Improved json parsing (#232)

paperqa/docs.py

@@ -701,6 +701,8 @@ async def process(match):
                 success = True
                 if self.prompts.summary_json:
                     try:
+                        # fetch from markdown ```json if present
+                        context = context.split("```json")[-1].split("```")[0]
                         result_data = json.loads(context)
                     except json.decoder.JSONDecodeError:
                         # fallback to string
@@ -758,7 +760,7 @@ async def process(match):
         context_str = "\n\n".join(
             [
                 f"{c.text.name}: {c.context}"
-                + (f"\n\n Based on {c.text.doc.citation}" if detailed_citations else "")
+                + (f"\n\nBased on {c.text.doc.citation}" if detailed_citations else "")
                 for c in answer.contexts
             ]
         )

paperqa/prompts.py

@@ -59,12 +59,12 @@
 )
 
 summary_json_system_prompt = """\
-Respond with the following JSON format:
+Provide a summary of the relevant information that could help answer the question based on the excerpt. Respond with the following JSON format:
 
 {{
-  // summary relevant information from text - {summary_length} words
-  summary: string;
-  // relevance of this summary to answer question (out of 10)
-  relevance_score: number;
+  "summary": "...",
+  "relevance_score": "..."
 }}
-"""
+
+where `summary` is relevant information from text - {summary_length} words and `relevance_score` is the relevance of `summary` to answer question (out of 10)
+"""  # noqa: E501

paperqa/version.py

@@ -1 +1 @@
-__version__ = "4.0.0-pre.5"
+__version__ = "4.0.0-pre.6"