File size: 6,456 Bytes
b2ccf94
 
 
 
 
 
 
e51a05a
b2ccf94
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2bf862f
 
 
681ec3c
2bf862f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0bcbce0
 
 
 
 
 
 
2bf862f
 
 
 
 
681ec3c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b2ccf94
 
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
import duckdb
import json, sys
import re
from pathlib import Path
ROOT_FOLDER = Path(__file__).resolve().parent.parent

sys.path.append(str(ROOT_FOLDER))
from src.semantic import semantic_search

def decode_ratings(page_content):
    block_pattern = r'\[\d\.0★\].*'
    matches = re.findall(block_pattern, page_content)
    if matches:
        pattern = r'\[(\d\.0)★\]\s*(.*?)\s*—\s*(.*)'
        parsed = []

        for r in matches[:3]:
            match = re.match(pattern, r)
            if match:
                rating, title, text = match.groups()
                parsed.append({
                    'rating': float(rating),
                    'title': title.strip(),
                    'text': text.strip()
                })

        return(parsed)
    else:
        return {}

def enrich_search_results(vector_store, query: str, k: int, hf_dataset):
    """
    Perform similarity search and enrich results with HuggingFace dataset metadata.
    
    Args:
        vector_store: LangChain vector store instance
        query: Search query string
        k: Number of results to return
        filter: Filter dict for similarity search
        hf_dataset: HuggingFace Arrow dataset (datasets.Dataset)
    
    Returns:
        List of enriched metadata objects as dicts
    """
    results = semantic_search(query, vector_store, k=k)

    # 1. Extract parent_asins from metadata
    parent_asins = [doc.metadata.get("parent_asin") for doc, score in results]

    # 2. Query HuggingFace dataset via DuckDB
    con = duckdb.connect()
    arrow_table = hf_dataset.data.table  # Get underlying PyArrow table
    con.register("hf_table", arrow_table)

    asin_list = ", ".join(f"'{asin}'" for asin in parent_asins if asin)
    query_sql = f"SELECT * FROM hf_table WHERE parent_asin IN ({asin_list})"
    hf_rows = con.execute(query_sql).fetchdf()

    # Build lookup: parent_asin -> metadata dict
    asin_to_metadata = {
        row["parent_asin"]: row.to_dict()
        for _, row in hf_rows.iterrows()
    }

    enriched_results = []

    for doc, score in results:
        parent_asin = doc.metadata.get("parent_asin")
        total_reviews = doc.metadata.get("total_reviews")
        metadata_object = asin_to_metadata.get(parent_asin, {}).copy()
        metadata_object['score'] = score
        metadata_object['total_reviews'] = total_reviews

        # 3. Extract 3 lines after "Top Reviews\n" from page_content
        page_content = doc.page_content
        metadata_object["reviews"] = decode_ratings(page_content)

        enriched_results.append(metadata_object)

    con.close()

    # 4. Return JSON metadata objects
    return [json.loads(json.dumps(obj, default=str)) for obj in enriched_results]


def enrich_bm25_search_results(retriever, query: str, k: int, hf_dataset):
    """
    Perform BM25 search and enrich results with HuggingFace dataset metadata.
    
    Args:
        retriever: LangChain BM25Retriever instance
        query: Search query string
        k: Number of results to return
        hf_dataset: HuggingFace Arrow dataset (datasets.Dataset)
    
    Returns:
        List of enriched metadata objects as dicts
    """
    # Get BM25 scores via underlying rank_bm25 library
    query_tokens = query.split()
    scores = retriever.vectorizer.get_scores(query_tokens)  # numpy array

    top_k_indices = sorted(enumerate(scores), key=lambda x: x[1], reverse=True)[:k]
    results = [(retriever.docs[i], score) for i, score in top_k_indices]

    # 1. Extract parent_asins from metadata
    parent_asins = [doc.metadata.get("parent_asin") for doc, score in results]

    # 2. Query HuggingFace dataset via DuckDB
    con = duckdb.connect()
    arrow_table = hf_dataset.data.table
    con.register("hf_table", arrow_table)

    asin_list = ", ".join(f"'{asin}'" for asin in parent_asins if asin)
    query_sql = f"SELECT * FROM hf_table WHERE parent_asin IN ({asin_list})"
    hf_rows = con.execute(query_sql).fetchdf()

    # Build lookup: parent_asin -> metadata dict
    asin_to_metadata = {
        row["parent_asin"]: row.to_dict()
        for _, row in hf_rows.iterrows()
    }

    enriched_results = []

    for doc, score in results:
        parent_asin = doc.metadata.get("parent_asin")

        metadata_object = {
            **doc.metadata,
            **asin_to_metadata.get(parent_asin, {}),
            "score": score,
        }

        metadata_object['reviews'] = metadata_object.pop('top_reviews', {}) or {}

        enriched_results.append(metadata_object)

    con.close()

    # 4. Return JSON metadata objects
    return [json.loads(json.dumps(obj, default=str)) for obj in enriched_results]

def _format_docs(results, hf_dataset):
    """
    Perform similarity search and enrich results with HuggingFace dataset metadata.
    
    Args:
        vector_store: LangChain vector store instance
        query: Search query string
        k: Number of results to return
        filter: Filter dict for similarity search
        hf_dataset: HuggingFace Arrow dataset (datasets.Dataset)
    
    Returns:
        List of enriched metadata objects as dicts
    """

    # 1. Extract parent_asins from metadata
    parent_asins = [doc.metadata.get("parent_asin") for doc in results]

    # 2. Query HuggingFace dataset via DuckDB
    con = duckdb.connect()
    arrow_table = hf_dataset.data.table  # Get underlying PyArrow table
    con.register("hf_table", arrow_table)

    asin_list = ", ".join(f"'{asin}'" for asin in parent_asins if asin)
    query_sql = f"SELECT * FROM hf_table WHERE parent_asin IN ({asin_list})"
    hf_rows = con.execute(query_sql).fetchdf()

    # Build lookup: parent_asin -> metadata dict
    asin_to_metadata = {
        row["parent_asin"]: row.to_dict()
        for _, row in hf_rows.iterrows()
    }

    enriched_results = []

    for doc in results:
        parent_asin = doc.metadata.get("parent_asin")
        total_reviews = doc.metadata.get("total_reviews")
        metadata_object = asin_to_metadata.get(parent_asin, {}).copy()
        metadata_object['total_reviews'] = total_reviews

        # 3. Extract 3 lines after "Top Reviews\n" from page_content
        page_content = doc.page_content
        metadata_object["reviews"] = decode_ratings(page_content)

        enriched_results.append(metadata_object)

    con.close()

    # 4. Return JSON metadata objects
    return [json.loads(json.dumps(obj, default=str)) for obj in enriched_results]