Enhance abstract analysis function with input validation and improved reasoning prompts; update UI text for clarity and relevance

app.py

@@ -283,11 +283,29 @@ def embed_text(text: str | list[str]) -> torch.Tensor:
 
 def analyse_abstracts(query_abstract: str, compare_abstract: dict) -> str:
     """Analyze the relationship between two abstracts and return formatted analysis"""
-    # Highlight the synergies in thesede papers that would justify further research
+    global reasoning_model
+    # Check if the compare_abstract is valid
+    if not isinstance(compare_abstract, dict) or "abstract" not in compare_abstract:
+        return "Invalid compare_abstract format. Expected a dictionary with 'abstract' key."
+    if not query_abstract or not compare_abstract["abstract"]:
+        return "Invalid input. Please provide both query_abstract and compare_abstract."
+    # Check if the query_abstract is a string
+    if not isinstance(query_abstract, str):
+        return "Invalid query_abstract format. Expected a string."
+    # Check if the compare_abstract is a string
+    if not isinstance(compare_abstract["abstract"], str):
+        return "Invalid compare_abstract format. Expected a string."
+    # Check if the query_abstract is empty
+    if not query_abstract.strip():
+        return "Invalid query_abstract format. Expected a non-empty string."
+    # Check if the compare_abstract is empty
+    if not compare_abstract["abstract"].strip():
+        return "Invalid compare_abstract format. Expected a non-empty string."
+
     messages = [
-        {
-            "role": "user",
-            "content": f"""You are trained in evaluating connections between research papers. Please **identify and analyze the links** between these two papers:
+    {
+        "role": "user",
+        "content": f"""You are trained in evaluating conceptual and methodological connections between research papers. Please **identify and analyze the reasoning-based links** between the following two papers:
 
 Paper 1 Abstract:
 {query_abstract}
@@ -295,58 +313,59 @@ Paper 1 Abstract:
 Paper 2 Abstract:
 {compare_abstract["abstract"]}
 
-Consider the following aspects in your evaluation:
+In your evaluation, consider the following dimensions:
 
-* **Methodological Cross-Pollination**: How do the methods or approaches from one paper **directly enhance or inform** the other?
-* **Principle or Mechanism Extension**: Do the papers **share underlying principles or mechanisms** that can be **combined or extended** to yield new insights?
-* **Interdisciplinary Connections**: Are there **clear opportunities** for interdisciplinary collaborations or knowledge transfer between the two papers?
-* **Solution or Application Bridge**: Can the solutions or applications presented in one paper be **directly adapted or integrated** with the other to create **novel, actionable outcomes**?
+* **Methodological Cross-Pollination**: Do the methods or approaches from one paper **directly inform, enhance, or contrast with** the other?
+* **Principle or Mechanism Extension**: Do the papers **share core principles, mechanisms, or assumptions** that could be **combined or extended** to generate new understanding or tools?
+* **Interdisciplinary Bridges**: Are there clear opportunities for **knowledge transfer or collaboration** across fields or problem domains?
+* **Solution or Application Overlap**: Can the solutions, frameworks, or applications in one paper be **adapted or repurposed** to benefit the work in the other, leading to **tangible, novel outcomes**?
 
-Consider the connections in either direction, that is, from Paper 1 -> Paper 2, or vice versa, from Paper 2 -> Paper 1
+Assess these connections in both directions (Paper 1 → Paper 2 and Paper 2 → Paper 1). Focus on **relevant and practically meaningful links** — especially those that might be **missed in practice** due to the sheer volume of publications or the separation between research communities. These are often connections that would be **immediately apparent to an expert** familiar with both papers, but easily overlooked otherwise.
 
-Return a valid JSON object with this structure:
+Return a valid JSON object in the following structure:
 {{
-    "reasoning": "Step-by-step analysis of the papers, highlighting **key established connections**, identified synergies, and **concrete complementarities**. Emphasize the most **critical, actionable insights** or **key takeaways** from the analysis using markdown bold.",
+    "reasoning": "Step-by-step conceptual analysis of how the papers relate, highlighting **key connections**, complementary methods, or shared ideas. Emphasize the most **relevant, practically useful takeaways**, and use markdown bold to highlight major points.",
 
-    # Main connecting concepts, methods, or principles
     "key_connections": [
         {{
             "connection": "connection 1",
-            "description": "Brief description (1-2 sentences) for the **established connection**, explaining its **direct relevance** to the synergy analysis."
+            "description": "1–2 sentence explanation of the **main conceptual or methodological link**, emphasizing its practical or theoretical relevance."
         }},
-       ...
+        ...
     ],
 
