update sentiment_tool.py

tools/sentiment_tool.py

@@ -11,124 +11,122 @@ OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
 client = OpenAI(api_key=OPENAI_API_KEY)
 
 # -------------------------
-# Input schema
+# Input Schema
 # -------------------------
 class SentimentInput(BaseModel):
-    query: str = Field(default="bitcoin", description="Cryptocurrency to analyze sentiment for.")
+    query: str = Field(default="bitcoin", description="Cryptocurrency name to evaluate sentiment for.")
+
 
 # -------------------------
-# Sentiment Tool (Option C)
+# Sentiment Tool
 # -------------------------
 class SentimentTool(BaseTool):
     name: str = "get_crypto_sentiment"
     description: str = (
-        "Fetch crypto sentiment using Serper Search only. "
-        "News headlines come from Serper's News API. "
-        "Reddit-like sentiment is extracted from URLs or snippets containing 'reddit.com'. "
-        "Returns structured JSON: sentiment, reasoning, headlines, reddit_titles."
+        "Fetches cryptocurrency sentiment from Google News and Reddit "
+        "via Serper.dev, then classifies sentiment as bullish, bearish, or neutral "
+        "and returns structured JSON."
     )
     args_schema: Type[BaseModel] = SentimentInput
 
+    # -------------------------
+    # Internal execution
+    # -------------------------
     def _run(self, query: str = "bitcoin") -> dict:
+        news_headlines = []
+        reddit_titles = []
+        news_error = None
+        reddit_error = None
+
+        # -------------------------
+        # 1) Fetch Google News (Serper News API)
+        # -------------------------
         try:
-            # -----------------------------------------
-            # Fetch NEWS via Serper News API
-            # -----------------------------------------
-            news_headlines = []
-            news_error = None
-            try:
-                news_url = "https://google.serper.dev/news"
-                headers = {"X-API-KEY": SERPER_API_KEY, "Content-Type": "application/json"}
-                payload = {"q": f"{query} crypto", "num": 5}
-
-                news_response = requests.post(news_url, headers=headers, json=payload, timeout=10)
-                news_response.raise_for_status()
-
-                news_items = news_response.json().get("news", [])
-                news_headlines = [n.get("title", "") for n in news_items[:5]]
-
-            except Exception as e:
-                news_error = str(e)
-
-            # -----------------------------------------
-            # Extract REDDIT-like content from Serper Search results
-            # -----------------------------------------
-            reddit_titles = []
-            reddit_error = None
-            try:
-                search_url = "https://google.serper.dev/search"
-                headers = {"X-API-KEY": SERPER_API_KEY, "Content-Type": "application/json"}
-                payload = {"q": f"{query} reddit crypto", "num": 8}
-
-                search_response = requests.post(search_url, headers=headers, json=payload, timeout=10)
-                search_response.raise_for_status()
-                organic = search_response.json().get("organic", [])
-
-                for item in organic:
-                    url = item.get("link", "")
-                    snippet = item.get("snippet", "")
-                    title = item.get("title", "")
-
-                    # Accept if URL or snippet mentions Reddit
-                    if "reddit.com" in url.lower() or "reddit" in snippet.lower():
-                        reddit_titles.append(title)
-
-            except Exception as e:
-                reddit_error = str(e)
-
-            # -----------------------------------------
-            # Build combined text for LLM classification
-            # -----------------------------------------
-            combined_text = (
-                "News: " + " | ".join(news_headlines) +
-                "\nReddit-like: " + " | ".join(reddit_titles)
-            )
+            news_url = "https://google.serper.dev/news"
+            headers = {"X-API-KEY": SERPER_API_KEY, "Content-Type": "application/json"}
+            payload = {"q": f"{query} crypto", "num": 5}
+
+            resp = requests.post(news_url, headers=headers, json=payload, timeout=10)
+            resp.raise_for_status()
+
+            news_items = resp.json().get("news", [])
+            news_headlines = [item["title"] for item in news_items[:5]]
 
-            # -----------------------------------------
-            # LLM sentiment classification (JSON enforced)
-            # -----------------------------------------
-            sentiment_prompt = f"""
-You are a crypto sentiment analyst.
+        except Exception as e:
+            news_error = str(e)
+
+        # -------------------------
+        # 2) Fetch Reddit Discussions (Serper Search API)
+        # -------------------------
+        try:
+            reddit_url = "https://google.serper.dev/search"
+            headers = {"X-API-KEY": SERPER_API_KEY, "Content-Type": "application/json"}
+
+            # OPTION C — hybrid search (most accurate)
+            payload = {
+                "q": (
+                    f'"{query}" '
+                    f'(site:reddit.com/r/CryptoCurrency OR '
+                    f'site:reddit.com/r/{query} OR '
+                    f'site:reddit.com/r/Bitcoin OR '
+                    f'site:reddit.com)'
+                ),
+                "num": 5
+            }
+
+            resp = requests.post(reddit_url, headers=headers, json=payload, timeout=10)
+            resp.raise_for_status()
+
+            organic = resp.json().get("organic", [])
+            reddit_titles = [item["title"] for item in organic[:5]]
 
-Based on the following news headlines and Reddit-like search results,
-give an *overall sentiment classification* for "{query}" as:
-- bullish
-- bearish
-- neutral
+        except Exception as e:
+            reddit_error = str(e)
 
-Respond ONLY in **valid JSON** using this schema:
+        # -------------------------
+        # 3) Sentiment Analysis (OpenAI)
+        # -------------------------
+        sentiment_prompt = f"""
+You are a cryptocurrency sentiment analyst.
+
+Based on the following data:
+News Headlines: {news_headlines}
+Reddit Titles: {reddit_titles}
+
+Classify the overall sentiment toward "{query}" as bullish, bearish, or neutral.
+Write a concise reasoning sentence.
+
+Return ONLY valid JSON exactly in this format:
 
 {{
-  "sentiment": "bullish/bearish/neutral",
-  "reasoning": "short explanation summarizing why",
+  "sentiment": "<bullish/bearish/neutral>",
+  "reasoning": "<short explanation>",
   "news_headlines": [...],
   "reddit_titles": [...],
-  "news_error": null or error string,
-  "reddit_error": null or error string
+  "news_error": "<error or null>",
+  "reddit_error": "<error or null>"
 }}
-
-Here is the text:
-
-{combined_text}
 """
 
+        try:
             completion = client.chat.completions.create(
                 model="gpt-4.1",
                 messages=[
-                    {"role": "system", "content": "Return ONLY strict JSON. No commentary."},
-                    {"role": "user",  "content": sentiment_prompt}
-                ],
-                temperature = 0.2
+                    {"role": "system", "content": "You return ONLY valid JSON, no prose."},
+                    {"role": "user", "content": sentiment_prompt}
+                ]
             )
 
-            return completion.choices[0].message.content
+            raw_output = completion.choices[0].message.content
+
+            # The model returns correct JSON; just forward it:
+            return raw_output
 
         except Exception as e:
             return {
-                "sentiment": "unknown",
-                "reasoning": "Sentiment tool failed to run.",
-                "news_headlines": [],
-                "reddit_titles": [],
-                "news_error": str(e),
-                "reddit_error": str(e)
+                "error": f"Sentiment classification failed: {str(e)}",
+                "news_headlines": news_headlines,
+                "reddit_titles": reddit_titles,
+                "news_error": news_error,
+                "reddit_error": reddit_error
             }