ai-cfia · JolanThomassin · Mar 7, 2024 · Nov 2, 2023 · Nov 8, 2023 · Nov 8, 2023
diff --git a/ailab/db/finesse/prompt/json_template.json b/ailab/db/finesse/prompt/json_template.json
@@ -1,5 +1,6 @@
 {
     "page_score": "",
+    "question_quality": "",
     "crawl_id": "",
     "chunk_id": "",
     "title": "",

diff --git a/ailab/db/finesse/prompt/qna_system_prompt.txt b/ailab/db/finesse/prompt/qna_system_prompt.txt
@@ -1 +1 @@
-Generate Q&A pairs from search results, ensuring that the language in the Q&A pairs matches that of the search results. Create a single JSON file with appropriate formatting and indentation. Always leave "text_content_ empty. If you got more than one query, only treat and return result for the first one. Always return a SINGLE JSON file.
+Generate Q&A pairs from search results, ensuring that the language in the Q&A pairs matches that of the search results. Create a single JSON file with appropriate formatting and indentation. Always leave "text_content" empty. If you got more than one query, only treat and return result for the first one. Always return a SINGLE JSON file.
diff --git a/ailab/db/finesse/prompt/qna_user_prompt.txt b/ailab/db/finesse/prompt/qna_user_prompt.txt
@@ -1 +1 @@
-Your task is to process a JSON file containing text relevant to a search query. Produce a new JSON file completing all existing keys in the JSON template. Specifically, generate entries for 'question' and 'answer' where 'question' represents a question derived from the text and 'answer' provides an appropriate answer addressing the generated question.
+Your task is to process a JSON file containing text relevant to a search query. Produce a new JSON file completing all existing keys in the JSON template. Specifically, generate entries for 'question' and 'answer' where 'question' represents a question derived from the text, and 'answer' provides an appropriate response addressing the generated question. After that, determine a score into "question_quality" between 0.0 and 1.0 for the question you generated. The more complex the question is, the higher the score it receives. Complexity can be determined by the depth of knowledge required to understand the question, the number of concepts involved, and the sophistication of the language used. Every forms or PDF question like: "how do I fill a form?" or "how to open this document?" will be low scored.
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		Generate Q&A pairs from search results, ensuring that the language in the Q&A pairs matches that of the search results. Create a single JSON file with appropriate formatting and indentation. Always leave "text_content_ empty. If you got more than one query, only treat and return result for the first one. Always return a SINGLE JSON file.
		Generate Q&A pairs from search results, ensuring that the language in the Q&A pairs matches that of the search results. Create a single JSON file with appropriate formatting and indentation. Always leave "text_content" empty. If you got more than one query, only treat and return result for the first one. Always return a SINGLE JSON file.
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		Your task is to process a JSON file containing text relevant to a search query. Produce a new JSON file completing all existing keys in the JSON template. Specifically, generate entries for 'question' and 'answer' where 'question' represents a question derived from the text and 'answer' provides an appropriate answer addressing the generated question.
		Your task is to process a JSON file containing text relevant to a search query. Produce a new JSON file completing all existing keys in the JSON template. Specifically, generate entries for 'question' and 'answer' where 'question' represents a question derived from the text, and 'answer' provides an appropriate response addressing the generated question. After that, determine a score into "question_quality" between 0.0 and 1.0 for the question you generated. The more complex the question is, the higher the score it receives. Complexity can be determined by the depth of knowledge required to understand the question, the number of concepts involved, and the sophistication of the language used. Every forms or PDF question like: "how do I fill a form?" or "how to open this document?" will be low scored.
JolanThomassin marked this conversation as resolved. Show resolved Hide resolved