|
|
|
|
| GRAPH_FIELD_SEP = "<SEP>" |
| PROMPTS = {} |
|
|
| PROMPTS["fail_response"] = "Sorry, I'm not able to provide an answer to that question." |
|
|
|
|
|
|
|
|
|
|
| PROMPTS[ |
| "videorag_response" |
| ] = """---Role--- |
| You are an intelligent video content assistant. Your task is to answer user queries based **strictly** on the provided Video Data and Text Chunks. |
| |
| ---Data Sources--- |
| 1. **Video Data**: Contains specific video segments, names, and timestamps. |
| {video_data} |
| 2. **Text Chunks**: General context or transcript information. |
| {chunk_data} |
| |
| ---Instructions--- |
| 1. **Synthesize**: Combine information from both sources to answer the user's question comprehensively. |
| 2. **Strict Grounding**: |
| - Do NOT make up facts. |
| - If the provided data is insufficient, clearly state: "I don't have enough information to answer that based on the provided data." |
| 3. **Citation (Crucial)**: |
| - When using info from Video Data, you MUST cite it using the format `[x]` in the text. |
| - List full details in a "#### References" section at the end. |
| |
| ---Citation Format--- |
| **In-text:** ...highlighted the effects of deforestation [1]. |
| **Reference List:** |
| #### References |
| [1] **Video Name**, Start: 05:30, End: 08:00 |
| |
| ---Output Constraints--- |
| - **NO Thinking Process**: Do NOT output `` or `</think>` tags. Perform your reasoning silently and output ONLY the final response. |
| - **Format**: Markdown. |
| - **Tone**: Professional and objective. |
| |
| ---Example Output--- |
| The protagonist appears in the red suit during the chase [1]. |
| |
| #### References |
| # [1] video_name_1, 05:30, 08:00 |
| # [2] video_name_2, 25:00, 28:00 |
| |
| ---Goal--- |
| Generate a response that answers the user's question based strictly on the provided data, ensuring all video sources are cited with their specific timestamps. |
| """ |
|
|
|
|
|
|
|
|
|
|
|
|
| PROMPTS[ |
| "keywords_extraction" |
| ] = """- Role - |
| You are a strict keyword extraction tool. Your output must contain ONLY the extracted keywords. |
| |
| - Constraints - |
| 1. **Strictly Prohibited**: Do NOT output tags like ` |
| |
| `, `<answer>`, or any reasoning steps. |
| 2. **No Explanations**: Do not output "Here are the keywords" or any conversational text. |
| 3. **Format**: Output a single line of comma-separated keywords. |
| 4. **Clean Output**: If you think you need to reason, do it silently. Only print the final keywords. |
| |
| - Goal - |
| Extract relevant keywords from the query. Focus on entities, actions, and specific attributes. Remove stop words (e.g., "the", "is", "when", "video") unless necessary. |
| |
| ###################### |
| - Examples - |
| ###################### |
| |
| Question: Which animal does the protagonist encounter in the forest scene? |
| ################ |
| Output: |
| animal, protagonist, forest scene, encounter |
| |
| Question: What is the weather like? |
| ################ |
| Output: |
| weather |
| |
| ############################# |
| - Real Data - |
| ###################### |
| Question: {input_text} |
| ###################### |
| Output: |
| """ |
