raghavNCI
commited on
Commit
·
826a1b8
1
Parent(s):
aefa1e1
google search functionality
Browse files- nuse_modules/google_search.py +32 -0
- question.py +37 -26
nuse_modules/google_search.py
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# nuse_modules/google_search.py
|
2 |
+
|
3 |
+
import os
|
4 |
+
import requests
|
5 |
+
|
6 |
+
GOOGLE_API_KEY = os.getenv("GOOGLE_SEARCH_API_KEY")
|
7 |
+
GOOGLE_CX_ID = os.getenv("GOOGLE_CX_ID")
|
8 |
+
|
9 |
+
def search_google_news(keywords: list[str], num_results: int = 5):
|
10 |
+
query = " ".join(keywords)
|
11 |
+
url = (
|
12 |
+
f"https://www.googleapis.com/customsearch/v1"
|
13 |
+
f"?key={GOOGLE_API_KEY}&cx={GOOGLE_CX_ID}"
|
14 |
+
f"&q={query}&num={num_results}"
|
15 |
+
)
|
16 |
+
|
17 |
+
try:
|
18 |
+
res = requests.get(url, timeout=10)
|
19 |
+
res.raise_for_status()
|
20 |
+
data = res.json()
|
21 |
+
results = []
|
22 |
+
|
23 |
+
for item in data.get("items", []):
|
24 |
+
results.append({
|
25 |
+
"title": item.get("title"),
|
26 |
+
"link": item.get("link"),
|
27 |
+
"snippet": item.get("snippet"),
|
28 |
+
})
|
29 |
+
|
30 |
+
return results
|
31 |
+
except Exception as e:
|
32 |
+
return {"error": str(e)}
|
question.py
CHANGED
@@ -10,6 +10,7 @@ from urllib.parse import quote
|
|
10 |
import json
|
11 |
from nuse_modules.classifier import classify_question, REVERSE_MAP
|
12 |
from nuse_modules.keyword_extracter import keywords_extractor
|
|
|
13 |
|
14 |
load_dotenv()
|
15 |
|
@@ -27,6 +28,9 @@ HEADERS = {
|
|
27 |
"Content-Type": "application/json"
|
28 |
}
|
29 |
|
|
|
|
|
|
|
30 |
def is_relevant(article, keywords):
|
31 |
text = f"{article.get('title', '')} {article.get('content', '')}".lower()
|
32 |
return any(kw.lower() in text for kw in keywords)
|
@@ -79,40 +83,47 @@ async def ask_question(input: QuestionInput):
|
|
79 |
print("Intent ID:", qid)
|
80 |
print("Category:", REVERSE_MAP.get(qid, "unknown"))
|
81 |
|
82 |
-
|
83 |
|
84 |
-
|
|
|
|
|
85 |
|
86 |
-
|
87 |
-
|
|
|
|
|
88 |
|
89 |
-
|
90 |
-
|
91 |
-
articles = fetch_gnews_articles(query_and)
|
92 |
|
93 |
-
|
94 |
-
|
95 |
-
articles = fetch_gnews_articles(
|
96 |
|
97 |
-
|
|
|
|
|
98 |
|
99 |
-
|
100 |
-
a.get("content") or ""
|
101 |
-
for a in relevant_articles
|
102 |
-
])[:15000]
|
103 |
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
|
|
|
|
|
|
|
|
|
|
110 |
|
111 |
# Step 3: Ask Mistral to answer using the context
|
112 |
answer_prompt = (
|
113 |
f"You are a concise news assistant. Answer the user's question clearly using the context below if relevant. "
|
114 |
f"If the context is not helpful, you may rely on your own knowledge, but do not mention the context or question again.\n\n"
|
115 |
-
f"Context:\n{context}\n\n"
|
116 |
f"Question: {question}\n\n"
|
117 |
f"Answer:"
|
118 |
)
|
@@ -125,8 +136,8 @@ async def ask_question(input: QuestionInput):
|
|
125 |
return {
|
126 |
"question": question,
|
127 |
"answer": final_answer.strip(),
|
128 |
-
"sources": [
|
129 |
-
|
130 |
-
|
131 |
-
]
|
132 |
}
|
|
|
10 |
import json
|
11 |
from nuse_modules.classifier import classify_question, REVERSE_MAP
|
12 |
from nuse_modules.keyword_extracter import keywords_extractor
|
13 |
+
from nuse_modules.google_search import search_google_news
|
14 |
|
15 |
load_dotenv()
|
16 |
|
|
|
28 |
"Content-Type": "application/json"
|
29 |
}
|
30 |
|
31 |
+
def should_extract_keywords(type_id: int) -> bool:
|
32 |
+
return type_id in {1, 2, 3, 4, 5, 6, 7, 10}
|
33 |
+
|
34 |
def is_relevant(article, keywords):
|
35 |
text = f"{article.get('title', '')} {article.get('content', '')}".lower()
|
36 |
return any(kw.lower() in text for kw in keywords)
|
|
|
83 |
print("Intent ID:", qid)
|
84 |
print("Category:", REVERSE_MAP.get(qid, "unknown"))
|
85 |
|
86 |
+
necessary = should_extract_keywords(qid)
|
87 |
|
88 |
+
if necessary:
|
89 |
+
keywords = keywords_extractor(question)
|
90 |
+
print("Raw extracted keywords:", keywords)
|
91 |
|
92 |
+
if not keywords:
|
93 |
+
return {"error": "Keyword extraction failed."}
|
94 |
+
|
95 |
+
results = search_google_news(keywords)
|
96 |
|
97 |
+
for r in results:
|
98 |
+
print(r["title"], r["link"])
|
|
|
99 |
|
100 |
+
# Step 2: Fetch articles using AND, then fallback to OR
|
101 |
+
# query_and = " AND ".join(f'"{kw}"' for kw in keywords)
|
102 |
+
# articles = fetch_gnews_articles(query_and)
|
103 |
|
104 |
+
# if not articles:
|
105 |
+
# query_or = " OR ".join(f'"{kw}"' for kw in keywords)
|
106 |
+
# articles = fetch_gnews_articles(query_or)
|
107 |
|
108 |
+
# relevant_articles = [a for a in articles if is_relevant(a, keywords)]
|
|
|
|
|
|
|
109 |
|
110 |
+
# context = "\n\n".join([
|
111 |
+
# a.get("content") or ""
|
112 |
+
# for a in relevant_articles
|
113 |
+
# ])[:15000]
|
114 |
+
|
115 |
+
# if not context.strip():
|
116 |
+
# return {
|
117 |
+
# "question": question,
|
118 |
+
# "answer": "Cannot answer – no relevant context found.",
|
119 |
+
# "sources": []
|
120 |
+
# }
|
121 |
|
122 |
# Step 3: Ask Mistral to answer using the context
|
123 |
answer_prompt = (
|
124 |
f"You are a concise news assistant. Answer the user's question clearly using the context below if relevant. "
|
125 |
f"If the context is not helpful, you may rely on your own knowledge, but do not mention the context or question again.\n\n"
|
126 |
+
# f"Context:\n{context}\n\n"
|
127 |
f"Question: {question}\n\n"
|
128 |
f"Answer:"
|
129 |
)
|
|
|
136 |
return {
|
137 |
"question": question,
|
138 |
"answer": final_answer.strip(),
|
139 |
+
# "sources": [
|
140 |
+
# {"title": a["title"], "url": a["url"]}
|
141 |
+
# for a in relevant_articles
|
142 |
+
# ]
|
143 |
}
|