-    "synergies_and_complementarities": [
+    "complementarities": [
         {{
-            "type": ["Methodological Cross-Pollination", "Principle or Mechanism Extension", "Interdisciplinary Connections", "Solution or Application Bridge"],  # Choose only one type per entry, and only include relevant types to this analysis
-            "description": "Brief explanation (1-2 sentences) of the **identified, concrete synergy** or **complementarity**, and a **specific, actionable example** to illustrate the concept."
+            "type": ["Methodological Cross-Pollination", "Principle or Mechanism Extension", "Interdisciplinary Bridges", "Solution or Application Overlap"],  # Use only the most relevant label per entry
+            "description": "A concise explanation (1–2 sentences) of the **identified complementarity** or **productive relationship**, including a specific example or outcome it could enable."
         }},
-      ...
+        ...
     ],
 
-    # Novel, actionable outcomes or applications emerging from the synergies
     "research_potential": [
         {{
-            "potential": "Actionable outcome or application 1",
-            "description": "Brief description (1-2 sentences) of the **concrete potential outcome** or **application**, and a **specific scenario** to illustrate its **direct impact**."
+            "potential": "Potential application or outcome 1",
+            "description": "1–2 sentence explanation of the **concrete potential impact**, framed in terms of a **realistic scenario or use case**."
         }},
-       ...
+        ...
     ],
 
-    "rating": 1-5,  # Overall rating of the papers' synergy potential, where:
-                      # 1 = **No synergy or connection** (definitely no link between the papers)
-                      # 2 = **Low potential for synergy** (some vague or speculative connection, but highly uncertain)
-                      # 3 = **Plausible synergy potential** (some potential connections, but requiring further investigation to confirm)
-                      # 4 = **Established synergy with potential for growth** (clear connections with opportunities for further development)
-                      # 5 = **High established synergy with direct, clear opportunities** (strong, concrete links with immediate, actionable outcomes)
+    "rating": 1-5,  # Overall strength of the connection:
+                    # 1 = No meaningful connection
+                    # 2 = Weak or speculative connection
+                    # 3 = Plausible but unproven connection
+                    # 4 = Solid connection with future potential
+                    # 5 = Strong, well-aligned connection with immediate, valuable implications
 
-    "confidence": 0.0-1.0,  # Confidence in your analysis, as a floating-point value representing the probability of your assessment being accurate
+    "confidence": 0.0-1.0  # Confidence score in your assessment (e.g., 0.85 for high confidence, 1.0 for absolute certainty)
+    # Note: The confidence score should reflect your level of certainty in the analysis, not the strength of the connection itself.
+    # A score of 0.0 indicates no confidence in the analysis, while 1.0 indicates absolute certainty.
 }}
 
-Return only the JSON object, with double quotes around key names and all string values.""",
-        },
-    ]
+Return only the JSON object. All key names and string values must be in double quotes.
+""",
+    },
+]
 
     # Generate analysis
     try:
@@ -507,9 +526,9 @@ def create_interface():
         gr.HTML(
             """
             <div style="text-align: center; margin-bottom: 1rem">
-                <h1>Research Compass</h1>
-                <p>Find synergistic papers to enrich your research</p>
-                <p>An experiment in AI-driven research synergy analysis</p>
+                <h1>Inkling</h1>
+                <p>Discover papers with deep conceptual connections to your research</p>
+                <p>An experiment in AI-assisted research discovery and insight generation</p>
             </div>
         """
         )
@@ -517,22 +536,25 @@ def create_interface():
         with gr.Accordion(label="Instructions and Privacy Policy", open=False):
             gr.Markdown(
                 """
-                This tool helps you find papers that are synergistic with your research. 
-                It uses AI to analyze the connections between papers and provides insights into their potential synergies.
-                
-                Please ask any questions or provide feedback on the tool to help us improve it by starting a discussion on
-                the [Community Tab](https://huggingface.co/spaces/nomadicsynth/research-compass/discussions).
+                This tool helps you uncover research papers with **deep, meaningful connections** to your ideas.
+                It uses AI to go beyond keyword or semantic similarity — analyzing how papers relate **conceptually** and **contextually**, 
+                even when the surface topics differ.
+
+                The focus is on surfacing *novel insights* — connections that may not be obvious at a glance, 
+                but could **spark new perspectives**, **deepen understanding**, or **highlight relationships that might otherwise be overlooked**.
 
+                It’s designed to act more like a research collaborator than a search engine — helping you explore conceptual bridges and 
+                unexpected pathways in the literature.
+
+                Please ask any questions or provide feedback on the tool to help us improve it by starting a discussion on
+                the [Community Tab](https://huggingface.co/spaces/nomadicsynth/inkling/discussions).
 
                 **Privacy Policy**: The abstract or research description you provide will be included in any feedback
                 you submit and may be used to improve the model, and published in a public dataset.
                 Please ensure that you have the right to share this information.
                 By submitting feedback, you agree to the use of this information for research purposes.
-                Your feedback will be used to improve the model and the research process.
-                Please do not include any personally identifiable information, proprietary or sensitive information,
-                or any other information you do not want to share.
+                Do not include personally identifiable, proprietary, or sensitive information.
 
