR
from src.utils.logging import get_logger
logger = get_logger(__name__)
def load_raw_reviews(path: Path) -> pd.DataFrame:
"""
Load raw reviews JSONL into a DataFrame, keeping only the columns we need.
Expected columns: user_id, asin (=parent_asin if available), rating, timestamp.
Args:
path: Path to raw reviews JSONL.
Returns:
DataFrame with columns [user_id, item_id, rating, timestamp].
"""
logger.info("Loading reviews from %s", path)
df = pd.read_json(path, lines=True)
# Use parent_asin when available as the canonical item identifier
if "parent_asin" in df.columns:
df["item_id"] = df["parent_asin"].fillna(df["asin"])
else:
df["item_id"] = df["asin"]
keep = ["user_id", "item_id", "rating", "timestamp"]
df = df[keep].copy()
# Normalise types
df["rating"] = pd.to_numeric(df["rating"], errors="coerce")
df["timestamp"] = pd.to_numeric(df["timestamp"], errors="coerce")
# Drop rows with missing essentials
df = df.dropna(subset=["user_id", "item_id", "rating"])
df = df.drop_duplicates(subset=["user_id", "item_id"])
logger.info("Loaded %d reviews (%d users, %d items)",
len(df), df["user_id"].nunique(), df["item_id"].nunique())
return df
def load_raw_metadata(path: Path) -> pd.DataFrame:
"""
Load raw metadata JSONL into a DataFrame.
Args:
path: Path to raw metadata JSONL.
Returns:
DataFrame with item-level attributes.
"""
logger.info("Loading metadata from %s", path)
df = pd.read_json(path, lines=True)
if "parent_asin" in df.columns:
df["item_id"] = df["parent_asin"].fillna(df.get("asin", df["parent_asin"]))
elif "asin" in df.columns:
df["item_id"] = df["asin"]
else:
raise ValueError("Metadata JSONL has neither 'asin' nor 'parent_asin' column.")
# Flatten categories: take first listed category as primary
if "categories" in df.columns:
df["category"] = df["categories"].apply(
lambda x: x[0] if isinstance(x, list) and len(x) > 0 else "Unknown"
)
else:
df["category"] = "Unknown"
# Normalise price to float
if "price" in df.columns:
df["price"] = pd.to_numeric(df["price"], errors="coerce")
else:
df["price"] = float("nan")
keep_cols = ["item_id", "title", "category", "price"]
# Include brand/store if present
for col in ("store", "brand"):
if col in df.columns:
df["brand"] = df[col].fillna("Unknown")
break
else:
df["brand"] = "Unknown"
keep_cols.append("brand")
# Include description for text features
for col in ("description", "features"):
if col in df.columns:
df["description"] = df[col].apply(
lambda x: " ".join(x) if isinstance(x, list) else str(x)
)
break
else:
df["description"] = ""
keep_cols.append("description")
df = df[[c for c in keep_cols if c in df.columns]].drop_duplicates("item_id")
logger.info("Loaded metadata for %d items", len(df))
return df
def filter_interactions(
df: pd.DataFrame,
min_user_reviews: int,
min_item_reviews: int,
) -> pd.DataFrame:
"""
Iteratively filter users and items below minimum interaction counts.
Args:
df: Interactions DataFrame.
min_user_reviews: Minimum reviews per user.
min_item_reviews: Minimum reviews per item.
Returns:
Filtered DataFrame.
"""
prev_size = len(df) + 1
while len(df) < prev_size:
prev_size = len(df)
user_counts = df["user_id"].value_counts()
df = df[df["user_id"].isin(user_counts[user_counts >= min_user_reviews].index)]
item_counts = df["item_id"].value_counts()
df = df[df["item_id"].isin(item_counts[item_counts >= min_item_reviews].index)]
logger.info("After filtering: %d reviews, %d users, %d items",
len(df), df["user_id"].nunique(), df["item_id"].nunique())
return df
def encode_ids(
df: pd.DataFrame,
user_encoder_path: Path,
item_encoder_path: Path,
) -> Tuple[pd.DataFrame, LabelEncoder, LabelEncoder]:
"""
Encode user_id and item_id as zero-based integers and save encoders.
Args:
df: Interactions DataFrame with string user_id and item_id.
user_encoder_path: Where to save the user LabelEncoder.
item_encoder_path: Where to save the item LabelEncoder.
Returns:
Tuple of (encoded_df, user_encoder, item_encoder).
"""
PROCESSED_DIR.mkdir(parents=True, exist_ok=True)
user_enc = LabelEncoder()
item_enc = LabelEncoder()
df = df.copy()
df["user_idx"] = user_enc.fit_transform(df["user_id"])
df["item_idx"] = item_enc.fit_transform(df["item_id"])
with open(user_encoder_path, "wb") as f:
pickle.dump(user_enc, f)
with open(item_encoder_path, "wb") as f:
pickle.dump(item_enc, f)
logger.info("Encoded %d users and %d items", len(user_enc.classes_), len(item_enc.classes_))
return df, user_enc, item_enc
def preprocess(cfg: DataConfig) -> Tuple[pd.DataFrame, pd.DataFrame]:
"""
Full preprocessing pipeline: load → filter → encode → save.
Args:
cfg: DataConfig.
Returns:
Tuple of (interactions_df, metadata_df).
"""
interactions = load_raw_reviews(cfg.raw_reviews_path)
metadata = load_raw_metadata(cfg.raw_metadata_path)
# Optional debug subsetting before filtering
if cfg.debug:
top_users = interactions["user_id"].value_counts().head(cfg.debug_n_users).index
interactions = interactions[interactions["user_id"].isin(top_users)].copy()
logger.info("Debug mode: trimmed to %d users", interactions["user_id"].nunique())
interactions = filter_interactions(
interactions, cfg.min_user_reviews, cfg.min_item_reviews
)
interactions, user_enc, item_enc = encode_ids(
interactions, cfg.user_encoder_path, cfg.item_encoder_path
)
# Keep only metadata for items present after filtering
valid_items = set(interactions["item_id"].unique())
metadata = metadata[metadata["item_id"].isin(valid_items)].copy()
# Add item_idx to metadata
item_id_to_idx = dict(zip(user_enc.classes_, range(len(user_enc.classes_))))
item_id_to_idx = dict(zip(item_enc.classes_, range(len(item_enc.classes_))))
metadata["item_idx"] = metadata["item_id"].map(item_id_to_idx)
# Save
PROCESSED_DIR.mkdir(parents=True, exist_ok=True)
interactions.to_parquet(cfg.interactions_path, index=False)
metadata.to_parquet(cfg.metadata_path, index=False)
logger.info("Saved interactions → %s", cfg.interactions_path)
logger.info("Saved metadata → %s", cfg.metadata_path)
return interactions, metadata
|