task_lite_clean_en/363/metadata.json

{
  "absolute_id": 363,
  "persona": "Researcher",
  "task": "Based on the memory-related papers on the desktop, write a \"Review of Research Progress in Large Language Model Personalization and Memory Management\"\n\nTask background: As an AI researcher, you are required to write a review for the team on the latest progress in LLM personalization and memory management. You have four related papers on hand, covering continuous personalization (SPRInG), memory evaluation benchmark (MemoryRewardBench), ethnographic text annotation (Ethnographic Annotation), and code understanding ability (Beyond Accuracy). You need to read these papers, extract key ideas, methods, experimental results, and summarize research trends and challenges in the current field.\n\nTask goal: Please complete a review report (PDF format) based on the four PDF papers in the data/input/ directory. Specific requirements:\n\nWrite an abstract for each paper, including research questions, methods, and main findings.\n\nCompare the similarities and differences between these papers in terms of research objectives, technical routes, and evaluation indicators.\n\nAnalyze the common challenges and future directions of LLM in personalization, memory management, text understanding, etc.\n\nAppropriately cite the paper content in the review and add hyperlinks to relevant chapters in the original text (such as the methods section).\n\nThe output file is saved as output/LLM_personalized_memory_survey.pdf.",
  "task_diff": "medium",
  "output_files": [
    "LLM_personalized_memory_survey.pdf"
  ],
  "rubrics": [
    "Is the output file LLM_personalized_memory_survey.pdf successfully generated and in legal pdf format?",
    "Does the review include separate abstract sections for the four papers, each covering the research questions, methods, and main findings?",
    "Does the abstract of the SPRInG paper correctly state that the research question is to address the problem of continuous personalization of user preferences evolving over time?",
    "Are the two core methods of drift-driven selective adaptation and logarithmic interpolation correctly described in the SPRInG paper abstract?",
    "Does the abstract of the MemoryRewardBench paper state that this is the first benchmark to evaluate the long-term memory management capabilities of reward models?",
    "Does the MemoryRewardBench paper abstract correctly state that the benchmark covers 10 different settings, with context lengths ranging from 8K to 128K tokens?",
    "Does the abstract of the EthnographicAnnotation paper correctly conclude that LLM cannot currently replace human experts for ethnographic text annotation?",
    "Does the abstract of the EthnographicAnnotation paper indicate that longer texts, ordinal distinguishing features, and fuzzy constructs are particularly difficult for LLM?",
    "Is the research question correctly asked in the BeyondAccuracy paper abstract: Can testing human metrics predict LLM code comprehension success or failure?",
    "Are the results given correctly in the BeyondAccuracy paper abstract: human metric prediction AUROC 0.63, shadow model AUROC 0.86?",
    "Does the review include a comparative analysis chapter, comparing four papers from the three dimensions of research objectives, technical routes, and evaluation indicators?",
    "Does the comparative analysis chapter include a comparison table of research objectives and correctly categorize the research areas of the four papers?",
    "Does the comparative analysis point out that SPRInG and MemoryRewardBench directly focus on personalization and memory, while the other two studies study the boundary of capabilities from the side?",
    "Does the review analysis point out that SPRInG is algorithmic innovation, MemoryRewardBench is benchmark construction, EthnographicAnnotation is empirical evaluation boundary, and BeyondAccuracy is methodological innovation?",
    "Has the common challenge of balancing dynamics and stability been analyzed in the review?",
    "Does the review discuss the common challenge of fundamental difficulties in assessment, pointing out problems with traditional aggregation accuracy?",
    "Has the challenge of the tension between domain expertise and generalization abilities been discussed in the review?",
    "Has the challenge of interpretability and lack of transparency been analyzed in the review?",
    "Does the review propose at least five future research directions (theoretical framework of continuous personalization, science of memory management assessment, cross-cultural understanding of personalization, beyond accuracy assessment methodology, exploration of semi-parametric architectures)?",
    "Have all four papers in the review added hyperlinks to the original arXiv papers?",
    "Does the review include a reference list giving complete information on all four papers?",
    "Is the PDF file format correct, including titles and chapter levels, and is the table format displayed correctly?"
  ],
  "rubric_types": [
    "Basic Evaluation",
    "Basic Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Outcome Evaluation",
    "Basic Evaluation",
    "Basic Evaluation",
    "Basic Evaluation"
  ],
  "file_dep_graph": [
    {
      "from": "2601.09974v1.pdf",
      "to": "LLM_personalized_memory_survey.pdf"
    },
    {
      "from": "2601.11969v2.pdf",
      "to": "LLM_personalized_memory_survey.pdf"
    },
    {
      "from": "2601.12099v1.pdf",
      "to": "LLM_personalized_memory_survey.pdf"
    },
    {
      "from": "2601.12951v1.pdf",
      "to": "LLM_personalized_memory_survey.pdf"
    }
  ],
  "data_manifest": [
    {
      "filename": "2601.09974v1.pdf",
      "stored_relpath": "data/2601.09974v1.pdf"
    },
    {
      "filename": "2601.11969v2.pdf",
      "stored_relpath": "data/2601.11969v2.pdf"
    },
    {
      "filename": "2601.12099v1.pdf",
      "stored_relpath": "data/2601.12099v1.pdf"
    },
    {
      "filename": "2601.12951v1.pdf",
      "stored_relpath": "data/2601.12951v1.pdf"
    }
  ],
  "tested_capabilities": [
    "Task-Providing File Utilization",
    "Result-Providing Files Utilization",
    "Semantic Content Relations Understanding"
  ]
}