-                
                 **Disclaimer**: This tool is in alpha testing and is not intended for production use.
                 The results are not guaranteed to be accurate or reliable. Use at your own risk.
                 The tool is provided "as is" without any warranties or guarantees.
@@ -542,11 +564,11 @@ def create_interface():
             )
             gr.Markdown(
                 """
-                1. **Enter Abstract**: Paste an abstract or describe your research details in the text box.
-                2. **Search for Synergistic Papers**: Click the button to find papers with similar themes.
-                3. **Select a Paper**: Click on a row in the results table to view paper details.
-                4. **Analyze Connection Potential**: Click the button to analyze the synergy potential between the papers.
-                5. **Synergy Analysis**: View the detailed analysis of the connection potential between the papers.
+                1. **Enter Abstract**: Paste an abstract or describe your research question or idea in the text box.
+                2. **Find Related Papers**: Click the button to explore conceptually related research.
+                3. **Select a Paper**: Click on a row in the results table to view more details.
+                4. **Analyze Connection**: Click the analysis button to explore the potential connection between the papers.
+                5. **Insight Analysis**: Review the model’s reasoning about how and why these papers may relate meaningfully.
                 """
             )
 
@@ -556,7 +578,7 @@ def create_interface():
             lines=8,
             key="abstract",
         )
-        search_btn = gr.Button("Search for Synergistic Papers", variant="primary")
+        search_btn = gr.Button("Find Related Papers", variant="primary")
 
         # Store full paper data
         paper_data_state = gr.State([])
@@ -572,7 +594,7 @@ def create_interface():
             headers=["Title", "Authors", "Categories", "Date", "Match Score"],
             datatype=["markdown", "markdown", "str", "date", "str"],
             latex_delimiters=latex_delimiters,
-            label="Synergistic Papers",
+            label="Related Papers",
             interactive=False,
             wrap=False,
             line_breaks=False,
@@ -589,25 +611,26 @@ def create_interface():
                     show_copy_button=True,
                     key="paper_details",
                 )
-                analyze_btn = gr.Button("Analyze Connections", variant="primary", visible=False)
+                analyze_btn = gr.Button("Analyze Connection", variant="primary", visible=False)
                 with gr.Accordion(label="Feedback and Flagging", open=True, visible=False) as paper_feedback_accordion:
                     gr.Markdown(
                         """
-                        Please provide feedback on the paper match and analysis to improve the model.
+                        Please provide feedback on the relevance of this paper to your input.
+                        This helps us improve how well the system identifies meaningful research connections.
                         """
                     )
-                    # Add thumbs up/down for paper match
                     paper_feedback = gr.Radio(
-                        ["👍 Good Match", "👎 Poor Match"], label="Is this paper a good match?",
+                        ["👍 Good Match", "👎 Poor Match"],
+                        label="Is this paper meaningfully related to your query?",
                     )
                     paper_expert = gr.Checkbox(label="I am an expert in this field", value=False)
-                    paper_comment = gr.Textbox(label="Additional feedback on this paper match (optional)")
+                    paper_comment = gr.Textbox(label="Additional feedback on this match (optional)")
                     flag_paper_btn = gr.Button("Submit Paper Feedback")
 
             with gr.Column(scale=1):
                 analysis_output = gr.Markdown(
-                    value="# Synergy Analysis",
-                    label="Synergy Analysis",
+                    value="# Connection Analysis",
+                    label="Connection Analysis",
                     latex_delimiters=latex_delimiters,
                     show_copy_button=True,
                     key="analysis_output",
@@ -615,17 +638,16 @@ def create_interface():
                 with gr.Accordion(label="Feedback and Flagging", open=True, visible=False) as analysis_feedback_accordion:
                     gr.Markdown(
                         """
-                        The analysis is generated by an AI model and may not be perfect.
-                        If you find any errors or have suggestions for improvement, please let us know.
-                        Your feedback is valuable and will be used to improve the model and prompt.
+                        This connection analysis was generated by an AI model trained to reason about conceptual links between research papers.
+                        If you find the explanation helpful, unclear, or off-base, your feedback will help refine the model’s reasoning process.
                         """
                     )
-                    # Add thumbs up/down for analysis
                     analysis_feedback = gr.Radio(
-                        ["👍 Helpful", "👎 Not Helpful"], label="Was this analysis helpful?",
+                        ["👍 Helpful", "👎 Not Helpful"],
+                        label="Was this explanation useful in understanding the connection?",
                     )
                     analysis_expert = gr.Checkbox(label="I am an expert in this field", value=False)
-                    analysis_comment = gr.Textbox(label="Additional feedback on analysis (optional)")
+                    analysis_comment = gr.Textbox(label="Additional feedback on the analysis (optional)")
                     flag_analysis_btn = gr.Button("Submit Analysis Feedback")
 
         # Set up logging directories